2018 Prem Et Al - Killing of Social Leaders For Territorial Control - The Unintended Consequences of Peace

LACEA WORKING PAPER SERES. No.
0019
KILLING SOCIAL LEADERS FOR TERRITORIAL CONTROL: THE UNINTENDED

CONSEQUENCES OF PEACE
Mounu Prem
Andrés Rivera
Darío Romero
Juan F. Vargas
LATIN AMERICAN AND THE CARIBBEAN ECONOMIC ASSOCIATION

December 2018
The views expressed herein are those of the authors and do not necessarily reflect the views of the
Latin American and the Caribbean Economic Association. Research published in this series may
include views on policy, but LACEA takes no institutional policy positions.
LACEA working papers are circulated for discussion and comment purposes. Citation of such a paper
should account for its provisional character. A revised version may be available directly from the
author.
© 2018 by Mounu Prem, Andrés Rivera, Darío Romero and Juan F. Vargas. All rights reserved. Short
sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided
that full credit, including © notice, is given to the source.
LACEA WORKING PAPER SERIES No. 0019 December 2018
Killing social leaders for territorial control: the unintended consequences of peace
Mounu Prem
Universidad del Rosario, School of Economics
[email protected]
Andrés Rivera
[email protected]
Darío Romero
Department of Economics, Columbia University
[email protected]
Juan F. Vargas
[email protected]
ABSTRACT
Incomplete peace agreements may inadvertently increase insecurity if they trigger violent territorial
contestation. We study the unintended consequences of the Colombian peace process and find that
the permanent ceasefire declared by the FARC insurgency during peace negotiations with the
government triggered a surge in the targeting of local community leaders. Leaders were killed by
armed groups excluded from the peace process to thwart collective action and civilian mobilization,
thus consolidating their dominance in formerly FARC-controlled areas. These results are exacerbated
in places with judicial inefficiency and where peasants dispossessed during the conflict have started
administrative process to reclaim their land.
JEL Classification: D72, D74.
Keywords: Social leaders, Peace process, Armed conflict, Territorial control.
ACKNOWLEDGEMENTS AND FINANCIAL DISCLOSURE

We thank Adriana Camacho, Héctor Galindo, Jorge Gallego, Felipe González, Dorothy Kronick, Luis
Martinez, Fabio Sánchez, Andrés Rosas, Santiago Saavedra, Jacob Shapiro, Abbey Steele, Mateo
Uribe and seminar participants at the Columbia University Development Colloquium, the 2018
Annual Workshop of the Households in Conflict Network, the 2018 LACEA-LAMES Annual
Conference, Rosario-Andes Taller Applied (RATA), and Universidad Javeriana for helpful
comments and suggestions. We are especially grateful to Charu Prem for in depth comments at
various stages of the project.
KILLING SOCIAL LEADERS FOR TERRITORIAL CONTROL 1
1. Introduction
Peace agreements are usually imperfect and far from comprehensive. They need to
address the specificities of particular conflicts, and are shaped by both internal and
external political constraints (Doyle and Sambanis, 2000). This constitutes a magnif-
icent challenge. Indeed, the concept of ‘peace’ goes well beyond the absence of war
and should “incorporate the conditions under which states have little need or incentive
to use violence against their citizens, and conversely citizens have little motivation or
incentive to challenge the state by force of arms” (Regan, 2014).
The limitations of peacemaking are likely to be exacerbated when, in internal con-
flicts with multiple actors, peace deals are made with only a fraction of the active
armed groups. In such circumstances, there is no guarantee that violence will end, and
in fact conflict levels may even increase.1 The fragility and short duration of peace in
countries such as the Democratic Republic of Congo constitutes an eloquent example
of this observation.2
This paper studies the unintended consequences of partial peacemaking in Colombia.
After over five decades of civil war, at the end of 2016 the government signed a peace
agreement with the country’s largest and oldest guerrilla, the Revolutionary Armed
Forces of Colombia (FARC from the Spanish acronym). While the conflict with FARC
ended as a result of the agreement, other groups such as the National Liberation Army
(ELN from the Spanish acronym), criminal bands of former paramilitary groups, and
FARC dissidences that opposed an agreement with the government, were excluded from
the negotiations. Moreover, as government forces largely failed to occupy and build
institutional capacity in FARC’s former strongholds, a vacuum of power was created in
these valuable territories that other armed groups rushed to fill. Looking at the recent
experience of Colombia as a case study, we argue that incomplete peace building efforts
can have unintended negative consequences.3
Specifically, we study the systematic killing of local social leaders that has taken
place in Colombia in the last few years. From January 2009 to December 2017 over 550
1
Franke and Öztürk (2015) and König et al. (2017) show theoretically that, when there are more than
two parties involved in conflicts with complex network structures, partial peace deals may backfire.
2
See DR Congo: Peace Process Fragile, Civilians at Risk, Human Rights
Watch, 07/27/2008. Available from: https://www.hrw.org/news/2008/07/27/
dr-congo-peace-process-fragile-civilians-risk-updated-version-august-28-2008 (last
accessed June 22, 2018).
3
Steele and Schubiger (2018) show that past attempts to negotiate peace in Colombia under the
Betancur administration (1982-1986) also backfired and led to more violence.
social leaders were killed in Colombia.4 Using a triple differences strategy, we find that
the permanent ceasefire introduced at the end of 2014 by FARC in the context of the
peace negotiations encouraged the targeting of local community leaders by illegal armed
groups not involved in the peace process, who sought to occupy valuable territories,
previously controlled by FARC.5
The final peace agreement, signed at the end of 2016, is probably the most important
political achievement of a country that faced over five decades of internal armed conflict.
However, the simultaneous unprecedented surge in the assassination of social leaders,
which we study in this paper, casted shadow over the euphoria generated by the end
of the conflict with FARC.
A vast academic literature agrees that state capacity is an important determinant
of economic development.6 Thus, a relevant question is how to build capable states. A
key element of state formation, emphasized at least since the work of Weber (1946),
is the consolidation of the monopoly of violence within a given territory (see also
Huntington, 1968; Tilly, 1990). This function, however, is not limited to a central
institutionalized authority. Specifically, in the context of internal conflict, non-state
actors with long-term horizons can also establish social order within specific peripheral
territories, regulating most public and private affairs and enforcing specific commands,
thus de facto establishing a local “state” (Arjona, 2016).
In this type of situations, the de facto withdrawal of the ruling actor –due for instance
to a peace agreement followed by disarmament or a permanent ceasefire – generates
a vacuum of power that other armed groups often rush in to fill. In turn, consoli-
dating territorial dominance often entails the use of violence against civilians due to
their strategic role in sharing information, providing resources and services, mobilizing
supporters to increase a group’s military strength, prevent defections, etc. (Kalyvas,
2006).7 Winning over the cooperation of the local population thus encourages the use
4
Social leaders are local activists representing citizens’ organizations of various sources, including local
community councils, ethnic groups, unionists, environmental activists, etc. Section 3 describes this
variable as well as all other variables used in this study.
5
The ceasefire was declared on December 20th 2014 and resulted from the peace negotiations that
FARC held with the Government of Colombia since October 2012. The ceasefire was largely met and
extremely comprehensive, as it included the avoidance of disputes against other illegal armed groups.
6
See for instance Amsden (1992), Evans (1995), Migdal (1998), Herbst (2000), Gennaioli and Rainer
(2007), Acemoglu (2005), Besley and Persson (2011, 2009), Acemoglu and Robinson (2012) and Ace-
moglu et al. (2015).
7
Other incentives may explain the killing of civilians in different contexts. For instance, Alesina et al.
(forthcoming) and Robinson and Torvik (2009) emphasize the electoral incentives that motivate the
use of violence against civilians.
of selective civilian killings.8 Indiscriminate violence is usually ineffective in achieving

this objective as it generally backfires.9
Consistent with this, we show that the killing of social leaders increased dispropor-
tionally after the start of FARC’s permanent ceasefire in places previously dominated
by this insurgency and located in the proximity of areas with presence of other armed
groups. The killing of social leaders is not driven by a differential trend of the overall
homicide rate, and thus it is not explained by either a strategy of indiscriminate killings
of civilians or a differential change of reporting rates in previously FARC-controlled ar-
eas after the ceasefire.10 In addition, we show that the killing of leaders is exacerbated
in areas with a weaker state capacity in the form of an inefficient local judiciary, and
in places with demands for land restitution.11 Our results are driven by the window of
opportunity for territorial control given by the permanent ceasefire, but are not exac-
erbated (or attenuated) during the implementation stage of the peace agreement that
started at the beginning of 2017.
This paper contributes to the existing literature in several ways. First, we emphasize
how peace agreements may backfire if they generate territorial vacuums of power that
are not quickly filled by the legitimate state. In particular, our results suggest that
partial pacification processes can exacerbate violence by other existing armed groups,
aimed at controlling pacified territories. Indeed, the killing of social leaders in Colombia
has largely undermined the legitimacy of the peace agreement. Second, we identify
how civilians may be differentially targeted according to their role in society. Our data
allows us to identify the scope of selective targeting according to the victims’ activities.
Specifically, we study the killing of local community activists, who represent vulnerable
groups. Third, we contribute to the recent literature about the factors that help
8
This has been recognized at least since the work of Galula (1964), Clutterbuck (1966) and Thompson
(1966). See also, Kalyvas (2006); Humpreys and Weinstein (2006); Mason and Krane (1989); Goodwin
(2001); Wood (2003); Downes (2007); Kalyvas and Kocher (2007); Lyall (2009); Kocher et al. (2011);
Condra and Shapiro (2012); Lyall and Imai (2013); Christensen et al. (forthcoming); and Toft and
Zhukov (2015) among several scholars who have studies this phenomenon. Berman and Matanoc
(2015) provide a recent thorough review.
9
According to Eck and Hultman (2007), most civilian killings are deliberately planned by both state
and non-state actors, and by and large are not the result of indiscriminate killings or collateral damage.
See Steele (2009) for a discussion on collective targeting compared to selective and indiscriminate
killings in terms of different types of strategic violence that armed groups can opt for against civilians.
10
Within a year of the peace agreement, Colombia’s homicide rate fell to its lowest level since 1975.
11
A salient feature of the Colombian conflict has been the dispossession of land to smallholders and
peasants from right-wing paramilitary groups. This has resulted in high levels of land concentration
in conflict-affected areas. In the context of Law 1448 of 2011 (called the victims and land restitution
law), formerly dispossessed agents can claim back their land, and the law places the burden of the
proof on current title holders.
the success of violence reduction security programs12 by exploring what mechanisms

exacerbate or attenuate the killing of social leaders following the cease fire. Specifically,
we emphasize the importance of state capacity, judicial effectiveness and well-specified
land property rights in reducing the incentives of other armed groups to target social
leaders for territorial domination. Fourth, our paper also relates to the literature
studying how policies aimed at reducing illegal activities can increase violence in the
form of armed territorial disputes (see for example Werb et al., 2011 and Dell, 2015).
We also contribute to the literature that highlights how changes in the expected rents
of armed groups can exacerbate conflict (Angrist and Kugler, 2008). In our case the
cost of controlling a previously FARC controlled area is reduced after the ceasefire,
making these strtegic territories more attractive to other armed groups.
The rest of the paper is organized as follows. Section 2 provides some context on the
Colombian conflict and the recent peace process, section 3 describes the data sources,
section 4 discusses the identification strategy, section 5 reports the main results, ro-
bustness and potential mechanisms and section 6 concludes.
2. Background
2.1. The Colombian conflict and the recent peace process. The Colombian civil
war started with the foundation of left-wing guerrillas FARC and ELN in the mid 1960s.
Both groups claim to represent the rural poor and have fought for over 50 years with
the stated aim of overthrowing the government. In order to finance the protracted war,
both groups have been profiting from several forms of illegal activities localized within
the Colombian territory (Richani, 1997). This implies that sub-national territorial
dominance is an important intermediate objective of the armed groups.
The conflict was a Cold War proxy until the end of the 1980s, but escalated during
the 1990s fueled by the involvement of the guerrillas in illegal drug trafficking and the
consolidation of right wing paramilitary groups. The formation of paramilitary groups
dates back to the late 1960s. As part of the war against “internal enemies,” the US
National Security Doctrine legitimized the military as the force ultimately responsible
for security and development in Latin America. In Colombia, this encouraged the
enactment of Decree 3398 of 1965 and Law 48 of 1968, which allowed civilians to be
trained and armed by the military to fight the newly created communist insurgencies.13
12
See for instance Fearon et al. (2009); Berman et al. (2013); Blair et al. (2018)
13
An additional small number of paramilitary groups emerged as self-defense forces, organized by rural
elites to oppose guerrilla extortion.
In the mid 1990s, the paramilitaries effectively became a third force in the conflict,
when splintered paramilitary armies colluded under the umbrella organization of the
United Self-Defense Groups of Colombia (AUC by its Spanish acronym). Through
the end of the 1990s and the first half of the 2000s, the counterinsurgency strategy
of paramilitaries was based on perpetrating massacres targeted at civilians, thought
to constitute the local ‘infrastructure of guerrillas’ (Restrepo et al., 2004; Aranguren,
2001).
In 2003, the AUC declared a partial ceasefire and a started conversations with the
government of Alvaro Uribe that ultimately led to its demobilization between 2005
and 2007. However, this process did not effectively disarm all paramilitary units of the
country, as some factions did not lay down all their weapons. In many aspects, the
reintegration of former paramilitaries was largely ceremonial, and many combatants
reconvened to form criminal bands and neo-paramilitary groups, that exist to date
(Human Rights Watch, 2010).
In October 2012 the Colombian government and FARC started peace negotiations in
Havana, with the oversight of the Norwegian and Cuban government. While the four-
year long process was characterized by constant ebb and flow, one of the most significant
milestones was the establishment of a permanent ceasefire by FARC on December 20th,
2014. In fact, as a result of the ceasefire, FARC withdrew their troops to more remote
areas where military contact with government security forces and other armed groups
was unlikely to take place. Likewise, although the ceasefire involved primarily the
government security forces, a clash with another armed structure, in the midst of a
ceasefire, would have impacted the negotiations and the public opinion greatly. This
explains why FARC’s offensive activities drop by 98% during this period (CERAC,
2016). Indeed, the ceasefire was largely met until followed by the bilateral definitive
ceasefire and then by the final disarmament in 2016.14
We argue that FARC’s inability to respond violently during the ceasefire constituted
a window of opportunity for other armed groups (specifically the ELN and former
paramilitary criminal bands) to try to establish their dominance in previously FARC-
controlled territories.
The 5-decade long, three-sided Colombian conflict resulted in the largest number
internally displaced persons (IDPs) globally (United Nations High Commissioner for
Refugees, 2017). To date, over 8.5 million people are formally registered with the
14
The final peace agreement was endorsed by Congress on November 30th, 2016 after a previous
version of it got rejected by a 0.5% vote margin in a referendum that took place on October 2nd that
year.
state as victims of the conflict.15 Local social leaders have contributed their share to
these figures. Given the high number of social leaders assassinated during the armed
conflict, the Victims’ Unit (a government’s institution in charge of keeping a registry
and providing assistance and reparations to victims) established a specific program
for the collective reparation of social groups systematically victimized through killings,
threats, and forced disappearances. This included unionists, journalists, and human
rights defenders.
2.2. The killing of social leaders in Colombia. The persecution of social leaders
dates back at least to the emergence of the paramilitary groups, when leaders were
seen as an instrument of the communits “subversion” (Gallón et al., 2013). Steele
(2017) argues that, historically, leaders were selectively assassinated in Colombia at
the same time that “regular” people were collectively displaced as complement strate-
gies used especially by paramilitaries to facilitate territorial control.16 During Uribe’s
administration (2002-2010) violence against union members was at the center of the
public debate. For instance, union leaders argued against the free trade agreement
with the US because of the incapacity of the government to curbing violence against
trade unionists.17 The government, in turn, argued that most of the violence was a
byproduct of the armed conflict (Mejı́a and Uribe, 2011).
The targeting of social leaders was exacerbated by the territorial dispute triggered by
FARC’s de facto withdrawal from its former strongholds after the start of the perma-
nent ceasefire.18 This encouraged community leaders and activists to raise their voices
15
Source: Victims’ Registry, from the Unit for the Victims Assistance and Reparation (herein the
Victims’ Unit), March 2018 figure (https://www.unidadvictimas.gov.co/en).
16
One important difference with the wave of killings studied in this paper is that, during the late 1980s
and the first half of the 1990s most of the targeting wast directed at a specific (and quite visible) group,
namely members of the Patriotic Union (UP from the Spanish acronym) political party. A second
important difference is that the current wave of killings has taken place after the ceasefire and posterior
demobilization of FARC, which took place for the first time since the group was founded in the mid
1960s. This implies that the killings are likely not driven by the alleged support of civilian leaders to
an insurgent opponent.
17
This argument was even picked up by US democratic presidential candidate
Barack Obama in the final presidential debate with John McCain, to oppose the
free trade agreement with Colombia. See: “Candidates Obama and McCain dis-
agree on Colombian FTA agreement in final presidential debate”, Semana magazine,
10/16/2008, available from https://www.semana.com/international/headlines/articulo/
candidates-obama-and-mccain-disagree-on-colombian-fta-agreement-in-final-presidential-debate/
96274-3 (last accessed June 2018).
18
In recent years, almost all of the killings have taken place in the regions that FARC abandoned.
See “Peacetime Spells Death for Colombia?s Activists”, by Nicholas Casey. Published by The New
York Times on 10/13/2028. Available from: https://nyti.ms/2QQp2Rb (last accessed November 30,
2018).
to demand basic services and infrastructure from the government as they thought the
conflict has ended. Formerly dispossessed peasants who fled the conflict also returned
to claim their lands and re-unite with family and friends. But the central state failed to
take control over these areas and brought neither development projects nor security.19
Instead, other armed groups stepped in to replace FARC’s rule and take over its il-
legal activities. Local leaders and activists constitute a threat for the interests of these
groups as they mobilize people, attract attention, and demand services that would ulti-
mately increase access and state presence in their territories. This is especially the case
of the leaders of local community councils, which constitute the primary organizational
structure of local communities in both rural hamlets and urban neighborhoods. These
councils are the main intermediary between peasant and the government, and are key
in helping the state implementing micro-level policies, including those agreed in the
peace settlement with FARC, such as illegal crops substitution and the promotion of
local development initiatives. Council leaders channel the demands of the community,
oversee the execution of projects and report cases of corruption and criminal activity
affecting the community.20
Since the start of the ceasefire, the category of leaders that has been targeted the
most are local community council leaders (37%), and most of them have been killed in
former FARC territories disputed by other armed groups. The next section describes
the source of these and other data used in our empirical analysis.
3. Data
3.1. Killing of social leaders. The killings of social leaders comes from a Colombian
Human Right NGO called Somos Defensores. This NGO was created in 1999 with
19
For instance, the Special Administrative Unit for Territorial Consolidation (UACT from its Spanish
acronym) was created in 2011 with the objective of promoting the institutional presence of the state
in the territories formerly controlled by armed groups, in order to foster their consolidation and
reconstruction. However, UACT never had enough leverage to address its mission. Shapiro et al.
(2019) evaluate the impacts of the national territorial consolidation policy and find no significant
effects on almost any outcome.
20
The Ministry of Interior estimates that the circa 64,000 local community councils in Colom-
bia affiliate about 7 million people. Councils are present throughout the entire country except
in the indigenous territories, which feature other type of organizations (albeit also targeted by
armed groups seeking territorial control). See “Si no protegen a los lı́deres comunales el Acuerdo
de Paz fracasa”, La Silla Vacı́a, 08/13/2018. Available from: https://m.lasillavacia.com/
si-no-protegen-los-lideres-comunales-el-acuerdo-de-paz-fracasa-67442?utm_source=
newsletter&utm_medium=email&utm_campaign=Las2520cuatro2520patas2520de2520La2520Silla
(last accessed November 30, 2018).
the aim of protecting social leaders in Colombia by reporting the abuses targeted at
them in the context of the armed conflict. A social leader is defined by the NGO as
an activist that represents the interests of local vulnerable communities. Social leaders
include local community council members, representatives of ethnic (indigenous and
Afro-Colombian) communities, unionists and environmental advocates among others.
Since 2006, Somos Defensores created an information system that records all the
killings of social leaders, with the objective of producing permanent statistics about
this type of violence in order to lobby national authorities and generate awareness on
what they call a systematic (and intentional) practice.21 The registry is filled with
the input of a large network of Human Rights organizations (over 500) with presence
throughout the Colombian territory (especially in conflict-affected areas) and supple-
mented with fieldwork carried out by Somos Defensores to verify that assassinations
of alleged leaders are indeed so. Efforts are made to avoid double counting.
Somos Defensores uses these data to publish bi-annual reports with details of each
one event. For each murder case the report includes: the date and place of the event,
the victim’s name, the organization represented by the leader, and the presumed per-
petrator. The NGO records a total of 563 killings up to the first semester of 2018,
91% of which remain unresolved by the judicial system.22 Most of the murdered lead-
ers were part of local community councils (33%), indigenous communities (22%), or
peasant organizations (12%), see table 1. For the purpose of our statistical analysis,
we aggregate this information at the municipality-bi-annual level.
Our analysis covers the period 2011:1 to 2017:2, since the start Juan Manuel Santos’
presidential term.23 During this period, 490 leaders were murdered (35 per semester).
Before the ceasefire (until 2014:2), 250 killing cases are recorded (31 per semester).
After the ceasefire there are 240 cases recorded (40 per semester). This increase can
be seen in Panel A of Figure 1, that shows the evolution of the number of leaders
killed during our sample period. In turn, Figure 2 presents the spatial distribution of
assassinations by municipality during the entire period of analysis. Overall, killings
are concentrated in the periphery of the country, in places relatively far from the big
cities and characterized by a rather weak presence of the state. This is consistent with
21
See Programa Somos Defensores - PNGPDDH (2008).
22
See “La mayorı́a de asesinatos de lı́deres sociales quedan impunes: ONG Somos Defensores”, El
Espectador, 11/13/2018. Available from: https://www.elespectador.com/noticias/judicial/
la-mayoria-de-asesinatos-de-lideres-sociales-quedan-impunes-ong-somos-defensores-articulo-823451
(last accesses November 30, 2018).
23
Event-level data is not available after the second semester of 2017.
our interpretation that leaders are targeted in areas that are being violently disputed
by armed groups after the de facto withdrawal of FARC.
3.2. Armed groups presence and exposure. Turning to our measures of armed
groups presence, we use the violence dataset originally compiled by Restrepo et al.
(2004), and updated through 2014 by Universidad del Rosario. This dataset codes
violent events recorded in the Noche y Niebla reports from the NGO Centro de Inves-
tigación y Educación Popular (CINEP) of the Company of Jesus in Colombia, which
provides a detailed description of the violent event, date, the municipality in which
it occurred, the identity of the perpetrator, and the count of victims involved in the
incident.24 Specifically, we create a dummy for FARC presence if there was at least one
violent case by FARC in the period 2011:1–2014:2, after president Juan Manuel Santos
took office and before the beginning of the ceasefire.
Measuring the influence exercised by an armed group over a specific location is
extremely challenging. Indicators of presence and non-violent coercion over a large
set of municipalities cannot be systematically recorded in an objective way. Violence,
on the other hand, while more easily observed, is only imperfectly correlated with
territorial dominance. However, non-violent dominance is unlikely to occur without any
violence inflicted in the past, either as a way to legitimize influence with the citizenry or
to oust any contesting (legal or illegal) group. It is thus reasonable to assume that the
ability to inflict localized violence over a certain period could be expected to translate
into influence in different ways. We thus follow a growing empirical literature on the
Colombian conflict (see e.g. Ch et al., 2018; Acemoglu et al., 2013; Fergusson et al.,
2018a,b), and use past violence over a period of years as an (imperfect) indicator of
influence.25
To measure the intention of other armed groups to dispute the control of a specific
area, we follow Acemoglu et al. (2015) to create a measure of exposure to other armed
groups (neo-paramilitary criminal bands and the ELN guerrilla). This is obtained from
the interaction of a presence dummy equivalent to that of FARC (and during the same
24
Noche y Niebla sources include (Restrepo et al. 2004, p. 404) “1. Press articles from more than 20
daily newspapers of both national and regional coverage. 2. Reports gathered directly by members
of human rights NGOs and other organizations on the ground such as local public ombudsmen and,
particularly, the clergy.” Notably, since the Catholic Church is present in even the most remote areas
of Colombia, we have extensive coverage of violent events across the entire country.
25
Arjona and Otálora (2011) compare existing databases of civil war violence in Colombia to survey
evidence on armed groups’ presence (for the small subsample of municipalities for which the latter is
available) and conclude that while violence is likely to underestimate –by roughly the same magnitude-
both guerrilla and paramilitary control, there is a non-negligible correlation between both measures.
period, 2011:1–2014:2), and a vector of (distance-penalized) neighboring municipalities.

Specifically, we define the matrix F with entries fij as:
1
(3.1) fij =
1 + dij (1 + eij )
where dij is the euclidean distance between municipalities i and j, and eij is the dif-
ference in altitude between municipality i and j (used to compute the an “effective
distance”, corrected by terrain ruggedness). Let N (i) be the set of neighbors that
share a common border with municipality i and constitutes the ith row of the matrix
N, which summarizes all the neighbors of every municipality and has entries:
Y
] 0 if j œ ” N (i)
(3.2) nij =
[ fij if j œ N (i)
Note that the resulting measure varies between 0 an 1 and accounts for the differential
exposure of other armed groups given the neighborhood of a specific municipality.
Using these definitions we end up with a sample composed by 129 municipalities with
FARC presence, 172 municipalities with presence of other armed groups, and 49 with
presence of both FARC and other groups. In addition, using the distance-penalized
neighboring criterion, there are 564 municipalities exposed to the influence of other
armed groups, 92 of which have FARC presence.
3.3. Other data. We complement these data with a large set of municipality-level
characteristics from an annual panel constructed by Centro de Estudios sobre De-
sarrollo Económico (CEDE) at Universidad de los Andes. This dataset includes so-
cioeconomic and geographical information for all the municipalities in Colombia. We
gathered information on population, presence of coca plantations, altitude, size of the
municipality, distance to the closest mayor city, tax revenue, an index for sound fiscal
policy, literacy rate, and an index of rurality.
Table 2 presents summary statistics for our sample of 1,069 municipalities that in-
cludes all municipalities with a population of less than 200,000. We drop mayor cities
and capitals that are mainly urban and less affected by the conflict.26 On a given
semester 2.1% of the municipalities experience the murder of a social leader, and the
26All our results are robust to including all Colombian municipalities.
homicide rate (per 1,000 inhabitants) of this group is on average 0.11 in a given se-
mester.27 In our sample 9% of the municipalities are exposed to FARC, 52% to other
armed groups, and 8% to both of them.
We also present differences in observables between type of armed group presence
at the municipality level before the ceasefire in Table 3. We focus our attention on
differences between municipalities with only FARC presence and municipalities with
both FARC and other armed groups’ exposure (column 3). We find that in general
both types of municipalities are similar in terms of geographic and socioeconomic char-
acteristics before the ceasefire. There is some evidence that municipalities with FARC
presence only had more killings of social leaders before the ceasefire.
Finally we split the evolution of killings by type of armed group presence since
2011. We divide the municipalities in two groups: presence of both FARC and other
groups and presence of FARC only (see Figure 1 panel B). In general, we do not
see any differential time pattern between these two types of municipalities before the
ceasefire. However, there is a large increase in the number of killings in municipalities
with presence of both FARC and other armed groups after the ceasefire. This already
suggests that FARC areas exposed to the influence of other armed groups experienced
an increase in killings after the ceasefire. The next section describes how we explore
this idea more formally.
4. Empirical strategy
4.1. Main specification. Our identification strategy exploits the timing of the per-
manent ceasefire announced by FARC on December 20, 2014, during the peace nego-
tiations with the Colombian government, and the spatial distribution of illegal armed
groups in Colombia prior to the ceasefire. Since we are interested in how the killing of
social leaders changed after the ceasefire in places with FARC presence that, in addi-
tion, are exposed to the influence of other armed groups, the main empirical strategy
is based on a difference-in-difference-in-differences or triple differences model. More
formally, using the subindex m to denote municipalities and t to denote time, we
estimate:
27We do not have the universe of social leaders at the municipal level, so we use as denominator the
total municipal population.
(4.1)
ymt =–m + ”t + —1 ◊ FARCm ◊ ExposureOthersm ◊ Ceaset + —2 ◊ FARCm ◊ Ceaset
ÿ
+ —3 ◊ ExposureOthersm ◊ Ceaset + (c ◊ –t ) + ‘mt
cœXm
where ymt is one of our measures of leaders killed,28 FARCm is a dummy that takes
the value one for municipalities with FARC presence as measured before the ceasefire,
and ExposureOthersm is our measure of exposure to other armed groups which, as
explained in the previous section, comes from the interaction of a dummy of presence
of other armed groups and the vector Nm of distance-penalized vicinity. Ceaset is a
dummy that takes the value one after the start of the permanent ceasefire, in the first
semester of 2015. –m and ”t are municipal and time fixed effects that capture any time-
invariant municipal-level heterogeneity and any aggregate time shock, respectively. Xm
are municipality characteristics measured before the ceasefire that we interact with the
time fixed effects to flexibly control for differential trends parametrized by each one
of the municipal attributes. Finally, the error term ‘mt is allowed to be spatially and
timely correlated, using the structure suggested by Conley (1999) and Conley (2016).
Our coefficient of interest is —1 which captures the differential change in the killing
of social leaders after the ceasefire in municipalities with FARC presence and that are
exposed to the influence of other armed groups, relative to the change in municipalities
with only FARC presence (but not exposed) or in municipalities exposed (but without
FARC presence), taking into account: i) any differential effects driven by fixed mu-
nicipality characteristics over time; ii) any aggregate time shock; and iii) differential
municipal trends based on a large set of pre-treatment characteristics. The main iden-
tification assumption is that, in the absence of the ceasefire, the killing of social leaders
in municipalities with FARC presence and exposed to other armed groups would have
evolved in a similar way than the killing of leaders in other municipalities.
This “parallel trends” assumption can be asessed by estimating following dynamic
version of (4.1):
ÿ ÿ
(4.2) ymt =–m + ”t + FARCm ◊ ExposureOthersm ◊ ”j1 + FARCm ◊ ”j2
jœT jœT
ÿ ÿ
+ ExposureOthersm ◊ ”j3 + (c ◊ ”t ) + ‘mt
jœT cœXm
28These include the total number of killings, a dummy variable for any leader being killed in a
municipality, or the rate of killings per 100,000 municipal inhabitants.
where T includes all semester of our sample period but the second semester of 2014,
which is the period right before the ceasefire. The parameters ”j1 can be interpreted as
the differential killings in municipalities with FARC presence that are exposed to the
influence of other armed groups in year-semester j, relative to the year-semester right
before the ceasefire.
The signature of the final peace agreement at the end of 2016 was followed by a
mobilization of FARC fronts away from their areas of operation and into specific zones
where, under the monitoring of a UN Peace Mission, FARC disarmed and started their
reincorporation process. Thus, one could argue that, while the permanent ceasefire
opened a window of opportunity for other armed groups to dispute the control of
FARC-dominated territories, the mobilization of FARC personnel during the agreement
implementation stage further facilitated the occupation of former FARC areas by other
armed groups. If this is the case, then there should be a differential effect on the killing
of social leaders starting in 2017:1, relative to that observed during the post ceasefire,
pre-implementation period (2015:1 to 2016:2).
However, as mentioned in the introduction, the quest for territorial control by armed
groups entails the selective killing of civilians to induce fear and encourage allegiance
and support, and this strategy is independent of whether a ceasefire-compliant FARC
is present or not. This, on the other hand, argues against any differential effect in
the killing of social leaders after the implementation of the agreement relative to the
ceasefire period. We take a skeptical view and estimate this potential differential effect
across the post cease-fire period through the following model:
(4.3)
ymt = –m + ”t + —1 ◊ FARCm ◊ ExposureOthersm ◊ Implementationt
+ —2 ◊ FARCm ◊ Implementationt + —3 ◊ ExposureOthersm ◊ Implementationt
+ —4 ◊ FARCm ◊ ExposureOthersm ◊ Ceaset + —5 ◊ FARCm ◊ Ceaset
ÿ
+ —6 ◊ ExposureOthersm ◊ Ceaset + (c ◊ ”t ) + ‘mt
cœXm
where Implementationt is a dummy that takes the value of one after beginning of the
peace agreement implementation phase, in first semester of 2017. Relative to —4 , —1
captures the differential change in the killing of social leaders during the implementa-
tion stage in comparison with the ceasefire period for the interaction of interest.
4.2. Testing potential mechanisms. We can use municipal-level variation across

specific characteristics to estimate heterogenous effects that can shed some light re-
garding the underlying mechanisms of the main effect of interest. In particular, the
killing of leaders may be exacerbated in municipalities that are more economically at-
tractive for the controlling armed group. Moreover, municipalities with better state
capacity and a more effective judiciary are likely to attenuate the unintended violent
incentive provided by the ceasefire in formerly FARC strongholds. We thus divide a
set of potential mechanisms into these two categories (attractiveness and state capac-
ity) and test whether the estimated average effects entail some variation across these
dimensions.
To that end, we augment the main specification in equation (4.1) by adding a fourth
interaction term. Specifically, let the municipality characteristic Zm (measured before
the ceasefire) be a measure of the relative attractiveness or else the relative cost of
disputing a FARC stronghold. We estimate:
(4.4)
ymt = –m + ”t + —1 ◊ FARCm ◊ ExposureOthersm ◊ Zm ◊ Ceaset + —2 ◊ ExposureOthersm ◊ Zm ◊ Ceaset
+ —3 ◊ FARCm ◊ Zm ◊ Ceaset + —4 ◊ FARCm ◊ ExposureOthersm ◊ Ceaset + —5 ◊ FARCm ◊ Ceaset
ÿ
+ —6 ◊ ExposureOthersm ◊ Ceaset + —7 ◊ Zm ◊ Ceaset + (c ◊ –t ) + ‘mt
cœXm
Our coefficient of interest, —1 , captures the differential killing of social leaders in

places with FARC presence and exposed to other armed groups in municipalities with
characteristic Zm . Note that the results coming from this test are suggestive about
potential mechanisms, but not necessarily causal. They have to be interpreted with
caution.
Using the above specifications we estimate the impact of the December 2014 perma-
nent ceasefire on the killing of social leaders in areas previously dominated by FARC
and exposed to other armed groups (equation 4.1), the dynamic persistence of this effect
(equation 4.2), the potential differential effect given by the start of the implementation
of the peace agreement at the beginning of 2017 (equation 4.3), and heterogeneous
effects given by the relative attractiveness of disputed municipalities as well as their
institutional capacity (equation 4.4). The next section reports the estimated results.
5. Results
5.1. Main results. We start by describing, on Table 4, the empirical estimates of the
main specification given by regression model 4.1. Recall that our main coefficient of
interest is the (triple) interaction between a (pre-ceasefire) FARC presence indicator,
the municipal “exposure” to the influence of other armed groups –given by the (distance
penalized) vicinity of either neo-paramilitary criminal bands or ELN strongholds- and
a dummy that captures the period after the announcement of the permanent ceasefire.
We measure the killing of social leaders in different ways. Columns 1 and 2 of Table
4 compute the rate of killings by 100,000 inhabitants (of the municipality where the
death is recorded). Columns 3 and 4 use the non-normalized count of social leaders
killed. Columns 5 and 6 focus on the extensive margin, coding a dummy variable that
takes value one if at least one single leader is killed in a municipality-year.29 While
all specifications include both municipality and and time fixed effects, even columns
include all the predetermined municipal controls (described in section 3) interacted
with the time fixed effects to flexibly control by differential trends parametrized by
each one of the municipal attributes.30
In all cases, the coefficient of interest is positive and significant. This suggests
that social leaders are differentially targeted after the ceasefire in areas both formerly
controlled by FARC and exposed to other armed groups. According to the magnitude of
the estimate reported in Column 1, in places controlled by FARC prior to the ceasefire,
a one standard deviation increase in the average municipal exposure to other armed
groups (equal to 0.325, see second Panel from the top of Table 2, Column 2) increases
the rate of leaders killed by 0.11 per 100,000 inhabitants (=0.348 ◊ 0.325) after the
start of the permanent ceasefire. This effect is statistically significant at 5%, and it
implies that the pre-ceasefire mean of the social leaders’ homicide rate doubles.31
The estimate reported in Column 3 of Table 4, which focuses on the count of leaders
killed, implies that in FARC-controlled areas a one standard deviation increase in the
average municipal exposure to other armed groups increases the number of leaders killed
29This attenuates concerns about potential measurement error in the count of leaders, or the possibility
that the results are driven by a higher density of social leaders in places with FARC presence and
exposed to other armed groups.
30In Table A.1 we estimate the main regression using two models that take into account the count
nature of our dependent variable, namely Negative Binomial and Conditional Poisson models. In both
cases we find similar results to the ones presented in Table 4.
31Adding the differential trends parametrized by the predetermined controls, the equivalent estimated
coefficient reported in Column 2 of Table 4 is slightly bigger in magnitude and implies an increase in
the rate of leaders killed of 0.12 per 100,000 inhabitants (=0.377 ◊ 0.325). It is also significant at the
5% level.
by 0.03 (=0.09 ◊ 0.325) after the ceasefire. Again, this implies that the pre-ceasefire
mean of the count of leaders killed doubles.32
Finally, one potential concern given the low average of killings per semester, could
be that our results are driven by a few outliers with a large number of killings in a
given semester. To deal with this concern we present results on the extensive margin of
killings. Column 5 implies that in places where FARC was present prior to the ceasefire
a one standard deviation increase in the average municipal exposure to other armed
groups increases the probability of a leader being killed in 1.7 percentage points.33
This is equivalent to an 80% increase in the probability of any leader being killed in a
municipality pre-ceasefire and it is significant at the 10% level.34
Table 4 also reveals that in municipalities exposed to the violent influence of other
armed groups, but not previously controlled by FARC, there is a statistically significant
drop in the killing of social leaders after the start of the permanent ceasefire. More-
over, in places dominated by FARC but not exposed to other armed groups, there is no
significant differential change in the targeting of leaders. These results are consistent
with our interpretation that it is the attempt at controlling territories previously dom-
inated by FARC what drives the targeting of social leaders when the ceasefire provides
the opportunity. Moreover, this evidence is also consistent with other armed groups
substituting their violent effort to places formerly controlled by FARC and away from
other places, after the ceasefire.
To partially test the identification assumption that, in the absence of the ceasefire,
the killing of social leaders in municipalities with FARC presence exposed to other
armed groups would have evolved in a similar way than the killing of leaders in other
municipalities, and at the same time get a sense of how persistent is the differen-
tial targeting of leaders during the post ceasefire period, we present the results from
estimating equation 4.2. This is a non-parametric version of the main empirical speci-
fication (equation 4.1).
The results are shown in Figure 3, where we plot the point estimates associated
with the triple interaction of interest, together with the 95% confidence interval. The
estimates plotted in Panel A included no controls and those of Panel B include the
32Allowing for differential trends parametrized by predetermined controls does not change the magni-
tude of the estimated coefficient substantially (a 16% of a standard deviation increase in then number
of leaders killed), but it does increase statistical precision (see Column 4).
33A similar strategy was implemented by Crost et al. (2016) in an empirical setting with a low average
of incidents per month.
34When the controls are added the estimated coefficient and the significance level remain the same
(Column 6).
pre-determined controls interacted with the time fixed effects. Both cases include the
municipality and time fixed effects. In neither case are there statistically significant
coefficients in the years prior to the ceasefire, and the point estimates move around 0.
This supports our choice of our difference-in-differences empirical strategy. However,
the point estimates increase in magnitude after the start of the permanent ceasefire
(with a slight decline in the last two semesters), and most of them are statistically
significant.
5.2. Robustness. Our measure of exposure to the violent influence of other armed
groups, based on a flexible neighborhood definition proposed by Acemoglu et al. (2015),
interacts a dummy of presence of either neo-paramilitary criminal bands or the ELN
guerrilla with a vector of (distance-penalized) neighboring municipalities (see section 3
fro details). Thus, in our baseline measure a municipality m is more or less exposed to
these groups depending on whether (and how many of) its neighboring municipalities
experience their presence, and on how far is the centroid of these municipalities from
that of m (after controlling for the average slope of the land between the two centroids).
Our results are not driven by using this specific measure of exposure. On the one
hand, a simpler alternative measure defines exposure as the share of m’s neighbors
with presence of other armed groups. On the other hand, a more general measure
does not restrict the distance-penalized indicator to m’s neighbors, and instead uses
all municipalities in Colombia.
Tables A.2 and A.3 in the Appendix are equivalent to Table 4 but use these two
alternative measures of exposure, respectively. In all cases the coefficient of interest
is positive and significant.35 Overall, this is reassuring of our territorial dispute inter-
pretation, as the surge in the killing of leaders in former FARC-dominated territories
after the permanent ceasefire are driven by the exposition to other armed groups.
Our results are not driven by lumping together neo-paramilitary criminal bands and
ELN in the “other armed groups” category. These illegal armed actors have several
differences, including their political objectives and their military strategy, which ar-
guably involve different relationships with civilians.36 Importantly, however, because
of the irregular nature of Colombia’s internal conflict, controlling valuable municipali-
ties is instrumental to both groups (Berman and Matanoc, 2015).
35The only exception is when the dependent variable is the dummy of any leader killed, differential
trends parametrized according to pre-determined controls are not included, and the alternative mea-
sure of exposure is the share of neighbors with presence of other armed groups (Column 5 of Table
A.2).
36Some of these differences are discussed in section 2.
Table 5 shows the results from estimating equation 4.1, using the rate of leaders killed
as dependent variable, but including in the measure of other armed groups only the
neo-paramilitary criminal bands (Columns 1 and 2) or only the ELN guerrilla (Columns
3 and 4). Interestingly, FARC-dominated municipalities experience a differential surge
in the rate of leaders killed after the start of the permanent ceasefire when they are
exposed to the violent influence of either group, as measured separately. Moreover, in
spite of the difference in the size of the reported estimated coefficients in Table 5, the
economic magnitude of the effect is essentially equivalent.
Focusing on the even columns, which flexibly control for municipal-specific pre-
determined characteristics, we find that in places with FARC presence prior to the
ceasefire, a one standard deviation increase in the average municipal exposure to
neo-paramilitary criminal bands (to the ELN) increases the rate of leaders killed by
0.353 ◊ 0.292 = 0.10 (0.515 ◊ 0.178 = 0.09) per 100,000 inhabitants after the start of
the permanent ceasefire. Recall that this effect, which in both cases is significant at the
5% level, is equivalent to doubling the rate of leaders killed relative to its pre-period
mean.
At the end of our sample period the peace agreement was signed and its imple-
mentation started with the movement of FARC combatants to special hosting areas
for disarmament and reincorporation.37 We estimate equation 4.3 to test whether the
implementation of the peace agreement further encouraged the killing of social leaders
by armed groups taking advantage of FARC’s internal migration. This is picked up
by the triple interaction between our FARC presence measure, the exposition to other
armed groups, and a time dummy that indicates the implementation period (the two
last semesters of our sample period).
Table 6 shows the estimated coefficients from specification. The coefficient of interest
is not statistically significant which suggests no differential effect on the rate of leader
killed in this sub-period.38 We interpret this as suggesting that the ceasefire, which
was by and large respected by FARC (to credibly signal their willingness of reaching
a peace agreement) was a high enough incentive for other armed groups to dispute
the control of this group’s territorial strongholds, and there was no differential such
incentive when the implementation stage of the peace agreement started.
5.3. Potential mechanisms.

37CalledTerritorial Spaces for Training and Reincorporation, ETCR from the Spanish acronym.
38The Table also shows that the level effect for this sub-period is positive and statistically significant
when the pre-determined controls are added.
5.3.1. Type of targeted leaders. The label ‘social leader’ encompasses several different
type activists that represent different ‘communities’ and hence have different motives
and work for different causes. Table 1 lists the different leaders included in the Somos
Defensores data, such as leaders of community councils, ethnic groups, labor unions,
teachers, sexual minorities, etc. Arguably, however, not all such leaders are equally
attractive as potential targets of armed groups seeking territorial control. The quali-
tative discussion of section 2.2, in fact, implies that a specially risky category is that
of leaders of local community councils or peasants dispossessed from their land.
To investigate the effect of the ceasefire on the killing of different type of leaders, we
repeat Panel B of Figure 1 by splitting the evolution of killings into four categories:
leaders of local community councils, leaders of peasant and conflict-related organiza-
tions, leaders of ethnic (indigenous or Afro-Colombian) communities, and a residual
category of ‘other’ leaders. This is reported in Figure A.1. We note a stark increase
in the killings of the first two categories (Panels A and B) right after the start of the
permanent ceasefire, in places with both FARC presence and exposed to other armed
groups in the pre-ceasefire period (but not in places with just FARC presence). How-
ever, we do not find a similar trend in either the category of ethnic leaders or in the
residual category (Panels C and D).
A more formal test is presented on Table A.4 of the appendix, where we re-run the
baseline empirical specification and study the effect of the ceasefire on the killing of
leaders of different types, in places with both FARC presence and exposed to other
armed groups. While the estimated coefficient of interest is positive throughout, it
is only statistically significant (at the 5%) for the case of local community council
leaders. While this is consistent with the anecdotal discussion provided above, these
results have to be interpreted with caution given that splitting the dependent variable
by type of leader leaves us with very few events per category.
5.3.2. Heterogeneous effects. Having established the type of leaders that are more likely
to be targeted, we continue investigating the potential mechanisms of our main result
by studying the type of municipalities that either make it more likely that leaders are
killed, or else that help attenuate the incentive of other armed groups to selectively
kill social leaders in areas previously dominated by FARC. We do so by estimating
heterogenous effects for a range of municipal characteristics using equation 4.4.
We look at three broad set of factors: the demand for land restitution, the (lack of)
state capacity and the economic attractiveness of the municipality.
First, the lack of land property rights in rural areas has been at the heart of the
Colombian conflict since its initial stages (Albertus and Kaplan, 2012; Flores, 2014).
Traditionally, left-wing guerrillas have been in favor of communal rural lands and the
right of peasants to appropriate idle land and peripheral ‘baldios’. On the other hand,
right-wing paramilitary groups have helped local landowners and drug lords concentrate
and formalize land, often through the use of violence and intimidation (Ch et al., 2018).
Indeed, most of the victims of the armed conflict (7.4 out of 8.5 million as recorded
by the Unique Victims’ Registry) are IDPs, and many of them were dispossessed from
their land by illegal armed groups, especially the paramilitary.
Law 1448 of 2011 (known as the ‘Victims and Land Restitution Law’) provided
the legal framework for conflict victims to obtain assistance and reparations from the
government, including humanitarian aid, psychological assistance and a large set of
material reparations. This package notably includes land restitution. To facilitate the
latter, the law created the Land Restitution Unit, a Presidential special unit in charge
of receiving all the land restitution requests and of overseeing the subsequent judicial
and administrative restitution processes.39
Given the above discussion, we posit that the incentive of illegal armed groups to
dispute the territories with prior FARC dominance following the start of the ceasefire is
larger in municipalities that, since the enactment of Law 1448, have had a larger share
of land claimed for restitution. In addition, we also expect that in this case the per-
petrators are more likely to be neo-paramilitary criminal bands, which either directly
benefited or represent groups of society who benefited from earlier land dispossession.
This goes in line with some anecdotal accounts, which suggest that a non-negligible
share of social leader killed in Colombia are leaders of local community councils who
specialize in mobilizing land-dispossessed victims to claim their land.40
Column 1 of Table 7 reports the estimated coefficient of the four-way interaction
term described in equation 4.4. In the case, the potential mechanism Z, is a dummy
39Between 2012 and 2017 over 204 thousand hectares of land had been restituted (Unidad Adminis-
trativa Especial de Gestión de Restitucioón de Tierras, 2018).
40See “Si no protegen a los lı́deres comunales el Acuerdo de Paz fracasa”,
La Silla Vacı́a, 08/13/2018. Available from: https://m.lasillavacia.com/
si-no-protegen-los-lideres-comunales-el-acuerdo-de-paz-fracasa-67442?utm_source=
newsletter&utm_medium=email&utm_campaign=Las2520cuatro2520patas2520de2520La2520Silla
(last accessed November 30, 2018).
variable that equals one for municipalities with land restitution claims above the me-
dian.41 We find that municipalities with FARC presence and that are exposed to other
armed groups experienced a larger boost in the killing of social leaders after the start
of the ceasefire if they also had a relatively large number of land restitution requests.
Moreover, as expected, this heterogeneous effect is entirely driven by the killings perpe-
trated by neo-paramilitary criminal bands, and not by the ELN guerrilla.42 Again, this
result suggests that other armed groups target local leaders whose activity constitutes
a threat group’s particular interests in a specific territory.
Second, we have argued that the very nature of the peace process with FARC –
that excluded other armed groups from the negotiations- constitutes a threat to the
sustainability of the achieved ‘peace’ if not accompanied by state-led efforts to bring
its capacity to the territory and consolidate its institutional presence. This argument
can be extended to the degree of existing state capacity (prior to the ceasefire) in
the municipalities with FARC presence. Areas with existing state institutions would
make it more costly for other armed groups to take control of the vacant territories by
targeting the local populations.
We explore this idea formally by testing the extent of which different measures of
pre-determined state capacity at the municipal level attenuate the targeting of social
leaders by other armed groups in previously FARC controlled areas after the start of
the ceasefire. Unlawful violence perpetrated by armed groups can be checked either by
the military hierarchy or by other branches of the government (such as the judiciary).
We thus use a measure of local judicial inefficiency (Column 2 of Table 7) and the
distance to the nearest military unit (Column 3) as proxies of state capacity to explore
potential heterogeneous effects of our main result according municipal-level variation
in state capacity.43
The four-way interaction with judicial inefficiency measure is positive and significant,
suggesting that when the local judiciary is inefficient (an inverse proxy of state capac-
ity), illegal armed groups find it easier to get away with the killing of local community
41Specifically,we measure the intensity of the demand for land restitution using the number of requests
for land restitution at the municipal level. Our dataset includes all the requests since the creation of
the Land Restitution Unit until June 2015. However, our measure is only for the pre-ceasefire period.
42Result available upon request.
43To measure judicial inefficiency, we follow Acemoglu et al. (2018) and use data from Colombia’s
Inspector General Office, the institution in charge of disciplinary oversight of all public servants.
Based on an event-based dataset with all processes arising from complaints against public servants
from 1995 to 2010, we compute judicial inefficiency as the ratio between the number of complaints
against judicial officials in a specific municipality and all the complaints against any public servant in
that same municipality.
leaders.44 Specifically, we find that a one standard deviation increase in the level of judi-
cial inefficiency (0.08) increases the rate of leaders killed in 0.10 (=3.893◊0.08◊0.325).
This is equivalent to doubling the sample mean. Moreover, as in the case of land resti-
tution we find that the effect is mainly driven by exposure to paramilitary groups,
which are the ones that have been more involved in scandals related to co-opting local
judges (see López 2007, Ávila and López 2010).45 We do not find a significant het-
erogenous effect related to the vicinity to military units, as measured by the logarithm
of the distance between the military base and the municipal centroid.
Third, to test for differential effects based on the availability (or potential) of illegal
rents, we add estimate a heterogeneous effect based on the coca suitability of each
municipality (see Mejı́a and Restrepo 2015).46 As reported on column 4, we do not
find that more leaders are killed in places with higher cocal suitability. Moreover,
this no-result is robust to measuring coca with actual coca availability (the share of
municipal land cultivated with coca) or the availability of (legal or illegal) natural-
resource mines. Overall our results do not support the idea that the economic value of
municipalities exacerbate the killing of social leaders.
5.3.3. Indiscriminate violence as an alternative explanation. As a final attempt to iden-

tify the potential mechanism explaining our main result we investigate the effect of the
ceasefire on the aggregate homicide rate of municipalities. Our story requires that
the killing of social leaders is driven by the selective targeting of leaders so as to
thwart collective action at the local level, and not by indiscriminate municipal vio-
lence. To rule out that our results are explained by an aggregate increase in insecurity
in FARC-dominated territories exposed to other armed groups after the ceasefire, that
mechanically translates into more leaders killed, Table 8 estimates equation 4.1 using
as dependent variable the overall municipal homicide rate. The coefficient of interest,
associated with the triple interaction, is not statistically significant. This is reassuring
that social leaders are being selectively targeted by other armed groups.
6. Conclusions
Territorial contestation by armed groups in the context of civil war often involves
the selective killing of civilians. This strategy, which has been documented by a vast
literature in political science and economics, is used to encourage allegiance, as well as
44Acemoglu et al. (2018) show that one source of local judicial inefficiency is the capture by groups
with de facto political power to get away with unlawful behavior.
45Results available upon request.
46We thank the authors for sharing their data.
to achieve informal collaboration, prevent defections, mobilize supporters, and increase

military strength. In this paper, we show that the recent surge in the systematic killing
of local social leaders in Colombia can be –at least partially- explained by the vacuum
of power that FARC’s permanent ceasefire left in this group’s controlled areas, which
encouraged other illegal armed groups seeking to occupy these areas to target local
community leaders.
Our estimation strategy exploits the temporal variation given by the ceasefire as
well as the cross-sectional variation given by the presence of FARC and the exposure
to the influence of other armed groups. We do so in a triple differences model that
controls for two-way fixed effects and for differential trends parametrized by a large set
of pre-determined municipal controls.
Our results are not explained by the overall municipal homicide rate which suggests
that they are not caused by either a differential change in reporting after the ceasefire
or by a strategy of indiscriminate violence against civilians, which is in line with the
literature. In addition, we show that the killing of leaders is exacerbated in areas
with high demand for land restitution and a weaker state capacity in the form of an
inefficient local judiciary. We also show that our results are driven by the window
of opportunity for territorial control given by the permanent ceasefire, but are not
exacerbated (or attenuated) during the implementation stage of the peace agreement,
that started at the beginning of 2017.
Overall, the killing of social leaders, we argue, constitutes an unintended negative
consequence of a partial pacification process that was not accompanied by an effort
to consolidate the state control in former FARC strongholds. Despite the historical
importance and the tremendous opportunity of the peace agreement with the FARC,
the killing of social leaders may be the beginning of a new and more sophisticated stage
of social disruption in Colombia. We hope to be wrong.
References
Acemoglu, D. (2005): “Politics and economics in weak and strong states,” Journal of Monetary
Economics, Swiss National Bank Special Issue, 52, 1199–1226.
Acemoglu, D., L. Fergusson, J. A. Robinson, D. Romero, and J. F. Vargas (2018): “The
Perils of Top-down State Building: Evidence from Colombia’s False Positives,” Unpublished man-
uscript.
Acemoglu, D., C. Garcia-Jimeno, and J. A. Robinson (2015): “State capacity and economic
development: A network approach,” American Economic Review, 105, 2364–2409.
Acemoglu, D. and J. Robinson (2012): Why Nations Fail: The Origins of Power, Prosperity and
Poverty, Crown, 1st ed.
Acemoglu, D., J. A. Robinson, and R. J. Santos (2013): “The monopoly of violence: Evidence
from colombia,” Journal of the European Economic Association, 11, 5–44.
Albertus, M. and O. Kaplan (2012): “Land Reform as a Counterinsurgency Policy,” Journal of
Conflict Resolution, 57, 198–231.
Alesina, A., S. Piccolo, and P. Pinotti (forthcoming): “Organized Crime, Violence, and Poli-
tics,” Review of Economic Studies.
Amsden, A. (1992): A theory of government intervention in late industrialization, Lynne Rienner
Publishers, 53–84.
Angrist, J. D. and A. D. Kugler (2008): “Rural windfall or a new resource curse? Coca, income,
and civil conflict in Colombia,” The Review of Economics and Statistics, 90, 191–215.
Aranguren, M. (2001): Mi Confesión: Carlos Castaño Revela sus Secretos, Oveja Negra.
Arjona, A. (2016): Rebelocracy: Social Order in the Colombian Civil War, Cambridge Studies in
Comparative Politics, Cambridge University Press.
Arjona, A. M. and L. Otálora (2011): “Presencia vs. violencia: problemas de medición de la
presencia de actores armados en Colombia,” Foco Económico.
Ávila, A. F. and C. N. López (2010): Y refundaron la patria–: de cómo mafiosos y polı́ticos
reconfiguraron el estado colombiano, Debate.
Berman, E., J. Felter, J. Shapiro, and E. Troland (2013): “Modest, Secure, and Informed:
Successful Development in Conflict Zones,” American Economic Review, 103, 512–517.
Berman, E. and A. Matanoc (2015): “The Empiricists’ Insurgency,” Annual Review of Political
Science, 18, 443–464.
Besley, T. and T. Persson (2009): “State Capacity, Conflict, and Development,” Econometrica,
78.
——— (2011): “Fragile States And Development Policy,” Journal of the European Economic Associ-
ation, 9, 371–398.
Blair, R., C. Blattman, and A. Hartman (2018): “Engineering Informal Institutions: Long-run
Impacts of Alternative Dispute Resolution on Violence and Property Rights in Liberia,” Working
Paper 24482, National Bureau of Economic Research.
CERAC (2016): “Un año de desescalamiento: conflicto casi detenido, pero que se resiste a desapare-
cer,” 12.
Ch, R., A. Steele, J. Shapiro, and J. Vargas (2018): “Endogenous taxation in ongoing internal
conflict: The case of Colombia,” American Political Science Review, 112, 996–1015.
Christensen, D., M. Nguyen, and R. Sexton (forthcoming): “Strategic Violence during Democ-
ratization: Evidence from Myanmar,” World Politics.
Clutterbuck, R. (1966): The Long, Long War: Counterinsurgency in Malaya and Vietnam.,
Praeger.
Condra, L. and J. Shapiro (2012): “Who Takes the Blame? The Strategic Effects of Collateral
Damage,” American Journal of Political Science, 56, 167–187.
Conley, T. G. (1999): “GMM estimation with cross sectional dependence,” Journal of econometrics,
92, 1–45.
——— (2016): Spatial Econometrics, London: Palgrave Macmillan UK, 1–9.
Crost, B., J. H. Felter, and P. B. Johnston (2016): “Conditional cash transfers, civil conflict
and insurgent influence: Experimental evidence from the Philippines,” Journal of Development
Economics, 118, 171–182.
Dell, M. (2015): “Trafficking networks and the Mexican drug war,” American Economic Review,
105, 1738–79.
Downes, A. (2007): “Restraint or Propellant? Democracy and Civilian Fatalities in Interstate Wars,”
Journal of Conflict Resolution, 51, 872–904.
Doyle, M. W. and N. Sambanis (2000): “International peacebuilding: A theoretical and quanti-
tative analysis,” American political science review, 94, 779–801.
Eck, K. and L. Hultman (2007): “One-Sided Violence against Civilians in War: Insights from New
Fatality Data,” Journal of Peace Research, 44, 233–246.
Evans, P. (1995): Embedded Autonomy: States and Industrial Transformation, Princeton University
Press.
Fearon, J., M. Humphreys, and J. Weinstein (2009): “Modest, Secure, and Informed: Successful
Development in Conflict Zones,” American Economic Review, 99, 287–291.
Fergusson, L., J. A. Robinson, R. Torvik, and J. F. Vargas (2018a): “The Real Winner’s
Curse,” Documento CEDE No. 2017-05.
Fergusson, L., J. F. Vargas, and M. Vela (2018b): “Sunlight Disinfects? Free Media in Weak
Democracies,” Working paper Universidad El Rosario.
Flores, T. (2014): “Vertical Inequality, Land Reform, and Insurgency in Colombia,” Peace Econom-
ics, Peace Science, and Public Policy, 20, 5–31.
Franke, J. and T. Öztürk (2015): “Conflict networks,” Journal of Public Economics, 126, 104–
113.
Gallón, G., H. Rodrı́guez, and D. F. Abonı́a (2013): “Desafiando la intransigencia,” .
Galula, D. (1964): Counterinsurgency Warfare; Theory and Practice, New York: Praeger.
Gennaioli, N. and I. Rainer (2007): “The modern impact of precolonial centralization in Africa,”
Journal of Economic Growth, 12, 185–234.
Goodwin, J. (2001): No Other Way Out: States and Revolutionary Movements 1945-1991, Cam-
bridge University Press.
Herbst, J. (2000): States and Power in Africa: Comparative Lessons in Authority and Control,
Princeton University Press.
Human Rights Watch (2010): “Paramilitarie’ Heirs: The New Face of Violence in Colombia.” .
Humpreys, M. and J. Weinstein (2006): “Handling and Manhandling Civilians in Civil War,”
American Political Science Review, 100, 429–447.
Huntington, S. (1968): Political order in changing societies, Yale University Press.
Kalyvas, S. and M. Kocher (2007): “How “Fre” Is Free Riding in Civil Wars? Violence, Insur-
gency, and the Collective Action Problem,” World Politics, 59, 177–216.
Kalyvas, S. N. (2006): The Logic of Violence in Civil War, Cambridge Studies in Comparative
Politics, Cambridge University Press.
Kocher, M., T. Pepinsky, and S. Kalyvas (2011): “Aerial Bombing and Counterinsurgency in
the Vietnam War,” American Political Science Review, 55, 201–218.
König, M. D., D. Rohner, M. Thoenig, and F. Zilibotti (2017): “Networks in Conflict: Theory
and Evidence From the Great War of Africa,” 85, 1093–1132.
López, C. (2007): “La ruta de la expansión paramilitar y la transformación polı́tica de Antioquia,”
Para polı́tica: la ruta de la expansión paramilitar y los acuerdos polı́ticos, Corporación Nuevo Arco
Iris, Bogotá.
Lyall, Jason, B. G. and K. Imai (2013): “Explaining Support for Combatants during Wartime:
A Survey Experiment in Afghanistan,” American Political Science Review, 107, 679–705.
Lyall, J. (2009): “Does Indiscriminate Violence Incite Insurgent Attacks?Evidence from Chechnya,”
Journal of Conflict Resolution, 53, 331–362.
Mason, D. and D. Krane (1989): “The Political Economy of Death Squads: Toward a Theory of
the Impact of State-Sanctioned Terror,” International Studies Quarterly, 33, 175–198.
Mejı́a, D. and P. Restrepo (2015): “Bushes and Bullets: Illegal Cocaine Markets and Violence in
Colombia.” .
Mejı́a, D. and M. J. Uribe (2011): “Is violence against union members in Colombia systematic
and targeted?” Economı́a, 12, 119–154.
Migdal, P. (1998): Strong societies and weak states : state-society relations and state capabilities in
the Third World, Princeton University Press.
Programa Somos Defensores - PNGPDDH (2008): “Defender y Proteger la Vida: La Acciión
de los Defensores de Derechos Humanos en Colombia,” .
Regan, P. M. (2014): “Bringing peace back in: Presidential address to the Peace Science Society,
2013,” Conflict Management and Peace Science, 31, 345–356.
Restrepo, J., M. Spagat, and J. Vargas (2004): “The dynamics of the columbian civil conflict:
A new dataset.” Homo Oeconomicus, 21, 396–429.
Richani, N. (1997): “The Political Economy of Violence: The War-System in Colombia.” Journal of
Interamerican Studies and World Affairs, 39, 37–81.
Robinson, T. and R. Torvik (2009): “Vertical Inequality, Land Reform, and Insurgency in Colom-
bia,” American Ecnomic Review, 99, 310–315.
Shapiro, J., A. Steele, and J. Vargas (2019): “State-building at the Community Level: Evalu-
ating the Impact of Colombia’s Consolidation Program,” Unpublished manuscript.
Steele, A. (2009): “Seeking safety: Avoiding displacement and choosing destinations in civil wars,”
Journal of Peace Research, 46, 419–429.
——— (2017): Democracy and Displacement in Colombia’s Civil War, Cornell University Press.
Steele, A. and L. I. Schubiger (2018): “Democracy and civil war: The case of Colombia,” Conflict
Management and Peace Science, 35, 587–600.
Thompson, R. (1966): Defeating Communist Insurgency: The Lessons of Malaya and Vietnam.,
Praeger.
Tilly, C. (1990): Coercion, Capital, and European States, AD 990-1990, Cambridge: Basil Blackwell.
Toft, M. and Y. Zhukov (2015): “Islamists and Nationalists: Rebel Motivation and Counterin-
surgency in Russia’s North Caucasus,” American Political Science Review, 109, 222–238.
Unidad Administrativa Especial de Gestión de Restitucioón de Tierras (2018): “Informe
Preliminar a la Audiencia Pública de Rendición de Cuentas: Vigencia 2017,” .
United Nations High Commissioner for Refugees (2017): “Global Report 2017,” .
Weber, M. (1946): Politics as a Vocation, Macmillian, 26–45.
Werb, D., G. Rowell, G. Guyatt, T. Kerr, J. Montaner, and E. Wood (2011): “Effect
of drug law enforcement on drug market violence: A systematic review,” International Journal of
Drug Policy, 22, 87–94.
Wood, E. J. (2003): Insurgent Collective Action and Civil War in El Salvador, Cambridge University
Press.
Figure 1. Evolution of social leaders killings
50
40
30
Implementation
20
Ceasefire
Agreement Agreement Agreement

on Land on illegal on Justice
drug trade
10
I I I I I
Agreement Agreement
on Political on End of
Participation Conflict
Peace Negotiation
Santos II
Santos I
I
0
2011-1
2011-2
2012-1
2012-2
2013-1
2013-2
2014-1
2014-2
2015-1
2015-2
2016-1
2016-2
2017-1
2017-2
A. Aggregate
B. Split by type of municipality
Notes: This figure presents the evolution of killings of social leaders from 2010 to 2017. Panel A presents
the distribution of total cases per semester and adds the description of the peace process. In panel B we split
the sample by type of municipality, distinguishing between municipalities with FARC presence and above the
median of exposure to other armed groups and municipalities with FARC presence but not exposed. In both
panels we show one-year moving averages to smooth the data.
Figure 2. Spatial distribution of social leaders killings
0
1
2
3
4 -16
Notes: This map presents the spatial distribution of killings of social leaders for the sample 2011 to 2017.
Figure 3. Dynamic estimation and parallel trends
2
1
Coefficient
0
-1
2011h1 2011h2 2012h1 2012h2 2013h1 2013h2 2014h1 2014h2 2015h1 2015h2 2016h1 2016h2 2017h1 2017h2
Semesters
A. Without controls
1.5
1
Coefficient
.5
0
-.5
-1
2011h1 2011h2 2012h1 2012h2 2013h1 2013h2 2014h1 2014h2 2015h1 2015h2 2016h1 2016h2 2017h1 2017h2
Semesters
B. With controls
Notes: This figure presents the coefficients from our dynamic specification presented in equation (4.2). We
present the point estimates of the regression and the confidence of interval at the 95%.
Table 1. Leaders killed by activity
Activity N %
(1) (2)
Local council 165 33.3

Indigenous 109 22.0
Peasant 59 11.9
Conflict victims 47 9.5
Union member 32 6.5
Afro 23 4.7
Human rights 19 3.8
LGBT 18 3.6
Student-teacher 20 4.0
Women 3 0.6
Notes: This table shows the dis-
tribution of homicides by type of
social leader during our period of
analysis, 2011:1 to 2017:2.
Table 2. Descriptive Statistics: Time-invariant variables
Mean Std. Dev. Min Max
Social leaders killings
Dummy of any killing 0.021 0.144 0.0 1.0

Number of killings 0.026 0.189 0.0 5.0
Rate of killings 0.113 1.051 0.0 38.4
Illegal groups presence
FARC 0.093 0.290 0.0 1.0

Exposure to other armed groups 0.128 0.325 0.0 1.3
Exposure to neo-paramilitary groups 0.103 0.292 0.0 1.3
Exposure to ELN 0.035 0.178 0.0 1.2
Geographic
Altitude (Km) 1.149 0.903 0.0 3.1

Distance to main city kms 80.772 55.551 0.0 376.1
Rural share 0.579 0.229 0.0 1.0
Municipal area in km2 865.268 2996.145 15.0 65674.0
Basic socioeconomic
Log (population) 9.489 0.948 6.9 12.2

Poverty index 69.924 15.631 14.3 100.0
Literacy rate 83.661 8.447 30.0 97.8
Language test scores 47.977 2.200 38.4 57.1
Math test scores 47.863 2.694 39.4 61.7
Fiscal
Log (Tax income) 6.726 1.408 0.0 12.1

Good fiscal policy index 66.239 9.359 0.0 94.2
Notes: Control variables measure before 2010. Altitude above sea level of the urban center of each
municipality. Distance is linear distance to the state’s capital. Rural share is the percentage of population
outside urban center. Municipal area official in km2 . Total municipal population (in logs). Proportion
of people in poverty according to multidimensional index. Percentage literate population. Math and
language scores is the municipal average scores per area for high-school graduates in the official standardized
test. Tax income is municipal total amount collected taxes. Good fiscal policy index of efficiency, legal
requirements and management of the fiscal resources.
Table 3. Descriptive Statistics by illegal groups presence
Presence Presence Only FARC vs.

FARC Other Groups Both FARC and
Other Groups
(1) (2) (3)
Social leaders killings
Dummy of any killing 0.058 0.052 0.038

[0.000] [0.000] [0.024]
Number of killings 0.073 0.061 0.049
[0.000] [0.000] [0.026]
Rate of killings 0.290 0.243 0.209
[0.000] [0.000] [0.099]
Geographic
Altitude (Km) -0.276 -0.542 -0.134

[0.001] [0.000] [0.436]
Distance to main city kms 13.673 31.071 25.678
[0.023] [0.000] [0.057]
Rural share 0.062 -0.099 -0.004
[0.006] [0.000] [0.944]
Municipal area in km2 1900.961 833.764 1140.908
[0.000] [0.001] [0.149]
Basic socioeconomic
Log (population) 0.552 0.760 0.322

[0.000] [0.000] [0.099]
Poverty index 9.026 5.026 7.552
[0.000] [0.001] [0.016]
Literacy rate -5.038 -3.873 -7.021
[0.000] [0.000] [0.000]
Language test scores -0.765 -0.888 -0.973
[0.001] [0.000] [0.014]
Math test scores -1.241 -0.774 -0.971
[0.000] [0.003] [0.061]
Fiscal
Log (Tax income) 0.363 0.557 0.250

[0.002] [0.001] [0.346]
Good fiscal policy index 0.184 0.263 -0.831
[0.813] [0.826] [0.560]
Notes: Control variables measured before 2010 and social leader killings before 2014:2. Column 2 reports
the differences between municipalities with FARC presence and municipalities with no presence of any
group. Column 2 reports differences between municipalities with presence of other armed groups and
municipalities with no presence of any group. Column 3 reports differences between municipalities with
presence of FARC only and municipalities with presence of both FARC and other armed groups. p-value
in square brakets.
Table 4. Killing of social leaders, FARC presence, and exposure to

other armed groups
Killing rate Number of killings Any killing

(1) (2) (3) (4) (5) (6)
Cease ◊ FARC ◊ ExposureOthers 0.348úú 0.377úú 0.090úú 0.095úúú 0.052ú 0.055ú

(0.149) (0.154) (0.037) (0.036) (0.030) (0.030)
Cease ◊ FARC ≠0.113 ≠0.144 ≠0.012 ≠0.020 0.001 ≠0.006
(0.098) (0.102) (0.014) (0.015) (0.012) (0.013)
Cease ◊ ExposureOthers ≠0.178úúú ≠0.225úúú ≠0.032úúú ≠0.042úúú ≠0.018úú ≠0.027úúú
(0.060) (0.067) (0.011) (0.012) (0.009) (0.009)
Observations 14966 14966 14966 14966 14966 14966

Municipalities 1069 1069 1069 1069 1069 1069
Municipality FE X X X X X X
Period FE X X X X X X
Controls X X X
Avg Dep Var 0.101 0.101 0.028 0.028 0.021 0.021
SD Dep Var 1.083 1.083 0.219 0.219 0.144 0.144
Notes: This table presents the results from the main specification in equation (4.1). We exploit municipal-
level variation by semester, over the period 2011:1 to 2017:2. Columns (1) and (2) use the number of
homicides of social leaders over total population, columns (3) and (4) use as dependent variable the total
number of homicides of social leaders, while columns (5) and (6) use a dummy that takes the value one
if there was at least one social leader assassinated. Cease is a dummy that takes the value one for the
period after 2015:1. FARC is a dummy for those municipalities with FARC presence. ExposureOthers
is a continuous variable that measures ELN or paramilitary groups presence in the municipality or their
(distance-penalized) vicinity. Predetermined municipal controls includes logarithm of the population in
2010, municipality area, average elevation, distance to the closest major city, share of population under
poverty, literacy rate, math and language test scores, index of rurality, log of tax income and index of good
fiscal policy. Errors in parentheses control for spatial and first-order time correlation (see Conley, 1999,
Conley, 2016). We allow spatial correlation to extend to up to 279 km from each municipality’s centroid to
ensure that each municipality has at least one neighbor. * is significant at the 10% level, ** is significant
at the 5% level, *** is significant at the 1% level.
Table 5. Killing of social leaders by exposure to different armed groups
Neo-Paramilitary ELN
(1) (2) (3) (4)
Cease ◊ FARC ◊ ExposureOthers 0.330úú 0.353úú 0.483úú 0.515úú

(0.165) (0.168) (0.230) (0.240)
Cease ◊ FARC ≠0.099 ≠0.144 ≠0.061 ≠0.088
(0.092) (0.096) (0.080) (0.083)
Cease ◊ ExposureOthers ≠0.180úúú ≠0.225úúú ≠0.293 ≠0.300
(0.068) (0.077) (0.189) (0.192)
Observations 14966 14966 14966 14966

Municipalities 1069 1069 1069 1069
Municipality FE X X X X
Period FE X X X X
Controls X X
Avg Dep Var 0.101 0.101 0.101 0.101
SD Dep Var 1.083 1.083 1.083 1.083
level variation by semester, over the period 2011:1 to 2017:2. The dependent variable is the number of
homicides of social leaders over total population. Cease is a dummy that takes the value one for the period
after 2015:1. FARC is a dummy for those municipalities with FARC presence. In columns (1) and (2)
ExposureOthers is a continuous variable that measures paramilitary groups presence in the municipality
or their (distance-penalized) vicinity, while in columns (3) and (4) is a continuous variable that measures
ELN presence. Predetermined municipal controls includes logarithm of the population in 2010, municipality
area, average elevation, distance to the closest major city, share of population under poverty, literacy rate,
math and language test scores, index of rurality, log of tax income and index of good fiscal policy. Errors
in parentheses control for spatial and first-order time correlation (see Conley, 1999, Conley, 2016). We
allow spatial correlation to extend to up to 279 km from each municipality’s centroid to ensure that each
municipality has at least one neighbor. * is significant at the 10% level, ** is significant at the 5% level,
*** is significant at the 1% level.
Table 6. Killing of social leaders during the cease fire and the imple-
mentation of the peace agreement
Killing rate
(1) (2)
Implementation ◊ FARC ◊ ExposureOthers ≠0.0121 0.0212

(0.2159) (0.2171)
Cease ◊ FARC ◊ ExposureOthers 0.352úú 0.369úú
(0.161) (0.165)
Implementation ◊ FARC 0.1269 0.0782
(0.1548) (0.1620)
Cease ◊ FARC ≠0.155 ≠0.170
(0.099) (0.105)
Implementation ◊ ExposureOthers 0.0101 ≠0.0198
(0.0577) (0.0649)
Cease ◊ ExposureOthers ≠0.182úúú ≠0.218úúú
(0.063) (0.070)
Implementation + Cease ◊ FARC ◊ ExposureOthers 0.340 0.391ú

(0.216) (0.221)
Observations 14966 14966

Municipalities 1069 1069
Municipality FE X X
Period FE X X
Controls X
Avg Dep Var 0.101 0.101
SD Dep Var 1.083 1.083
after 2015:1, Implementation is a dummy that takes the value one for the period after 2017:1. FARC
is a dummy for those municipalities with FARC presence. ExposureOthers is a continuous variable that
measures ELN or paramilitary groups presence in the municipality or their (distance-penalized) vicinity.
Predetermined municipal controls includes logarithm of the population in 2010, municipality area, average
elevation, distance to the closest major city, share of population under poverty, literacy rate, math and
language test scores, index of rurality, log of tax income and index of good fiscal policy. Errors in parenthe-
ses control for spatial and first-order time correlation (see Conley, 1999, Conley, 2016). We allow spatial
correlation to extend to up to 279 km from each municipality’s centroid to ensure that each municipality
has at least one neighbor. * is significant at the 10% level, ** is significant at the 5% level, *** is significant
at the 1% level.
Table 7. Heterogeneous effects by municipality characteristics
Land Judicial Military Coca

Restitution Inefficiency Presence Suitability
(1) (2) (3) (4)
Cease ◊ FARC ◊ ExposureOthers ◊ Z 0.669* 3.893* -0.123 0.030

(0.345) (2.096) (0.108) (0.060)
Cease ◊ FARC ◊ Z 0.009 0.084 0.033 -0.031
(0.230) (1.102) (0.068) (0.084)
Cease ◊ ExposureOthers ◊ Z -0.200 -1.510** 0.009 -0.010
(0.128) (0.684) (0.037) (0.083)
Cease ◊ FARC ◊ ExposureOthers 0.001 0.035 0.436*** 0.474***
(0.266) (0.241) (0.168) (0.162)
Cease ◊ FARC -0.154 -0.136 -0.138 -0.161
(0.163) (0.120) (0.100) (0.103)
Cease ◊ ExposureOthers -0.133 -0.120 -0.240*** -0.217***
(0.109) (0.073) (0.067) (0.071)
Cease ◊ Z 0.070** 0.184 -0.007 0.013
(0.034) (0.180) (0.019) (0.029)
Observations 14966 14966 14966 14154

Period FE X X X X
Controls X X X X
Avg Dep Var 0.101 0.101 0.101 0.101
SD Dep Var 1.083 1.083 1.083 1.083
after 2015:1, Implementation is a dummy that takes the value one for the period after 2017:1. FARC
is a dummy for those municipalities with FARC presence. ExposureOthers is a continuous variable that
measures ELN or paramilitary groups presence in the municipality or their (distance-penalized) vicinity.
Land restitution is a dummy for those municipalities with the number of request for land restitution
over the size of the municipality being above the median. Judicial inefficiency is the share of justice
employees under disciplinary investigations. Military presence is the logarithm of the distance to the
closest military unit. Coca suitability is a dummy that takes the value one if the index for coca suitability
from Mejı́a and Restrepo (2015) is above the median of the empirical distribution weighted by the exposure
to coca plantations in neighboring municipalities. Predetermined municipal controls includes logarithm of
the population in 2010, municipality area, average elevation, distance to the closest major city, share of
population under poverty, literacy rate, math and language test scores, index of rurality, log of tax income
and index of good fiscal policy. Errors in parentheses control for spatial and first-order time correlation
(see Conley, 1999, Conley, 2016). We allow spatial correlation to extend to up to 279 km from each
municipality’s centroid to ensure that each municipality has at least one neighbor. * is significant at the
10% level, ** is significant at the 5% level, *** is significant at the 1% level.
Table 8. Overall homicides rate, FARC presence, and exposure to other

armed groups
Homicide rate
(1) (2)
Cease ◊ FARC ◊ ExposureOthers 1.511 1.499

(3.222) (3.131)
Cease ◊ FARC ≠0.686 ≠0.011
(2.097) (2.126)
Cease ◊ ExposureOthers ≠2.339 úú
≠2.457úú
(1.066) (1.111)
Observations 14966 14966

Municipalities 1069 1069
Municipality FE X X
Period FE X X
Controls X
Avg Dep Var 12.595 12.595
SD Dep Var 28.347 28.347
level variation by semester, over the period 2011:1 to 2017:2. The dependent variable is the total number
of homicides excluding social leaders over total population. Cease is a dummy that takes the value one for
the period after 2015:1. FARC is a dummy for those municipalities with FARC presence. ExposureOthers
fiscal policy. Errors in parentheses control for spatial and first-order time correlation (see Conley, 1999,
Conley, 2016). We allow spatial correlation to extend to up to 279 km from each municipality’s centroid to
ensure that each municipality has at least one neighbor. * is significant at the 10% level, ** is significant
at the 5% level, *** is significant at the 1% level.
Online Appendix
Figure A.1. Evolution of killings by leader type
A. Community leaders B. Conflict related organizations
C. Afro-Indigenous D. Others
Notes: This figure presents the evolution of killings of social leaders from 2010 to 2017 by type of leader. We
split the sample by type of municipality into those with FARC and other groups exposure, and only FARC.
We define those municipalities with exposure as those that have any positive exposure. Panel A presents the
evolution for community leaders (30%), panel B for leaders of peace related organizations (20%), panel C for
afro and indigenous leaders (27%), while panel D present the rest (23%). In all the panels we show one-year
moving averages to smooth the data.
Table A.1. Killing of social leaders, FARC presence and exposure to

other armed groups: Using Negative Binomial and Conditional Poisson
models
Negative Binomial Conditional Poisson

(1) (2) (3) (4)
Cease ◊ FARC ◊ ExposureOthers 1.612*** 1.586*** 1.681*** 1.663***

(0.435) (0.465) (0.421) (0.451)
Cease ◊ FARC -0.633** -0.873*** -0.803*** -0.976***

(0.296) (0.327) (0.285) (0.320)
Cease ◊ ExposureOthers -1.165*** -1.343*** -1.232*** -1.459***

(0.289) (0.326) (0.282) (0.319)
Observations 2786 2786 2786 2786

Period FE X X X X
Controls X X
Avg Dep Var 0.008 0.008 0.008 0.008
SD Dep Var 0.105 0.105 0.105 0.105
level variation by semester, over the period 2011:1 to 2017:2. All the columns use as dependent variable the
total number of homicides of social leaders. Columns (1) and (2) estimate a negative binomial model while
columns (3) and (4) estimate a conditional poisson model. Cease is a dummy that takes the value one for
the period after 2015:1. FARC is a dummy for those municipalities with FARC presence. ExposureOthers
fiscal policy. Standard errors in parentheses. * is significant at the 10% level, ** is significant at the 5%
level, *** is significant at the 1% level.
Table A.2. Measuring exposure to other armed groups using the pres-
ence in neighboring municipalities
Homicides rate Homicides Any homicides

(1) (2) (3) (4) (5) (6)
Cease ◊ FARC ◊ NeighOthers 0.252úú 0.289úú 0.054úú 0.060úú 0.033 0.036ú

(0.114) (0.120) (0.027) (0.026) (0.022) (0.021)
Cease ◊ FARC ≠0.130 ≠0.172 ≠0.013 ≠0.021 ≠0.001 ≠0.007
(0.105) (0.110) (0.015) (0.016) (0.013) (0.014)
Cease ◊ NeighOthers ≠0.131úúú ≠0.171úúú ≠0.019úú ≠0.028úúú ≠0.010ú ≠0.018úúú
(0.046) (0.053) (0.008) (0.008) (0.006) (0.007)
Observations 14966 14966 14966 14966 14966 14966

Municipalities 1069 1069 1069 1069 1069 1069
Controls X X X
Avg Dep Var 0.101 0.101 0.028 0.028 0.021 0.021
SD Dep Var 1.083 1.083 0.219 0.219 0.144 0.144
level variation by semester, over the period 2011:1 to 2017:2. Columns (1) and (2) use the number of
homicides of social leaders over total population, columns (3) and (4) use as dependent variable the total
number of homicides of social leaders, while columns (5) and (6) use a dummy that takes the value one
if there was at least one social leader assassinated. Cease is a dummy that takes the value one for the
period after 2015:1. FARC is a dummy for those municipalities with FARC presence. NeighOthers is the
share of ELN or paramilitary groups presence among the municipality and their neighbors. Predetermined
municipal controls includes logarithm of the population in 2010, municipality area, average elevation,
distance to the closest major city, share of population under poverty, literacy rate, math and language test
scores, index of rurality, log of tax income and index of good fiscal policy. Errors in parentheses control
for spatial and first-order time correlation (see Conley, 1999, Conley, 2016). We allow spatial correlation
to extend to up to 279 km from each municipality’s centroid to ensure that each municipality has at least
one neighbor. * is significant at the 10% level, ** is significant at the 5% level, *** is significant at the 1%
level.
Table A.3. Measuring exposure to other armed groups using all mu-
nicipalities penalized by distance
Homicides rate Homicides Any homicides

(1) (2) (3) (4) (5) (6)
Cease ◊ FARC ◊ ExposureOthers 0.300úú 0.330úú 0.073úú 0.078úú 0.044ú 0.047ú

(0.138) (0.143) (0.031) (0.031) (0.025) (0.025)
Cease ◊ FARC ≠0.188 ≠0.227ú ≠0.028 ≠0.037úú ≠0.009 ≠0.017
(0.121) (0.126) (0.017) (0.018) (0.015) (0.015)
Cease ◊ExposureOthers ≠0.147úúú ≠0.209úúú ≠0.025úú ≠0.038úúú ≠0.014ú ≠0.025úúú
(0.051) (0.061) (0.010) (0.011) (0.007) (0.008)
Observations 14966 14966 14966 14966 14966 14966

Municipalities 1069 1069 1069 1069 1069 1069
Controls X X X
Avg Dep Var 0.101 0.101 0.028 0.028 0.021 0.021
SD Dep Var 1.083 1.083 0.219 0.219 0.144 0.144
Notes: This table presents the results from the main specification in equation (4.1). We exploit
municipal-level variation by semester, over the period 2011:1 to 2017:2. Columns (1) and (2) use the
number of homicides of social leaders over total population, columns (3) and (4) use as dependent
variable the total number of homicides of social leaders, while columns (5) and (6) use a dummy that
takes the value one if there was at least one social leader assassinated. Cease is a dummy that takes
the value one for the period after 2015:1. FARC is a dummy for those municipalities with FARC
presence. ExposureOthers is a continuous variable that measures ELN or paramilitary groups presence
in the municipality and (distance-penalized) Colombian whole municipalities. Predetermined municipal
controls includes logarithm of the population in 2010, municipality area, average elevation, distance to
the closest major city, share of population under poverty, literacy rate, math and language test scores,
index of rurality, log of tax income and index of good fiscal policy. Errors in parentheses control for
spatial and first-order time correlation (see Conley, 1999, Conley, 2016). We allow spatial correlation
to extend to up to 279 km from each municipality’s centroid to ensure that each municipality has at
least one neighbor. * is significant at the 10% level, ** is significant at the 5% level, *** is significant
at the 1% level.
Table A.4. Killing of social leaders by type of leader
Community Conflict related

councils organizations Afro-Indigenous Others
(1) (2) (3) (4) (5) (6) (7) (8)
Cease ◊ FARC ◊ ExposureOthers 0.170** 0.177** 0.067 0.070 0.086 0.099 0.024 0.030
(0.081) (0.086) (0.093) (0.091) (0.079) (0.085) (0.029) (0.029)
Cease ◊ FARC -0.010 -0.021 0.006 0.005 -0.113* -0.122* 0.004 -0.007
(0.057) (0.057) (0.054) (0.058) (0.065) (0.066) (0.017) (0.018)
Cease ◊ ExposureOthers -0.063 -0.087 -0.011 -0.018 -0.049 -0.050 -0.055*** -0.070***
(0.046) (0.055) (0.018) (0.018) (0.032) (0.033) (0.018) (0.021)
Observations 14966 14966 14966 14966 14966 14966 14966 14966

Municipalities 1069 1069 1069 1069 1069 1069 1069 1069
Municipality FE X X X X X X X X
Period FE X X X X X X X X
Controls X X X X
Avg Dep Var 0.003 0.003 0.006 0.006 0.011 0.011 0.003 0.003
SD Dep Var 0.117 0.117 0.179 0.179 0.341 0.341 0.123 0.123
Notes: This table presents the results from the main specification in equation (4.1) by splitting the killings into types of leaders. We
exploit municipal-level variation by semester, over the period 2011:1 to 2017:2. All columns use as dependent variable the number
of killings over population. Cease is a dummy that takes the value one for the period after 2015:1. FARC is a dummy for those
municipalities with FARC presence. ExposureOthers is a continuous variable that measures ELN or paramilitary groups presence in
the municipality or their (distance-penalized) vicinity. Predetermined municipal controls includes logarithm of the population in 2010,
KILLING SOCIAL LEADERS FOR TERRITORIAL CONTROL
municipality area, average elevation, distance to the closest major city, share of population under poverty, literacy rate, math and
language test scores, index of rurality, log of tax income and index of good fiscal policy. Errors in parentheses control for spatial
and first-order time correlation (see Conley, 1999, Conley, 2016). We allow spatial correlation to extend to up to 279 km from each
municipality’s centroid to ensure that each municipality has at least one neighbor. * is significant at the 10% level, ** is significant at
the 5% level, *** is significant at the 1% level.
43

2018 Prem Et Al - Killing of Social Leaders For Territorial Control - The Unintended Consequences of Peace

Uploaded by

Copyright:

Available Formats

2018 Prem Et Al - Killing of Social Leaders For Territorial Control - The Unintended Consequences of Peace

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2018 Prem Et Al - Killing of Social Leaders For Territorial Control - The Unintended Consequences of Peace

Uploaded by

Copyright:

Available Formats

LACEA WORKING PAPER SERES. No.

KILLING SOCIAL LEADERS FOR TERRITORIAL CONTROL: THE UNINTENDED

LATIN AMERICAN AND THE CARIBBEAN ECONOMIC ASSOCIATION

ACKNOWLEDGEMENTS AND FINANCIAL DISCLOSURE

of selective civilian killings.8 Indiscriminate violence is usually ineffective in achieving

the success of violence reduction security programs12 by exploring what mechanisms

period, 2011:1–2014:2), and a vector of (distance-penalized) neighboring municipalities.

4.2. Testing potential mechanisms. We can use municipal-level variation across

Our coefficient of interest, —1 , captures the differential killing of social leaders in

5.3. Potential mechanisms.

5.3.3. Indiscriminate violence as an alternative explanation. As a final attempt to iden-

to achieve informal collaboration, prevent defections, mobilize supporters, and increase

Figure 1. Evolution of social leaders killings

Agreement Agreement Agreement

B. Split by type of municipality

Figure 2. Spatial distribution of social leaders killings

Figure 3. Dynamic estimation and parallel trends

Table 1. Leaders killed by activity

Local council 165 33.3

Table 2. Descriptive Statistics: Time-invariant variables

Mean Std. Dev. Min Max

Social leaders killings

Dummy of any killing 0.021 0.144 0.0 1.0

Illegal groups presence

FARC 0.093 0.290 0.0 1.0

Altitude (Km) 1.149 0.903 0.0 3.1

Log (population) 9.489 0.948 6.9 12.2

Log (Tax income) 6.726 1.408 0.0 12.1

Table 3. Descriptive Statistics by illegal groups presence

Presence Presence Only FARC vs.

Social leaders killings

Dummy of any killing 0.058 0.052 0.038

Altitude (Km) -0.276 -0.542 -0.134

Log (population) 0.552 0.760 0.322

Log (Tax income) 0.363 0.557 0.250

Table 4. Killing of social leaders, FARC presence, and exposure to

Killing rate Number of killings Any killing

Cease ◊ FARC ◊ ExposureOthers 0.348úú 0.377úú 0.090úú 0.095úúú 0.052ú 0.055ú

Observations 14966 14966 14966 14966 14966 14966

Table 5. Killing of social leaders by exposure to different armed groups

Cease ◊ FARC ◊ ExposureOthers 0.330úú 0.353úú 0.483úú 0.515úú

Observations 14966 14966 14966 14966

Implementation ◊ FARC ◊ ExposureOthers ≠0.0121 0.0212

Implementation + Cease ◊ FARC ◊ ExposureOthers 0.340 0.391ú

Observations 14966 14966

Table 7. Heterogeneous effects by municipality characteristics

Land Judicial Military Coca

Cease ◊ FARC ◊ ExposureOthers ◊ Z 0.669* 3.893* -0.123 0.030

Observations 14966 14966 14966 14154

Table 8. Overall homicides rate, FARC presence, and exposure to other

Cease ◊ FARC ◊ ExposureOthers 1.511 1.499

Observations 14966 14966

Figure A.1. Evolution of killings by leader type

A. Community leaders B. Conflict related organizations

Table A.1. Killing of social leaders, FARC presence and exposure to

Negative Binomial Conditional Poisson

Cease ◊ FARC ◊ ExposureOthers 1.612*** 1.586*** 1.681*** 1.663***

Cease ◊ FARC -0.633** -0.873*** -0.803*** -0.976***

Cease ◊ ExposureOthers -1.165*** -1.343*** -1.232*** -1.459***

Cease ◊ FARC ◊ ExposureOthers 1.612* 1.586* 1.681* 1.663*

Cease ◊ FARC -0.633 -0.873* -0.803* -0.976*

Cease ◊ ExposureOthers -1.165* -1.343* -1.232* -1.459*