2023-09-cid-fellows-wp-155-violence-covid-outbreaks

Download as pdf or txt
Download as pdf or txt
You are on page 1of 50

Pretending to be the Law: Violence to Reduce the

COVID-19 Outbreak
Citation
Martin, Diego A., and Dario A. Romero. "Pretending to be the Law: Violence to Reduce the
COVID-19 Outbreak." CID Research Fellows and Graduate Student Working Paper Series
2023.155, Harvard University, Cambridge, MA, September 2023.

Published Version
https://growthlab.hks.harvard.edu/publications/pretending-be-law-violence-reduce-covid-19-
outbreak

Permanent link
https://nrs.harvard.edu/URN-3:HUL.INSTREPOS:37377356

Terms of Use
This article was downloaded from Harvard University’s DASH repository, WARNING: No
applicable access license found.

Share Your Story


The Harvard community has made this article openly available.
Please share how this access benefits you. Submit a story .

Accessibility
Pretending to be the Law:
Violence to Reduce the
COVID-19 Outbreak

Diego A. Martin and Dario A. Romero

The Growth Lab at Harvard University

CID Research Fellow and Graduate Student


Working Paper No. 155
September 2023

© Copyright 2023 Martin, Diego; Romero, Dario;


and the President and Fellows of Harvard College
PRETENDING TO BE THE LAW:
VIOLENCE TO REDUCE THE COVID-19 OUTBREAK

DIEGO A. MARTIN‡ AND DARIO A. ROMERO†

This version: September 2023

Abstract. Did the COVID-19 pandemic create an opportunity to earn population


control through illegal violence? We argue that criminal groups in Colombia portray as
de facto police by using mass killings to reduce the COVID-19 outbreak. They used
massacres as a threat to enforce social distance measures in places they consid-ered
worth decreasing mobility. Our results from an Augmented Synthetic Control Method
model estimated that commuting to parks fell 20% more in areas with mas-sacres than
in places without mass killings. In addition, we do not fnd a decline in mobility to
workplaces and COVID-19 deaths after the frst mass killing. These fndings are
congruent with the hypothesis that illegal armed groups used fear to enforce mobility
restrictions without hurting economic activities and their sources of revenue. However,
violence slightly impacted the virus’ spread. Treated areas had a decline of 35 cases per
100.000 inhabitants four months after the frst massacre.

JEL Codes: H75, D74, K42


Keywords: COVID-19, Social Distance, Lockdowns, Massacres, Governance


We thank Rodrigo Soares, Suresh Naidu, Reka Juhasz, Eric Verhoogen and seminar participants at
Purdue University seminar, the Annual LACEA Meeting 2021, and the Empirical Studies of Confict
(ESOC) at Princeton University for helpful comments and suggestions. All errors are our own.

Harvard Kennedy School, Havard University. Cambridge, MA 02138. E-mail:
dmartin@hks.harvard.edu.

Social Science Division, New York University - Abu Dhabi. Bldg A5-142 P.O. Box 129188, Abu
Dhabi, UAE. E-mail: drf312@nyu.edu.
PRETENDING TO BE THE LAW 1

1. Introduction

The COVID-19 pandemic was an immense shock that changed power balances be-
tween criminal bands, legal military forces, government, and civilians (Ali, Hassan, and
Hossain, 2021). These changes were particularly relevant in disputed environments
where legal and illegal groups battle for territory and population control (Abrams,
2021). We argue that criminal groups acted as de facto police by using violent strate-
gies when the national government dropped all lockdown measures. We focused on
non-state Colombian armies enforcing illegitimate stay-at-home orders by using mas-
sacres that is simultaneously killing three or more civilians not directly involved in the
confict.
Our argument is not that COVID-19 was the primary explanation for civilian vic-
timization in Colombia after 2020. The literature has shown that revenues from illicit
economics, like coca cultivation, drive mass killings in Colombia (Prem et al. (2022)
and Llanes (2022)).1 Instead, we argue that illegal armies use violence as a sign of
control and power, and this is especially true in areas where criminal payments do not
rely heavily on the use of local population in illicit activities. In the case of Colombia,
in areas where the population does not participate in coca growth, therefore we focused
our analysis on places where growing coca is more challenging due to low suitability
for coca cultivation. First, this strategy allows us to reduce concerns about alternative
channels that explain the use of violence during the pandemic since we concentrate the
analysis of the phenomena in places where illegal activities do not drive the motivation
of massacres. Second, it allows us to show how the pandemic had di erent e ects on
the balance of power depending on the relations structures between civilians and armed
groups.

1
Massacres have multiple motivations, such as electoral incentives, illicit dispute revenues, and popu-
lation alignment (Alesina, Piccolo, and Pinotti, 2018; Humpreys and Weinstein, 2006; Robinson and
Torvik, 2009)
PRETENDING TO BE THE LAW 2

We used an augmented synthetic control model (ASCM) to estimate the causal e ect
of the frst massacre on mobility and COVID-19 cases. Using the history of massacres
after the national lockdowns on March 24th , 2020. we built a weighted average of
non a ected units that match the pretreatment outcomes of areas with massacres
(Ben-Michael, Feller, and Rothstein, 2021a). The method creates contrafactual control
regions to model the behavior of mobility and COVID-19 behaviour as if it were not
a massacre. Before showing the results of the frst massacre on human movement and
the COVID-19 case, we presented evidence supporting the plausibility of conditional
treatment assumptions to estimate an unbiased parameter with the ASMC model.
We showed that neither previous COVID-19 cases nor past human mobility trends
predict mass killings. These results show that criminal bands did not target civilians
depending on pretreatment outcomes, physical attributes of the areas, or population
characteristics.
Our results showed that human mobility declined 20% more in areas with mass
killings than in synthetic control units after the frst massacre. The di erence in mobil-
ity was statistically signifcant for trips to parks but not for commuting to workplaces.
The reduction in mobility to low-risk transmission areas slowly translated into a de-
cline in the COVID-19 cases. Four months after a mass killing, treated units dropped
35 cases per 100.000 inhabitants compared with synthetic control units. However, the
decrease was only signifcant for the population between 15-44 years old. We did not
fnd a statistically signifcant e ect of mass killings on COVID-19-related deaths.
As robustness, we estimated a placebo test to show that massacres in 2022 did not
a ect population movement. We did not fnd e ects of massacres two years after the
frst COVID-19 case when non-lockdowns were in place in Colombia. This result sup-
ports our hypothesis about illegal groups using violence only during the pandemic to
demonstrate control over the population. Another possible story was that violence
a ected the detection of COVID-19 cases. However, we showed that the likelihood
PRETENDING TO BE THE LAW 3

of catching a new positive case was the same for all Colombian sub-regions indepen-
dently of the occurrence of massacres. While COVID-19 testing did not correlate with
violence, geographic characteristics explained the testing variation among provinces.
Finally, we reject the possibility that data disponibility explains our results. We arrive
at the same conclusions when using several samples of a ected areas.
This article contributes to the broad literature studying how diseases a ect the
decision of war-making. For example, infections can produce risker individuals or
increase the willingness to rebel by reducing life expectancy (Cervellati, Sunde, and
Valmori (2017) and Kustra (2017), respectively). Diseases can also become a natural
barrier for criminal groups who hide in contaminated zones, as in the case of Malaria
(Bagozzi, 2016). Our paper adds a new mechanism not previously analyzed: pretend
to be the law. We show that with the increment of concerns about public wellness,
illegal Colombian groups saw an opportunity to portray as a valid alternative force
that controls population and territory.
Our work also contributes to explaining non-state and state violence during the
COVID-19 pandemic. Since governments spend time and fnancial aid facing the pan-
demic, criminal groups could either react violently to reduce the COVID-19 outbreak
or avoid confrontation to save economic and human resources (Koehnlein and Koren,
2022). Authoritarian states also repressed domestic dissidents without the pressure of
the international community or citizen scrutiny (Barceló et al., 2022). African states,
for example, engaged in repressive campaigns after imposing lockdowns (Grasse et al.,
2021). In the Colombian case, we found that illegal groups acted as state actors to es-
tablish themselves as viable alternative government who took advantage of low scrutiny
to use violence for controlling the population.
Finally, we contribute to the growing literature on the explanation of use of violence
against civilians.2 While the role of civilians is essential to all sides in civil wars
2Fearon and Laitin (2003) and Blattman (2022) present the reasons behind violence and irregular
wars.
PRETENDING TO BE THE LAW 4

(Berman, Shapiro, and Felter, 2011; Clutterbuck, 1966; Galula, 1964; Thompson, 1966),
illegal groups still use violence against the population as an optimal strategy (Kalyvas,
2006). Little force could lead to loose authority (Balcells, 2017; Kocher, Pepinsky,
and Kalyvas, 2011; Lyall, 2009; Stoll, 1993; Vargas, 2016), but too much violence could
push civilians to leave the territory or join the enemy for revenge (Schwartz and Straus,
2018).3 Illicit armed groups use violence when people are not a principal component of
revenue production (la Calle, 2017). At the same time, weak insurgents have incentives
to target the local population with violence since they cannot ensure credible support
with other means (Wood, 2014). During the COVID-19 pandemic, we showed a high
level of violence in areas with low labor demand from illegal groups and several armed
groups fghting for territorial control. We also presented evidence that criminal bands
sought not to hurt economic activities but to control leisure activities.
The remainder of the article is as follows: Section 2 describes Colombia’s COVID-
19 situation and civil war background. Section 3 describes the di erent data sources
and descriptive statistics of the COVID-19 evolution and mass killings in Colombia.
Section 4 shows the synthetic control method and assumption to create a valid control
group. Section 5 shows the results for human movment and COVID-19 cases. Section
6 presents robustness tests. Finally, section 7 highlights the implications of the fndings
for policymakers.

2. Background

2.1. The Civil War in Colombia. The Colombian confict is a multi-party confict
that has lasted more than 60 years. Besides state forces, there are several left-wing
guerrillas groups in dispute, such as the Revolutionary Armed Forces of Colombia
(FARC) and the National Liberation Army (ELN). These organizations have a strong

3Civilianscan implement strategies to punish the previous criminal groups for extreme violence (Con-
dra and Shapiro, 2012; Shaver and Shapiro, 2021). However, the back fghting is not symmetric, and
some cases depend on the alignment or preferences of local inhabitants (Huber, 2019; Lyall, Blair, and
Imai, 2013; Toft and Zhukov, 2015).
PRETENDING TO BE THE LAW 5

presence in remote rural areas and have fnanced their activities using illegal activi-
ties and taxes on legal activities (Arango, 2020). Moreover, there are also right-wing
paramilitary criminal organizations in dispute. These illegal groups were born as a
counterinsurgency strategy under the approval of the sector of military forces (Arjona,
2016).
The Colombian civil world reached its highest point during the 1990s when most
groups were involved in drug cultivation and traÿcking (Oslender, 2007). During
this period several criminal organizations used massacres of civilians as a strategy to
consolidate their power. Specially paramilitary groups that unifed under the United
Self-Defense Forces of Colombia (AUC) used mass killings as a strategy to contain
the local support for left-wing guerilla (Aranguren, 2001). By 2002, Colombia faced
more than one hundred massacres per quarter (Restrepo, Spagat, and Vargas, 2004).
However, massacres decreased after the AUC’s peace process fnished in 2006. Al-
though Colombia still has several paramilitary groups that splintered from the leading
organization, the accord was successful to reduced massacres (Holmes et al., 2021).4
Recently there have been e orts to reduce violence. After four years of negotia-
tion between the Colombian government and the FARC, both sides signed a Peace
Agreement in 2016 (Parada, 2022). However, the government faced the challenge of es-
tablishing control in territories historically handled by FARC structures. Other illegal
armies established dominance in those previously controlled-FARC areas after fghts
with other bands and using violence against civilians (Arango, 2020).

2.2. COVID-19 and lockdowns. Like many countries around the world, the spread
of the COVID-19 virus heavily a ected Colombia. Beginning with the frst confrmed
case on March 6th , 2020, the virus spread rapidly to most Colombian areas. By Sep-
tember 2022, Colombia had more than six million confrmed COVID-19 cases with a
fatality rate of 2.25% (INS, 2022). However Colombia had fewer deaths per 100,000
4Figuire A.I in the Online Appendix shows the historical evolution of massacres in Colombia.
PRETENDING TO BE THE LAW 6

people than other countries in the region such as Peru, Brazil, Chile, and Argentina
(Sullivan, 2020).5
The Colombian government implemented several strategies to mitigate the pandemic
outbreak. On March 24th , 2020, the government announced a nationwide lockdown for
19 days. However, given the increase in transmission levels, the government progressible
extended the lockdown with adding exceptions to alleviate the economic cost associated
with the curfews. On May 4th , 2020, the government began an opening plan with
manufacturing and construction sectors. These lockdowns ended on September 1st ,
2020, when the government lifted all mobility restrictions (Arregocés, Rojano, and
Restrepo, 2021).
We argue that massacres became a strategic tool to handle the COVID-19 pandemic
under the presence of illicit actors competing for territory control and population sup-
port. According to qualitative interviews with social leaders, police oÿcers, and civil
society, armed groups imposed rules to contain the COVID-19 spread in Colombian
rural areas (HRW, 2020).6 In several regions, illegal groups saw the pandemic as an
opportunity to size control and change the power balance in dispute regions (Colombia
Investigative Unit, 2020).7

3. Data

3.1. Province defnition. The presence of spillovers is a challenge when analyzing the
impacts of policies to contain infectious diseases like COVID-19. An administrative
unit such as a county or a city does not retain the virus, and measures taken in a
5In the Online Appendix, Figure A.II shows that Colombia had four COVID-19 deaths peaks in
August 2020, February 2021, June 2021, and March 2022.
6The following groups set rules to reduce the COVID-19 outbreak: the National Liberation Army
(ELN), the Popular Liberation Army (EPL), the Gaitanist Self-Defenses of Colombia (AGC), Conta-
dores in Nariño; La Mafa in Putumayo; Oliver Sinisterra Front, the United Guerrillas of the Pacifc,
the Jaime Martı́nez mobile column, the Dagoberto Ramos mobile column, the 1st Front, 7th Front,
10th Front, and the Carolina Ramı́rez Front (FARC dissident groups) among others (HRW, 2020).
7
In the Online Appedix, Subsection B shows a list of qualitative examples where illegal armies used
massacres to reduce human movement and control the COVID-19 outbreak.
PRETENDING TO BE THE LAW 7

municipality a ect virus transmission in neighboring localities. We used as analysis


unit the province to overcome the spillover e ects. We grouped 1123 municipalities
into 154 sub-regions using the defnition proposed by Ramı́rez and De Aguas (2022),
who defned a sub-region as a set of places with similar environmental characteristics
and proximity to the closest urban center.8 We believe aggregating municipalities at
the providence level takes into account spillover.

3.2. Number of massacres. We used the Armed Confict Location and Event Data
Project (ACLED, Raleigh et al. (2010)). This project collects information about polit-
ical violence events in the world. For the Colombian case, the data has been available
since January 2018 and came from a wide range of national and local media with the
help of local NGOs and community networks.9 We focused our analysis on massacres,
defned as killing three or more unarmed civilians in one attack. We also excluded
events whose alleged perpetrator was a state force, such as the military or the po-
lice.10 Between January 2018 and March 2022, ony 62 municipalities in 36 sub-regions
had 94 massacres with 375 civilian causalities (5% of total municipalities in Colombia).
Around 50% of these sub-regions observed only one mass killing, 28% had two or three,
and 22% four or more massacres.
Figure 1 shows the quarterly evolution of massacres since 2018. Before the COVID-
19 outbreak in March 2020, there were ten massacres per semester, on average. After
the beginning of the pandemic, the mass killings increased to around 20 massacres per
semester. In the third quarter of 2020, when the government completely lifted the
8
Ramı́rez and De Aguas (2022) do not divide departments such as Arauca, Caquetá, Casanare,
Guainiı́a, Guaviare, Putumayo, Vaupeés and Vichada. A sub-region is neither a political nor an
administrative division in Colombia.
9
One crucial source of information about massacres is an NGO named the Institute of Studies for
Development and Peace (Indepaz) that monitors the confict in Colombia. The data comes from
di erent sources such as news, the police, the army, the Ministry of Defense and the Oÿce of the
Attorney General of the Nation, local agencies protecting the citizens’ rights, and other human rights
platforms (Indepaz, 2021).
10
See ACLED (2019) for more details about the defnitions of events and actors used during the
codifcations of the events.
PRETENDING TO BE THE LAW 8

curfews, massacres reached their maximum of 30 events, with 128 victims in a single
quarter. We also found that this increasing trend comes from non-coca growing areas.
In the Online appendix, Figure A.III shows that massacres in municipalities with high
coca suitability remained relatively stable, with around seven massacres per quarter
between 2018 and 2022. Mass killings in places with low coca suitability experienced
a boom after the second quarter of 2020, from 10 to 20 massacres.11
Figure 2 shows the time distributions of mass killings, victims, and provinces with
a massacre in our sample of low coca suitability regions. By March 31st, 2021, around
75% of mass killings happened in regions with coca suitability index below the na-
tional median. Mass killings, victims, and provinces with massacres increased after
the Colombian government eliminated all the lockdowns. About 30% of the massacres
from 2019 to 2022 occurred ten days after the total release of curfews. Overall, the
raw data aligns with the hypothesis that the increase in massacres was not associated
with the dispute over illegal activities. Mass killings targeted civilians in public spaces
rather than people linked to illicit economies.

3.3. Google community mobility. We hypothesized that massacres aimed to af-


fected COVID-19 evolution through changes in mobility patterns that the violence
produced on local communities. We used Google Community Mobility reports (GCM)
to address this hypothesis (LLC Google, 2021). Based on users that turned on the
location history settings, Google could measure the number of visits to di erent types
of locations and compare it to movement trends before the COVID-19 outbreak.12 We

11
We did not see an increase in killings in municipalities receiving government funding to reduce coca
cultivation (see Figure A.IV in the Online Appendix).
12
We do not have the raw number of visits to a place in a specifc week. We observe, for example,
-34% in week seven of 2020, which means a decline of 34% in trips to a particular place, comparing
the median number of trips between the frst six of the year and the median trips in week seven. The
baseline date is the median number of visits to a specifc place from January 3rd to February 6th in
2020 (LLC Google, 2021).
PRETENDING TO BE THE LAW 9

focused on movements to parks and workplaces that represent the primary behaviour
of mobility patterns in small cities (Duranton, 2016).13
Google data is available for only 321 municipalities and certain days per week. To
overcome these issues of data availability, we averaged the available days per week and
built weights from the municipality population to aggregate the data at the providence
level. When there is no information for a particular week, we interpolate or extrapo-
late the data to complete the missing observations. The fnal database recovered the
mobility index at workplaces for 119 providences and parks for 122 sub-regions or what
it is the same for 79% of Colombia’s total number of provinces.14
After the frst lockdown on March 24th , 2020, human mobility declined by about 60%
compared with the median trips in the frst six weeks of 2022. Although trips outside
the home gradually recovered over time, movements did not reach the levels before the
pandemic (see Figure 3). Before the total lift of lockdowns, human mobility decreased
more in areas with massacres than without mass killings. However, one month after
the complete release of curfews, the movement trend changed in small sub-regions with
mass killings. Trips to parks, for example, are below in areas with massacres than
without violence. The movement trends did not change in big provinces, regardless of
the level of these human rights violations. These pieces of raw data is consistent with
our hypothesis that illegal groups enforced mobility restrictions through massacres.

3.4. COVID-19 cases. Colombian Institute of Health (INS) centralizes the informa-
tion about the universe of COVID-19 events at the national level. The INS collects
information regarding the patient’s symptoms, location, and test result dates (INS,

13
The GCM also includes mobility to grocery stores, pharmacies, parks, transport stations, workplaces,
entertainment places, and residential areas (LLC Google, 2021).
14
For reference, we recovered mobility in types of places such as supermarkets or recreation places for
less than half of the provinces. Our fnal database recovered the mobility index for retail and residence
in 78 and 68 sub-regions, respectively.
PRETENDING TO BE THE LAW 10

2022). We used only the evolution of symptomatic cases to study the pandemic’s evo-
lution.15 We defned a new case using the self-reported date of the frst symptoms
instead of the date of diagnosis of the test. This date captures more precisely the
growth and circulation of the virus. Even more, considering that COVID-19 tests are
not widely available in all the regions, the timing for getting the results depends on
the region’s connectivity (INS, 2022).
During the frst months of the pandemic, COVID-19 cases followed a similar pattern
in small sub-regions with and without massacres. However, after the government com-
pletely lifted the lockdowns, small provinces with massacres reported a lower infection
rate than places without mass killings (see Figure 4 panel A). Big sub-regions with and
without massacres have a similar trend in COVID-19 events before and after the lift
of the curfews (see Figure 4 panel B).16
Figure 5 shows Colombia’s spatial distribution of massacres and COVID-19 cases.
First, we observe that most massacres occurred in regions with low coca suitability.
Second, when we looked at the areas with massacres, we observed that those regions
generally report lower cumulative COVID-19 cases, especially when compared with
their closest neighbors. These facts support our hypothesis that illegal armies use
massacres to reduce the virus spread and not directly control illicit markets.

3.5. Other data. We use a large set of predetermined municipal characteristics such
as the degree of rurality, population, area, altitude over the sea, distance to the de-
partmental capital, population density, total municipality income, suitability for coca
production, gold exploration, electoral risk, and justice ineÿciency index. We aggre-
gate these characteristics at the province level. For altitude, distance to the capital,
and coca suitability, we aggregate the measure weighting by the total population or the

15Only Bogota, the Colombian capital, constantly reported asymptomatic patients. Asymptomatic
cases in Bogota represent more than 90% of COVID-19 patients with non-symptoms in Colombia.
16Figure A.V in the Online Appendix shows that small provinces with massacres have fewer deaths
from the pandemic than areas without mass killings.
PRETENDING TO BE THE LAW 11

total size of each of the municipalities that belong to the province. The source of this
data is mainly Centro de Estudios sobre Desarrollo Económico (CEDE) at Universidad
de los Andes and the Colombian Census Bureau (DANE). In the Online Appendix, ta-
ble A.I shows in detail the defnition of each variable and the source and table A.II
presents the summary the descriptive statistics in our sample.

3.6. Final sample. We study the number of massacres from March 24th , 2020, when
the Colombian government started lockdowns to control the pandemic, to March 31st ,
2021, one month before generalized protests and riots (Uwishema et al., 2022). We
were concerned that including events after would have captured a di erent range of
motivations from illegal groups.17
We used the coca suitability index proposed by Mejı́a and Restrepo (2015) to defne
597 (53%) out of municipalities as highly suitable when 95% of the area supports
coca growth.18 The authors estimated the coca leaf yields depending on geographic
characteristics such as the height above the sea, soil erosion, soil nutrients, minerals,
topography, and rainfall index. We averaged the index at the province level, weighting
it by the municipality area, and calculated the distribution for each province.
Finally, we excluded Bogotá from our fnal sample since this city had a di erent
tracing COVID-19 strategy than other parts of the country. The city conducted an
active search in local communities of cases among positive patients and their contacts.
Even more, the city is mainly urban, and its mobility patterns di er from all the other
cities and towns of Colombia. We think that this exclusion allows us to perform a
better comparison within similar groups.

17Our COVID-19 measures include data until September 30th , 2021. That is around 30 weeks after
the events we were analyzing. The maximum period that we think the behavior change would a ect
the evolution of the infection.
18
A municipality where 95% of the area supports coca production corresponds to the 75th percentile
in the distribution of the province area suitable for coca cultivation.
PRETENDING TO BE THE LAW 12

4. Empirical strategy

Our objective is to evaluate the e ect of a massacre on community mobility and


COVID-19 transmission rates in sub-regions with a historically low level of violence.
Specifcally, we were interested in comparing our outcomes (YpT (1)) on each period after
a massacre T with a contrafactual cases as if there had been no massacres (ŶpT (0)). We
used an augmented synthetic control method (ASCM) to estimate a version of province
p treated (p 2 Wp = 1) that performed statistically equal before the frst massacres
(T0 < T ).
The seminal method synthetic control method (SCM) uses a weighted combination
of untreated units (Wp = 0) to build a synthetic unit, such the behavior of the out-
come resembles the original treated-unit before the treatment (Abadie, Diamond, and
Hainmueller, 2010; Abadie and Gardeazabal, 2003). This method constructs weights
(!pscm 2 [0, 1]) to minimize the di erence in pre-intervention trends between the treated
and the synthetic control. Once the weights are estimated, they are used to approxi-
mate the potential outcome Yˆp,T
syn
(0) of the treated unit in the post-intervention period.
Formally, the estimated synthetic outcome at time T is:

YˆpT
syn
!pscm YpT
X
(4.1) (0) =
Wp =0

However, this method does not guarantee a perfect balance in all the characteristics.19
To overcome this issue, we corrected the bias on estimations when the pre-treatment
ft was not perfect, following Ben-Michael, Feller, and Rothstein (2021a). Formally we
estimated the synthetic level of infection at the treated unit using the following model:
0 1
aug
!pascm YpT + @m̂pT (Xp ) − !pascm m̂pT (XP )A
X X
(4.2) ŶpT (0) =
Wp =0 Wp =0

19Indeed Appendix Table A.III shows some di erences between provinces without and with massacres.
A particular concern is that massacres occurred in a place with more share of gold exploitation, more
presence of coca substitution programs, provinces at lower altitudes and with higher density, and
further away from important cities.
PRETENDING TO BE THE LAW 13

Where m̂ is the outcome model that can be seen as an estimate of the bias due
to imbalance. The model we choose to de-bias the original SCM estimate is a ridge-
regularized linear regression that increases the pre-treatment ft using the variables set
Xp . This set included a series of pre-treatment outcomes and a set of fxed province
characteristics.20 The method’s cost is to employ negative weights to improve the pre-
treatment fts when negative weights are generally more sparse and less interpretable
(see Ben-Michael, Feller, and Rothstein, 2021a, sec 4.1).

4.1. Build the average treatment e ect on the treated (ATT). Our goal is to
identify the average e ect of mass killings on COVID-19 cases. Since we have multiple
treated units (i.e., sub-regions with massacres), we needed to aggregate the estimated
e ects for each unit to calcualte the ATT. This aggregation is particularly challenging
since we have several treated provinces in di erent weeks. Thus, the weight estimation
that minimizes the imbalance before the treatment must consider two forces. The frst
is the imbalance for each treated unit separately, and the second is the imbalance for
the whole average of the treated units.
We followed Ben-Michael, Feller, and Rothstein (2021b) and calculated ! ascm by
minimizing the two sources of imbalance in the average e ect instead of calculating the
mean individual e ects for each treated unit. The method is a partially pooled SCM
that weights the combination of these two measures. We allow the algorithm to choose
a combination of the two factors based on how well separate synthetic controls balance
the overall average. Formally, the parameter that governs the relative weight is .
 = 0 is equivalent to estimating separate SCM fts for each province, then estimating
the ATT by averaging those estimates.  = 1 is equivalent to fnding the weights that
minimize the ATT’s root mean squared placebo estimate.

20The characteristics are total population, area, the share of the rural population, women, coca
suitability, municipalities with governmental fnancial support to reduce the cultivation of illegal
crops, gold exploration area, population density, average altitude, total income and expenditures per
capita and distance to the capital.
PRETENDING TO BE THE LAW 14

4.2. Assumptions. This method correctly estimates unbiased ATT in the presence of
three assumptions. First, the treatment is stable across units, or what is the same, that
massacre only a ected the treated areas. An analysis at the municipality level violates
this assumption. Colombian illegal armies do not operate in isolated municipalities,
and massacres send signals to di erent municipalities beyond the location where they
happened. By aggregating the data at the province level, we overcome this issue. These
are units bigger than municipalities and represent neighboring places with physical
connections. Yet, they are smaller than departments where the sign of a mass killing
could get lost over large territorial extensions.
Second, massacres did not have e ects before their occurrence, or there were no
anticipation of mass killings. Given the unexpected nature of a massacre, we think
this is a realistic assumption. According to the literature, the population cannot fully
predict mass killings and therefore change their behavior before a massacre (Ibáñez
and Vélez, 2008; Steele, 2018).
Third, the treatment assignment was random, conditional on the observable covari-
ates and the pre-intervention path of the outcomes. We found that this assumption is
valid in our setting. That is, the previous levels of our outcomes (COVID-19 cases and
mobility indices) did not a ect the hazard ratio of observing a massacre. We tested the
assumption by estimating a discrete-time hazard model using the method described by
Jenkins (1995). We modeled the probability of having the frst massacre at a given
week as a function of province fxed characteristics and time-varying covariates in a
duration dependence equation. We used the following specifcation:

(4.3) hpt = exp( 0 Xp + 0


Lpt + ct )

Where hpt is the hazard rate for having at least one massacre between February
15th , 2020, and October 6th , 2022. Xp includes time-invariant characteristics of each
PRETENDING TO BE THE LAW 15

province. Lpt is a set of time-variable aspects measuring past COVID-19 levels and
community mobility. ct are week dummies that control for duration dependence.

5. Results

5.1. Addressing the randomized treatment assumption. Before showing the


results from the ASMC model, we presented evidence supporting that the level of
COVID-19 cases in the past did not predict massacres in the future (i.e., plausibility
of conditional treatment assumption). By estimating Equation 4.3, Table 1, Column 1
shows that positive cases from one to four weeks ago did not explain the likelihood of
having a massacre. The coeÿcient of previous infection levels was non-signifcant and
close to zero.
An alternative hypothesis was that illegal groups acted when they perceived a rise in
human movements. We tested this explanation by introducing average mobility levels
at workplaces and parks before the frst massacre. Table 1, Column 2 shows that the
likelihood of mass killings did not increase more in places with high levels of commutes
to workplaces than in areas with low trips to work. Similarly, we did not see a variation
in the probability of having a massacre depending on lag trips to parks (Column 3 in
Table 1).
We also tested the relation of massacres with other province-fxed characteristics.
We found population size, density, and share of the rural population increased the
probability of massacre. This evidence is going in line with our hypothesis that the
illegal groups used massacres to control the population, and they used this type of
violence to incentivize social distancing. Moreover, variables related to the presence of
illicit activities, such as coca cultivation, traÿcking routes, or gold exploitation, do not
explain the incidence of a massacre. This rejects the possibility that we are capturing
the incidence of unintended consequences of disputes for illegal economies (see Table
1). Finally, previous levels of victimization also do not explain the incidence of the
PRETENDING TO BE THE LAW 16

massacres in our sample 21. This fact shows the massacres observed were not a common
practice before COVID, and the increase was not the result of groups performing regular
activities before the pandemic.
All together, these results present that illegal groups did not choose to increase
violence in response to high or low levels of COVID-19 transmission. Criminal groups
acted as regulators of social interactions in the territory and used massacres as a tool to
enforce social distance measures. The treatment was randomly assigned to provinces
conditional on donor pool, observable covariates, and pre-massacre path of the outcome.
Both pre-levels of infections and mobility were unrelated to massacres. Altogether, the
evidence shows that the randomized treatment asscumption is valid in this setting.

5.2. Results on community mobility. Our frst hypothesis was that illegal groups
used massacres as signals to incentivize illicit stay-at-home orders and reduced mobility
among community members as a strategy to prevent COVID-19 transmission. By
estimating the SMC model in Equation 4.2, Figure 6 shows the di erence in percentage
change of human movements compared to the frst weeks of 2020 between treated
provinces and synthetic control pre- and post- the frst massacre.22 As evidence that
the ASMC model created syntethtic controls similar to the treated unit, we do not fnd
any di erence in human movement between control and treated sub-regions before the
frst massacre. The left panel in Figure 6 shows a three percentage points decrease
in trips to work when comparing places with and without massacres, but the decline
is statistically signifcant after week 15. For travels to parks, we found a statistically
signifcant decrease of three percentage points one week after the frst massacre. Park
trip reduction continued in the following weeks by about six percentage points (see the
right Panel in Figure 6).

21The only exception is the previous massacres of illegal groups, BACRIMS. This variable seems to
increase the probability of having a massacre.
22We built standard errors and confdence intervals using a Jackknife method (see Ben-Michael, Feller,
and Rothstein, 2021b, sec 5.3).
PRETENDING TO BE THE LAW 17

We found a di erent story when we replicated the same exercise using the sample
of high coca suitable provinces. Figure A.VI in the Online Appendix shows a non-
signifcant increase in human movement compared with and without massacres. Our
interpretation of the opposite e ects in low and high coca suitability areas is that crim-
inal groups balance controlling the population and keeping the earnings from illegal
activities such as coca cultivation. People became more critical in non-coca-growing
areas when coca leaves were not a reliable source of income. This maximization be-
havior is not unusual in the Latin American region. Brazilian criminal organizations,
for example, encouraged some businesses to remain open since they needed them as
a source of revenue while forbidding social events (Miagusko and Da Motta, 2021;
Sampaio, 2021).

5.3. Results on COVID-19 transmission. We evaluated if the reduction in human


movement translated to a decline in COVID-19 transmission. Figure 7 presents the
estimated average di erence in new cases per 100.000 inhabitants between provinces
with massacres (treatment) and synthetic estimated areas (control). As a test that
the control units from the ASMC model matched treated provinces, Figure 7 shows
that the COVID-19 trend is similar in treated and synthetic units before the frst mass
killing.
Four mounts after the frst massacre, treated sub-regions saw 35 cases per 100.000
per week less than control units. This result was considerable in magnitude. On
average, 5 cases per 100.000 inhabitants by day is around half of the rate of infection
observed in small provinces during the frst months of the pandemic and one-quarter
of the infection rate in big provinces in a period with a low transmission rate.
The nature of COVID-19 transmission explains the gradual reduction of cases after
the frst massacre and the delay to observe a decrease on aggregate levels. The positive
test decreases came from declining trips to parks, a place with low transmission risk
(Althouse et al., 2020; Tenforde, Fisher, and Patel, 2021). The restriction of human
PRETENDING TO BE THE LAW 18

mobility to low-risk contagion areas slowly translated to a drop in the total number of
cases.

5.3.1. E ects by Age. As heterogeneous e ects, we tested the model for di erent age
groups. Younger people were more likely to transmit COVID-19, while older people
were more likely to die from the virus (Davies et al., 2020). In Figure 8, we estimated
the di erence in COVID-19 cases between sub-regions with massacres and synthetic
control units by age. As a test that the ASCM mode built synthetic controls that
correctly emulate treated provinces, we did not fnd any statistically signifcant e ect
in the six categories before the frst massacre.
Overall, we found that the reduction in COVID-19 events after a mass killing comes
from a statistically signifcant decline in cases of people aged 15 to 29 and 30 to 34 (on
average, 30 cases and 50 cases per 100.000 inhabitants, respectively). We did not fnd
a signifcant decrease in children from 0 to 14 years old and adults from 45 years old
or older (see Figure 8). This result is congruent with our previous fnding showing a
reduction in travel to parks since people from 15 to 34 years old in general have more
social activities and networks.

5.3.2. E ect on Deaths. We checked if the reduction in a positive test also translated
into a decline in the CPVOD-19-related deaths. Figure 9 shows a non-statistically
signifcant di erence in fatalities between synthetic control and treated provinces. Only
20 weeks after the frst massacres, we estimated a reduction of the daily rate of around
0.4 cases per 100.000 inhabitants.
We looked at the di erential e ects in mortality rate by age. As before, we also did
not fnd any signifcant impact on COVID-19 deaths. We found only a slight decline
in deaths in people 45 years old or older (see Figure A.VII in the Online Appendix).23
This result implies that reducing virus transmission levels was insuÿcient to alter
23Estimation for younger groups was not informative, given the low mortality levels among young
individuals.
PRETENDING TO BE THE LAW 19

COVID-19 death. Massacres a ected the young population, who are less likely to die
from the pandemic.

6. Robustness cheeks

The frst robustness was a placebo time type. We argue that the principal purpose
of massacres during the pandemic was to reduce human movement. However, was
the decline in park trips a typical community response after a massacre? If it was,
we could not claim that mass killings aim to reduce mobility in the pandemic. To
test the alternative explanation that mobility typically decreases after a massacre, we
replicated the estimation of changes in visits to parks after a mass killing in 2022.24
Figure A.VIII in the Online Appendix shows that human movements did not change
after the mass killings in between October 2021 and October 2022. This result shows
that the illegal groups only used massacres to reduce mobility during the pandemic
and other reasons motivated mass killings after the COVID-19 outbreak.25
The second robustness test validated if massacres a ected the tracking of COVID-19
cases. If this were the situation, the observed lower COVID-19 cases in our main result
was not a product of a reduction in disease levels but a decline in tests conducted
to identify the illness. We conducted several experiments to evaluate this alternative
explanation. As a result, Figure A.IX in the Online Appendix shows that the time to
detect a new COVID-19 case was not di erent in areas with and without massacres.
Figure A.X presents non-changes in the number of tests conducted at the deparment
level in areas with mass killings.26 Finally, Table A.IV shows no correlation between
24Two years after the frst COVID-19 case in March 2022 and the lockdowns to control the pandemic
in September 2020, Colombia has completely opened curfews and eliminated restrictions to prevent
the spread of the virus.
25It is important to mention that this period includes general election camping and that, historically,
these periods are characterized by high levels of violence and therefore and increase in the number of
massacres.
26Colombia did not provide detailed data on the daily number of tests conducted at the municipality
level. INS only sporadically updated data with municipal cumulative tests. We use data on daily
tests performed and available from May 9th , 2021.
PRETENDING TO BE THE LAW 20

COVID-19 testing, massacres, and victims in massacres. Based on these results, mas-
sacres do not a ected access to COVID-19 tests.
The third robustness test validated if our results are consistent with a subsample
selection. We estimated Equation 2 for only the sample of areas where human mobility
data is available. As in the main results COVID-19 cases decreased more in provinces
with massacres than without massacres (see Figure A.XI in the Online Appendix). The
results were analogous to Figure 7 for all Colombian provinces and therefore we are
not concern that sample selections and the lack of full information in mobility trends
biased our results.

7. Conclusion

The COVID-19 pandemic drained the fnancial and operational resources of countries
around the planet. Weak states faced the trade-o between fghting crime or containing
the virus. In this context, illegal armies saw the pandemic as an opportunity to show
power and become a viable alternative government. In the Colombian case, we observed
how once the government dropped all the lockdowns to prevent the pandemic, criminal
groups became the facto police to control the COVID-19 outbreak. But these police
activities came with the use of massacres as a deterrence measure of social activities
but not economic activities. Illegal groups aimed to contain the spread without hurting
the economic activities that were the source of the funding.
We tested the e ect of massacres between March 24th , 2020, and March 31st , 2021,
on human movement and COVID-19 cases in areas. This period went from the frst
Colombian lockdown to one month before massive protests and riots in the country.
We focused our analysis on regions with low coca cultivation to rule out that massacres
were used to control illicit economies.
We used Augmented Synthetic Control Method (ASCM) to build a synthetic control
unit that matches the pre-levels of cases and mobility of sub-regions with massacres.
PRETENDING TO BE THE LAW 21

Our fndings showed that illegal groups used massacres to reduce mobility to parks
but not workplaces. We found that these reductions translated into a modest fall in
positive COVID-19 cases, but only after four months from the frst mass killing. The
decrease in cases came mainly from people between 15-44 years old, the population
group with more social activities. Finally, the decline in mobility did not translate to
a signifcant decrease in COVID-19 deaths.
Overall, we argue that the increase in massacres resulted from the absence of the
legitimate state’s e ort to control the territory. With the presence of illegal armies,
these groups saw an opportunity to gain the favor of local communities acting as
pandemic containers in their territories. However, given their lack of information, they
opted to use violence to deter the social gatherings that they considered dangerous for
the spread of the disease. We proved that these measures were ine ective in containing
the disease and preventing deaths. With this in mind, explaining the reasons for
implementing or withdrawing curfews is essential to reduce the lack of information to
justify violence against civilians.
After moving from pandemic to endemic, COVID-19 deepened poverty and vulnera-
ble conditions, mainly in areas where illegal armies fght for the control population. In
this context, a good explanation of government decisions is vital to avoid unintended
consequences of public policies. Good communication will prevent illegal groups’ use
of these measures to justify the use of violence against civilians. Territorial control is
not enough to reduce violence. The state must support its e orts to control territory
with good relations with local communities that allow them to convey the logic behind
its policies.
PRETENDING TO BE THE LAW 22

References
Abadie, A., A. Diamond, and J. Hainmueller. 2010. “Synthetic Control Methods for
Comparative Case Studies: Estimating the E ect of California’s Tobacco Control
Program.” Journal of the American Statistical Association 105:493–505.
Abadie, A., and J. Gardeazabal. 2003. “The Economic Costs of Confict: A Case Study
of the Basque Country.” American Economic Review 93:113–132.
Abrams, D.S. 2021. “COVID and crime: An early empirical look.” Journal of public
economics 194:104344.
Acemoglu, D., C. Garcı́a-Jimeno, and J.A. Robinson. 2015. “State Capacity and Eco-
nomic Development: A Network Approach.” American Economic Review 105:2364–
2409.
ACLED. 2019. “Codebook.” https://acleddata.com/acleddatanew//wp-
content/uploads/2021/11/ACLED Codebook v1 January-2021.pdf. Accessed
th
September 6 2022.
Alesina, A., S. Piccolo, and P. Pinotti. 2018. “Organized Crime, Violence, and Politics.”
The Review of Economic Studies 86:457–499.
Ali, T.O., M. Hassan, and N. Hossain. 2021. “The moral and political economy of the
pandemic in Bangladesh: Weak states and strong societies during Covid-19.” World
Development 137:105216.
Althouse, B.M., E.A. Wenger, J.C. Miller, S.V. Scarpino, A. Allard, L. Hébert-
Dufresne, and H. Hu. 2020. “Superspreading events in the transmission dynamics
of SARS-CoV-2: Opportunities for interventions and control.” PLOS Biology 18:1–
13.
Arango, C. 2020. FARC veinte años. De Marquetalia a la Uribe, vol. 15. Ediciones
Aurora.
Aranguren, M. 2001. Mi Confesion: Carlos Castano Revela sus Secretos. Oveja Negra.
Arjona, A. 2016. Rebelocracy: Social Order in the Colombian Civil War. Cambridge
Studies in Comparative Politics, Cambridge University Press.
Arregocés, H.A., R. Rojano, and G. Restrepo. 2021. “Impact of lockdown on particulate
matter concentrations in Colombia during the COVID-19 pandemic.” Science of The
Total Environment 764:142874.
Bagozzi, B.E. 2016. “On Malaria and the Duration of Civil War.” Journal of Confict
Resolution 60:813–839.
Balcells, L. 2017. Rivalry and Revenge: The Politics of Violence during Civil War.
Cambridge Studies in Comparative Politics, Cambridge University Press.
Barceló, J., R. Kubinec, C. Cheng, T.H. Rahn, and L. Messerschmidt. 2022. “Windows
of repression: Using COVID-19 policies against political dissidents?” Journal of
Peace Research 59:73–89.
Ben-Michael, E., A. Feller, and J. Rothstein. 2021a. “The augmented synthetic control
method.” Journal of the American Statistical Association, pp. 1–34.
—. 2021b. “Synthetic Controls with Staggered Adoption.” Working paper, arXiv. org.
Berman, E., J.N. Shapiro, and J.H. Felter. 2011. “Can Hearts and Minds Be Bought?
The Economics of Counterinsurgency in Iraq.” Journal of Political Economy 119:766–
819.
PRETENDING TO BE THE LAW 23

Blattman, C. 2022. Why We Fight: The Roots of War and the Paths to Peace. Viking.
Cervellati, M., U. Sunde, and S. Valmori. 2017. “Pathogens, Weather Shocks and Civil
Conficts.” Economic Journal 127:2581–2616.
Clutterbuck, R. 1966. The Long, Long War: Counterinsurgency in Malaya and Viet-
nam. Frederick A. Praeger.
Colombia Investigative Unit. 2020. “Criminal Governance Un-
der Coronavirus: How Colombian Groups Seized the Day.”
https://insightcrime.org/news/analysis/criminal-governance-coronavirus-
colombia/. Accessed September 6th 2022.
Condra, L., and J. Shapiro. 2012. “Who Takes the Blame? The Strategic E ects of
Collateral Damage.” American Journal of Political Science 56:167–187.
Davies, N.G., P. Klepac, Y. Liu, K. Prem, M. Jit, and R.M. Eggo. 2020. “Age-
dependent e ects in the transmission and control of COVID-19 epidemics.” Nature
medicine 26:1205–1211.
Duranton, G. 2016. “Agglomeration e ects in Colombia.” Journal of Regional Science
56:210–238.
Fearon, J.D., and D.D. Laitin. 2003. “Ethnicity, Insurgency, and Civil War.” The Amer-
ican Political Science Review 97:75–90.
Galula, D. 1964. Counterinsurgency Warfare; Theory and Practice. Praeger Security
International.
Grasse, D., M. Pavlik, H. Matfess, and T.B. Curtice. 2021. “Opportunistic Repression:
Civilian Targeting by the State in Response to COVID-19.” International Security
46:130–165.
Holmes, J.S., A.P. Mendizabal, D.S. De La Fuente, M. Callenes, and A. ´ Cárdenas.
2021. “Paramilitary violence in colombia: A multilevel negative binomial analysis.”
Defence and Peace Economics 32:193–219.
HRW. 2020. “Colombia: Armed Groups’ Brutal Covid-19 Measures.”
https://www.hrw.org/news/2020/07/15/colombia-armed-groups-brutal-covid-
19-measures. Accessed September 6th 2021.
Huber, L. 2019. “When Civilians Are Attacked: Gender Equality and Terrorist Tar-
geting.” Journal of Confict Resolution 63:2289–2318.
Humpreys, M., and J. Weinstein. 2006. “Handling and Manhandling Civilians in Civil
War.” American Political Science Review 100:429–447.
Ibáñez, A.M., and C.E. Vélez. 2008. “Civil confict and forced migration: The micro
determinants and welfare losses of displacement in Colombia.” World Development
36:659–676.
Indepaz. 2021. “Informe de Masacres en Colombia Durante el 2020 - 2021.”
http://www.indepaz.org.co/informe-de-masacres-en-colombia-durante-el-2020-
2021/. Accessed April 9th 2021.
INS. 2022. “Boletines Casos COVID-19 Colombia.”
https://www.ins.gov.co/Paginas/Boletines-casos-COVID-19-Colombia.aspx. Ac-
th
cessed Septemeber 30 2022.
Jenkins, S. 1995. “Easy Estimation Methods for Discrete-Time Duration Models.” Ox-
ford Bulletin of Economics and Statistics 57:129–38.
PRETENDING TO BE THE LAW 24

Kalyvas, S.N. 2006. The Logic of Violence in Civil War. Cambridge Studies in Com-
parative Politics, Cambridge University Press.
Kocher, M.A., T.B. Pepinsky, and S.N. Kalyvas. 2011. “Aerial Bombing and Counterin-
surgency in the Vietnam War.” American Journal of Political Science 55:201–218.
Koehnlein, B., and O. Koren. 2022. “COVID-19, state capacity, and political violence
by non-state actors.” Journal of Peace Research 59:90–104.
Kustra, T. 2017. “HIV/AIDS, Life Expectancy, and the Opportunity Cost Model of
Civil War.” Journal of Confict Resolution 61:2130–2157.
la Calle, L.D. 2017. “Compliance vs. constraints: A theory of rebal targeting in civil
war.” Journal of Peace Research 54:427–441.
Llanes, L.M. 2022. “The killing of social leaders: An unintended e ect of Colombia’s
illicit crop substitution program.” International Journal of Drug Policy 101:103550.
LLC Google. 2021. “Google COVID-19 Community Mobility Reports.”
https://www.google.com/covid19/mobility/. Accessed April 9th 2021.
Lyall, J. 2009. “Does Indiscriminate Violence Incite Insurgent Attacks?: Evidence from
Chechnya.” Journal of Confict Resolution 53:331–362.
Lyall, J., G. Blair, and K. Imai. 2013. “Explaining Support for Combatants during
Wartime: A Survey Experiment in Afghanistan.” American Political Science Review
107:679–705.
Mejı́a, D., and P. Restrepo. 2015. “Bushes and Bullets: Illegal Cocaine Markets and
Violence in Colombia.” Working paper.
Miagusko, E., and J.W. Da Motta. 2021. “The pandemic, social isolation and
government on the margins: notes from the West Zone of Rio de Janeiro.”
https://cgum.sites.uu.nl/the-pandemic-social-isolation-and-government-on-the-
margins-notes-from-the-west-zone-of-rio-de-janeiro/. Accessed September 6th 2022.
Oslender, U. 2007. “Violence in development: the logic of forced displacement on
Colombia’s Pacifc coast.” Development in Practice 17:752–764.
Parada, J.J. 2022. “Colombia’s Peace Process: A Case Study of a Vexing Society
Struggling for Institutional Adjustment.” Journal of Economic Issues 56:335–347.
Prem, M., A. Rivera, D. Romero, and J.F. Vargas. 2022. “Selective civilian targeting:
The unintended consequences of partial peace.” Quarterly Journal of Political Science
17:317–354.
Raleigh, C., A. Linke, H. Hegre, and J. Karlsen. 2010. “Introducing ACLED: An armed
confict location and event dataset: Special data feature.” Journal of Peace Research
47:651–660.
Ramı́rez, J.C., and J. De Aguas. 2022. “Confguración territorial de las provincias en
Colombia.”, pp. .
Restrepo, J., M. Spagat, and J.F. Vargas. 2004. “The Dynamics of the Colombian Civil
Confict: A New Data Set.” Homo Oeconomicus 21:396–429.
Robinson, J.A., and R. Torvik. 2009. “The Real Swing Voter’s Curse.” American Eco-
nomic Review 99:310–15.
Sampaio, A. 2021. “Criminal Governance in Rio de Janeiro. Protection
and Predation in the Pandemic.” https://https://globalinitiative.net/wp-
content/uploads/2021/10/GIZ-PB-Rio-web.pdf. Accessed September 6th 2022.
PRETENDING TO BE THE LAW 25

Sánchez, F., I. España, et al. 2013. “Estructura, potencial y desafı́os del impuesto
predial en Colombia.” Working paper, Universidad de los Andes-CEDE.
Schwartz, R.A., and S. Straus. 2018. “What drives violence against civilians in civil
war? Evidence from Guatemala’s confict archives.” Journal of Peace Research
55:222–235.
Shaver, A., and J.N. Shapiro. 2021. “The E ect of Civilian Casualties on Wartime
Informing: Evidence from the Iraq War.” Journal of Confict Resolution 65:1337–
1377.
Steele, A. 2018. “IDP resettlement and collective targeting during civil wars: Evidence
from Colombia.” Journal of Peace Research 55:810–824.
Stoll, D. 1993. Between Two Armies in the Ixil Towns of Guatemala. Congressional
Research Service.
Sullivan, M.P. 2020. Latin America and the Caribbean: impact of COVID-19 . Congres-
sional Research Service.
Tenforde, M.W., K.A. Fisher, and M.M. Patel. 2021. “Identifying COVID-19 Risk
Through Observational Studies to Inform Control Measures.” JAMA 325:1464–1465.
Thompson, R. 1966. Defeating Communist Insurgency;: The Lessons of Malaya and
Vietnam. Frederick A. Praeger.
Toft, M.M., and Y.M. Zhukov. 2015. “Islamists and Nationalists: Rebel Motivation and
Counterinsurgency in Russia’s North Caucasus.” American Political Science Review
109:222–238.
Uwishema, O., L. Lawal, J. Sun, N. Hamiidah, A. Nasir, I. Robert, H. Onyeaka, S. Jain,
B.T. Akın, and M.R. Tovani-Palone. 2022. “Colombia protests amidst the COVID-
19 pandemic: implications and recommendations.” Medicine, Confict and Survival,
pp. 1–6.
Vargas, J. 2016. Strategic Atrocities Civilians under Crossfre: Theory and Evidence
from Colombia, Oxford University Press. pp. 425–451.
Wood, R. 2014. “Opportunities to kill or incentives for restraint? Rebel capabilities,
the origins of support, and civilian victimization in civil war.” Confict Management
and Peace Science, pp. 461–480.
Wright, A.L. 2016. “Economic Shocks and Rebel.” HiCN Working Papers No. 232,
Households in Confict Network, Sep.
PRETENDING TO BE THE LAW 26

8. Figures and Tables

Figure 1. Massacres and victims evolution

150
Total lift lockdown
30

Total lift lockdown


25
Number of massacres

Number of massacres
100
15 20

50
10
5

0
2018-1

2018-2

2018-3

2018-4

2019-1

2019-2

2019-3

2019-4

2020-1

2020-2

2020-3

2020-4

2021-1

2021-2

2021-3

2021-4

2022-1

2022-2

2022-3

2018-1

2018-2

2018-3

2018-4

2019-1

2019-2

2019-3

2019-4

2020-1

2020-2

2020-3

2020-4

2021-1

2021-2

2021-3

2021-4

2022-1

2022-2

2022-3
A. Total massacres B. Total victims

Notes: This graph shows the quarterly evolution of massacres between the frst quarter of 2018 and the third
quarter of 2022. Panel A shows the massacres, while Panel B shows the total number of civilians killed in
those events.
Source: ACLED (Raleigh et al., 2010)

Figure 2. Massacres evolution


Provinces with one massacre Massacres Victims
300
30

Smooth lift measures

Total lift lockdown


Lockdown start
Provinces with at least one massacre

200
Massacres and victims
20

100
10 0

04/01/2020 07/01/2020 10/01/2020 01/01/2021 04/01/2021


Date

Notes: This graph shows the evolution of massacres in our sample (provinces with low coca
suitability). On the left axis, we show the cumulative number of provinces that have experienced
at least one massacre since the pandemic’s beginning (March 24st 2020). On the right axis, we
show the evolution of the total number of massacres and victims over the same period.
Source: ACLED (Raleigh et al., 2010)
PRETENDING TO BE THE LAW 27

Figure 3. Community mobility at workplaces by occurrence of


massacres

With massacres With massacres


No massacres No massacres

0
0

Smooth lift measures


Smooth lift measures

Total lift lockdown


Total lift lockdown

Lockdown start
Lockdown start
Mobility percentage change at parks

Mobility percentage change at parks


-20
-20

-40
-40

-60
-60 -80

-80
04/01/2020 07/01/2020 10/01/2020 01/01/2021 04/01/2021 07/01/2021 04/01/2020 07/01/2020 10/01/2020 01/01/2021 04/01/2021 07/01/2021
Date Date

A. Small Provinces B. Big Provinces

Notes: This graph shows the evolution of community mobility at workplaces in our sample (provinces with
low coca suitability) by province size. We defne small provinces if the total population is below the 75th
percentile of population distribution in the sample. For each province, we average the date weighting by the
total population of each municipality; then, if there are gaps, we interpolate the data. Finally, we aggregate it
by each group (with and without massacres), taking into account again the population of each province. We
show 15 days mobile average in new COVID-19 cases per 100.000 inhabitants.

Figure 4. COVID-19 new cases by occurrence of massacres

With massacres With massacres


No massacres No massacres
40

80
New covid-19 cases per 100.000 Inhabitants

New covid-19 cases per 100.000 Inhabitants


30

60
Smooth lift measures

Total lift lockdown


Lockdown start

Smooth lift measures

Total lift lockdown


Lockdown start
20

40
10

20
0

04/01/2020 07/01/2020 10/01/2020 01/01/2021 04/01/2021 07/01/2021 04/01/2020 07/01/2020 10/01/2020 01/01/2021 04/01/2021 07/01/2021
Date Date

A. Small Provinces B. Big Provinces

Notes: This graph shows the evolution of new cases in our sample (provinces with low coca suitability) by
province size. We defne small provinces if the total population is below the 75th percentile of population
distribution in the sample. We show 15 days mobile average in new COVID-19 cases per 100.000 inhabitants.
Figure 5. Spatial distribution of massacres and COVID-19 per 100.000 inhabitants

No massacres
1 massacre
2 and 3 massacres
More than 3 massacres
341 − 2349
2349 − 3581
3581 − 4918
4918 − 7943
7943 − 13832
PRETENDING TO BE THE LAW

Notes: The map on the left presents the spatial distribution of massacres from March 24th 2020 to March 31st 2021. Shaded regions represent places with
high coca suitability. The map on the right shows the spatial distribution of cumulative COVID cases by September 30st 2021.
28
PRETENDING TO BE THE LAW 29

Figure 6. E ect of a massacre mobility


5

5
Estimated effect on p.p. mobility comparison

Estimated effect on p.p. mobility comparison


-5 0
0

-10 -15
-5

-15 -10 -5 0 5 10 15 20 -15 -10 -5 0 5 10 15 20


Weeks since first massacre Weeks since first massacre

A. Workplaces B. Parks

Notes: This graph shows the estimated di erence in percentage change of community mobility compared
to the frst weeks of 2020 between treated provinces and synthetic control pre- and post- the frst massacre.
The sample includes only provinces with low coca suitability. The shaded region represents a 90% confdence
interval. For the workplaces parameter  is 0.19. The scale imbalance coming from the average of treated
units is 0.14 while coming only from treated units is 0.39. For the workplaces parameter  is 0.18. The scale
imbalance coming from the average of treated units is 0.18 while coming only from treated units is 0.37.

Figure 7. E ect of a massacre on COVID-19 transmission


50
Estimated effect on new cases per 100.000
-50 -100 0

-25 -20 -15 -10 -5 0 5 10 15 20 25 30


Weeks since first massacre

Notes: This graph shows the estimated di erence in the number of new cases per 100.000
inhabitants between treated provinces and synthetic control pre- and post- the frst massacre.
The sample includes only provinces with low coca suitability. The shaded region represents a
90% confdence interval. Parameter  is 0.34. The scale imbalance coming from the average of
treated units is 0.21 while coming only from treated units is 0.42.
PRETENDING TO BE THE LAW 30

Figure 8. E ect of a massacre on COVID-19 transmission by age

100
50
20
Estimated effect on new cases per 100.000

Estimated effect on new cases per 100.000

Estimated effect on new cases per 100.000


50
10

0
0

-50

-50
-10

-100
-100
-20

-25 -20 -15 -10 -5 0 5 10 15 20 25 30 -25 -20 -15 -10 -5 0 5 10 15 20 25 30 -25 -20 -15 -10 -5 0 5 10 15 20 25 30
Weeks since first massacre Weeks since first massacre Weeks since first massacre

A. Age 0 to 14 B. Age 15 to 29 C. Age 30 to 44


100

100

100
Estimated effect on new cases per 100.000

Estimated effect on new cases per 100.000

Estimated effect on new cases per 100.000


50

50

50
0

0
-50

-50

-50
-100

-100

-100
-25 -20 -15 -10 -5 0 5 10 15 20 25 30 -25 -20 -15 -10 -5 0 5 10 15 20 25 30 -25 -20 -15 -10 -5 0 5 10 15 20 25 30
Weeks since first massacre Weeks since first massacre Weeks since first massacre

D. Age 45 to 59 E. Age 60 to 74 F. Age more than 75

Notes: This graph shows the estimated di erence in the number of new cases per 100.000 inhabitants between
treated provinces and synthetic control pre- and post- the frst massacre. The sample includes only provinces
with low coca suitability. The shaded region represents a 90% confdence interval.
PRETENDING TO BE THE LAW 31

Figure 9. E ect of a massacre on COVID-19 related deaths

3
Estimated effect on new deaths per 100.000
-1 0 -2 1 2

-25 -20 -15 -10 -5 0 5 10 15 20 25 30 35


Weeks since first massacre

Notes: This graph shows the estimated di erence in the number of new deaths per 100.000
inhabitants between treated provinces and synthetic control pre- and post- the frst massacre.
The sample includes only provinces with low coca suitability. The shaded region represents a
90% confdence interval.
PRETENDING TO BE THE LAW 32

Table 1. Discrete-Time Hazard Estimate of the Probability of Having


at Least One Massacre

Previous Adding Workplace Adding Parks


Cases Mobility Mobility
(1) (2) (3)

Time-varying covariates:
Average cases t − 1 and t − 2 0.003 −0.000 0.001
(0.005) (0.006) (0.005)
Average cases t − 3 and t − 4 −0.007 −0.003 −0.003
(0.005) (0.006) (0.005)
Average percent change 0.009
work mobility t − 1 and t − 2 (0.048)
Average percent change 0.009
work mobility t − 3 and t − 4 (0.046)
Average percent change 0.012
parks mobility t − 1 and t − 2 (0.036)
Average percent change −0.005
parks mobility t − 3 and t − 4 (0.037)

Fixed characteristics before 2018:

Geographic

Altitude (km) −0.000 −0.000 −0.000


(0.000) (0.001) (0.001)
Province area (100 km2) −0.034 0.061 0.023
(0.064) (0.109) (0.099)
Distance to main city (km) 0.006 0.001 0.001
(0.005) (0.006) (0.007)
Density (inhabitants per km) 0.001 0.001 0.001
(0.001) (0.001) (0.001)

Basic socioeconomic

Log populations 1.656 1.481 1.520


(0.337) (0.451) (0.393)
Share of women 22.656 −10.187 −2.370
(31.250) (42.193) (48.067)
Rural share 3.668 3.900 5.151
(1.652) (2.347) (2.648)

Fiscal and state presence

Total income per capita (Thousand CLP) 4.958 2.623 3.221


(3.082) (5.103) (3.439)
Total expenditure per capita (Thousand CLP) −5.944 −4.673 −5.938
(3.435) (5.757) (4.220)
Justice ineÿciency index 0.740 6.237 8.395
(6.512) (7.296) (8.591)
Total number of institutions 1.941 2.812 2.157
(2.092) (2.726) (3.086)

Observations 3138 2261 2137


Duration dependence X X X
Continue...
PRETENDING TO BE THE LAW 33

Previous Adding Workplace Adding Parks


Cases Mobility Mobility
(1) (2) (3)

Illegal resources

Share of area suitable for coca 0.844 0.887 0.614


(0.936) (1.266) (1.137)
Share area with gold exploration 3.888 0.884 1.611
(3.619) (5.168) (5.406)
Share PNIS municipalities 1.353 0.710 −1.354
(1.211) (3.196) (2.555)
Illegal traÿcking routes −19.096 −22.934 −27.319
(15.191) (18.650) (18.022)

Violence and victimisation

Farc presence −3.027 −8.367 −3.936


(4.339) (12.475) (7.392)
Other illegal group presence −1.368 −0.575 −0.182
(0.893) (1.231) (1.236)
Share expelled population 1.103 1.899 1.242
(1.642) (2.025) (1.802)
Lands taking 0.133 0.042 0.173
(0.376) (0.531) (0.481)
Lands abandoned 0.041 0.051 0.053
(0.080) (0.081) (0.088)
Massacre victims: Guerilla 1.451 1.247 1.961
(1.141) (2.341) (2.205)
Massacre victims: Paramilitary 5.823 −2.332 3.797
(9.997) (14.288) (12.246)
Massacre victims: Bacrim 9.003 10.234 10.771
(3.589) (6.544) (5.942)

Observations 3138 2261 2137


Duration dependence X X X
Notes: This table presents proportional hazard estimates for the sample of low-coca suitability provinces.
Each column represents a separate regression. Column 2 adds previous work places mobility comparison.
Column 3 adds previous parks mobility comparison. * is signifcant at the 10% level, ** is signifcant at
the 5% level, *** is signifcant at the 1% level.
PRETENDING TO BE THE LAW i

ONLINE APPENDIX
Figure A.I. Massacres evolution 1996-2014

150
Number of massacres
50 100 0
1996q1

1997q1

1998q1

1999q1

2000q1

2001q1

2002q1

2003q1

2004q1

2005q1

2006q1

2007q1

2008q1

2009q1

2010q1

2011q1

2012q1

2013q1

2014q1
quarter

Notes: This graph shows the quarterly evolution of COVID-19 in Colombia between 1996 and
2014. Source is Restrepo, Spagat, and Vargas (2004) original data and updated through 2014
by Universidad del Rosario.

Figure A.II. COVID-19 evolution


1.5
60

New covid-19 deaths per 100.000 Inhabitants


New covid-19 cases per 100.000 Inhabitants

Smooth lift measures

Smooth lift measures


Total lift lockdown

Total lift lockdown


Lockdown start

Lockdown start
40

1
20

.5
0

01/01/2020 07/01/2020 01/01/2021 07/01/2021 01/01/2022 07/01/2022 01/01/2020 07/01/2020 01/01/2021 07/01/2021 01/01/2022 07/01/2022
Date Date

A. New COVID-19 cases B. New COVID-19 deaths

Notes: This graph shows the evolution of COVID-19 in Colombia. The graph shows the number
of new cases and deaths per 100.000 inhabitants. Numbers are the weekly moving average.
PRETENDING TO BE THE LAW ii

Figure A.III. Massacres and victims evolution by coca suitability

High coca suitability Low coca suitability High coca suitability Low coca suitability
Total lift lockdown
30

100
Total lift lockdown

80
Number of massacres

Number of massacres
20

40 60
10

20
0

0
2018-1

2018-2

2018-3

2018-4

2019-1

2019-2

2019-3

2019-4

2020-1

2020-2

2020-3

2020-4

2021-1

2021-2

2021-3

2021-4

2022-1

2022-2

2022-3

2018-1

2018-2

2018-3

2018-4

2019-1

2019-2

2019-3

2019-4

2020-1

2020-2

2020-3

2020-4

2021-1

2021-2

2021-3

2021-4

2022-1

2022-2

2022-3
A. Total massacres B. Total victims

Notes: This graph shows the quarterly evolution of massacres between January 1st 2020 and September
30th 2022 according if the massacre occurred in a municipality with high coca suitability. Panel A shows the
massacres, while Panel B shows the total number of civilians killed in those events.
Source: ACLED (Raleigh et al., 2010)

Figure A.IV. Massacres and victims evolution by PNIS status

PNIS Municipalities No PNIS PNIS Municipalities No PNIS


100

Total lift lockdown


30

Total lift lockdown


80
Number of massacres

Number of massacres
20

40 60
10

20
0

0
2018-1

2018-2

2018-3

2018-4

2019-1

2019-2

2019-3

2019-4

2020-1

2020-2

2020-3

2020-4

2021-1

2021-2

2021-3

2021-4

2022-1

2022-2

2022-3

2018-1

2018-2

2018-3

2018-4

2019-1

2019-2

2019-3

2019-4

2020-1

2020-2

2020-3

2020-4

2021-1

2021-2

2021-3

2021-4

2022-1

2022-2

2022-3

A. Total massacres B. Total victims

Notes: This graph shows the quarterly evolution of massacres between January 1st 2020 and September 30th
2022 according if the massacre occurred in PNIS municipality. Panel A shows the massacres, while Panel B
shows the total number of civilians killed in those events.
Source: ACLED (Raleigh et al., 2010)
PRETENDING TO BE THE LAW iii

Figure A.V. New deaths COVID by Occurrence of Massacres

With massacres With massacres


No massacres No massacres

2
1
New covid-19 deaths per 100.000 Inhabitants

New covid-19 deaths per 100.000 Inhabitants

Smooth lift measures

Total lift lockdown


Lockdown start
.8

1.5
Smooth lift measures

Total lift lockdown


Lockdown start
.6

1
.4

.5
.2 0

0
04/01/2020 07/01/2020 10/01/2020 01/01/2021 04/01/2021 07/01/2021 04/01/2020 07/01/2020 10/01/2020 01/01/2021 04/01/2021 07/01/2021
Date Date

A. Small Provinces B. Big Provinces

Notes: This graph shows the evolution of new deaths in our sample (provinces with low coca suitability) by
province size. We defne small provinces if the total population is below the 75th percentile of population
distribution in the sample. We show 15 days mobile average in death COVID-19 cases per 100.000 inhabitants

Figure A.VI. E ect of a massacre on community mobility high


coca suitable provinces sample
10

20
Estimated effect on p.p. mobility comparison

Estimated effect on p.p. mobility comparison


10
0 5

-10 0
-5 -10

-20

-15 -10 -5 0 5 10 15 20 -15 -10 -5 0 5 10 15 20


Weeks since first massacre Weeks since first massacre

A. Workplaces B. Parks

Notes: This graph shows the estimated di erence in percentage change of community mobility comparing
to the frst weeks of 2020 between provinces that have a massacre and synthetic control pre- and post- the
frst massacre. Sample includes only provinces with high coca suitability. The shaded region represents a 90%
confdence interval.
PRETENDING TO BE THE LAW iv

Figure A.VII. E ect of a massacre on COVID-19 death by age


2

4
Estimated effect on new deaths per 100.000

Estimated effect on new deaths per 100.000


2
1

0
0

-2
-1

-4
-2

-25 -20 -15 -10 -5 0 5 10 15 20 25 30 35 -25 -20 -15 -10 -5 0 5 10 15 20 25 30 35


Weeks since first massacre Weeks since first massacre

A. Age 30 to 44 B. Age 45 to 59
20

40
Estimated effect on new deaths per 100.000

Estimated effect on new deaths per 100.000


10

20
0

0
-10

-20
-20

-40

-25 -20 -15 -10 -5 0 5 10 15 20 25 30 35 -25 -20 -15 -10 -5 0 5 10 15 20 25 30 35


Weeks since first massacre Weeks since first massacre

C. Age 60 to 74 D. Age more than 75

Notes: This graph shows the estimated di erence in the number of new deaths per 100.000 inhabitants
between treated provinces and synthetic control pre- and post- the frst massacre. The sample includes only
provinces with low coca suitability. The shaded region represents a 90% confdence interval.
PRETENDING TO BE THE LAW v

Figure A.VIII. E ect of a massacre in 2022 on community mobility


10

20
Estimated effect on p.p. mobility comparison

Estimated effect on p.p. mobility comparison


0

10
-10

0
-20

-10
-30

-15 -10 -5 0 5 10 15 20 -15 -10 -5 0 5 10 15 20


Weeks since first massacre Weeks since first massacre

A. Workplaces B. Parks

Notes: This graph shows the estimated di erence in percentage change of community mobility comparing to
the frst weeks of 2020 between provinces that have a massacre in 2022 and synthetic control pre- and post-
the frst massacre. Sample includes only provinces with low coca suitability. The shaded region represents a
90% confdence interval.

Figure A.IX. E ect of a massacre on the gap between the day


of frst symptoms and diagnostic
4
Effect on days between diagnostic and symptoms
-2 0-4 2

-25 -20 -15 -10 -5 0 5 10 15 20 25 30


Weeks since first massacre

Notes: This graph shows the estimated di erence in number of days that passed between the
appearance of symptoms and diagnostic, between treated provinces and synthetic control pre-
and post- the frst massacre. Sample includes only provinces with low coca suitability. The
shaded region represents a 90% confdence interval.
PRETENDING TO BE THE LAW vi

Figure A.X. Test per 100.000 Inhabitants – Departments with


massacres

60
Test per 100.000 Inhabitants
20 0 40

-8 0 8 16 24
Days since first massacre

Notes: This graph shows the number of test conducted at departments that experienced a
massacre in the days around the date of occurrence of a massacre.

Figure A.XI. E ect of a massacre on COVID-19 transmission


Regions with Google data
50
50
Estimated effect on p.p. mobility comparison

Estimated effect on p.p. mobility comparison


0
0

-50
-50 -100

-100

-25 -20 -15 -10 -5 0 5 10 15 20 25 30 -25 -20 -15 -10 -5 0 5 10 15 20 25 30


Weeks since first massacre Weeks since first massacre

A. Workplaces sample B. Parks sample

Notes: This graph shows the estimated di erence in the number of new cases per 100.000 inhabitants between
treated provinces and synthetic control pre- and post- the frst massacre. The sample in the fgure on panel A
includes only provinces with complete information on workplace mobility. The sample in the fgure on panel
B includes only provinces with complete information on park mobility. The shaded region represents a 90%
confdence interval.
PRETENDING TO BE THE LAW vii

Table A.I. Variables and sources

Variables Defnitions Source


The Armed Confict Loca-
Total number of massacres in the municipal- tion & Event Data Project
Number of massacres ity. (ACLED) Raleigh et al.
(2010)

Accumulated levels of infec- Total COVID-19 cases on one municipality


Instituto Nacional de Salud
tions
Daily municipality movement variation at
Google community mobility
Mobility at workplaces workplaces comparing visiting time in rela-
report -Google
tion to a baseline day.

Daily municipality movement variation at


parks comparing visiting time in relation to
Google community mobility
Mobility at parks a baseline day. Parks include places like local
report -Google
parks, national parks, public beaches, mari-
nas, dog parks, plazas, and public gardens.

Average altitude of the municipality in me- Instituto Agustı́n Codazzi


Altitude
ters above sea level.

Area Area of municipality in square kilometres. Instituto Agustı́n Codazzi

Lineal distance to the Department capital for CEDE calculations based


Distance to main city on Instituto Agustı́n Co-
each municipality.
dazzi
Number of inhabitants per municipality Instituto Agustı́n Codazzi,
Population density
square kilometre. DANE.
Sánchez, España et al.
Municipality total income Total municipal GDP in Colombian pesos in.
(2013) and DANE, 2005
2018
census.
Total population Total population of municipality. DANE

Share of female population in the municipal-


Share of women DANE
ity.

Share of population outside urban


Share rural population DANE
centres in the municipality.

Sánchez, España et al.


Municipality total income Total municipal income in Colombian pesos.
(2013) and DANE, 2005
in 2018
census.
Sánchez, España et al.
Municipality total expendi- Total municipal expenditure in Colombian.
pesos in 2018 (2013) and DANE, 2005
ture
census.
Ratio of complaints against functionaries in Inspector General (Procu-
Justice ineÿciency index the judicial branch to total complaints. Mea- raduria).
sured from 2000 to 2010
Total number of municipality police post,
courts registry oÿces, public phones services Fundación Social - Colom-
oÿces, health center and hospitals, schools, bian NGO available in Ace-
Number of institutions moglu, Garcı́a-Jimeno, and
libraries, fre stations, public mail service of-
fces, jails and tax collection oÿces. Mea- Robinson (2015)
sured in 1995

Continue...
PRETENDING TO BE THE LAW viii

Variables Defnitions Source


Ecological time-invariant coca suitability
Share of area suitable for
measure using municipal geographic and Mejı́a and Restrepo (2015)
coca
weather characteristics.
Share of municipality area conceded for gold
Area with gold exploration Instituto Agustı́n Codazzi
mining.

Indicator if the municipality some families United Nations Oÿce on


PNIS municipality participated in the coca substitution pro- Drugs and Crime (UN-
gram. ODC).

Indicator if the municipality some families United Nations Oÿce on


PNIS municipality participated in the coca substitution pro- Drugs and Crime (UN-
gram. ODC).

Raw data from Ministry of


Route optimization from rebel unit location
Illegal traÿcking routes Defence. Calculation by
and drug transit points using road networks
Wright (2016)

Presence of FARC. Municipality with activi-


FARC presence ties (e.g. attacks , clashes) of FARC between Prem et al. (2022)
2011 and 2012
Presence of other illegal groups. Municipal-
Other illegal groups pres- ity with activities (e.g. attacks , clashes) of Prem et al. (2022)
ence other groups not FARC between 2012 and
2014
Sistema de Información Ge-
Share of expelled popula- Total number of expelled population between ográfca para la Planeación
tion 1984 and 2012. y el Ordenamiento Territo-
rial – SIGOT
Unidad Administrativa Es-
Share of land area grabbing by violence un- pecial de Gestión de Resti-
Lands taking
der the register of Colombia authorities tución de Tierras Despo-
jadas UAE-GRTD

Unidad Administrativa Es-


Share of land area abandoned after violence pecial de Gestión de Resti-
Lands abandoned under the register of Colombia authorities tución de Tierras Despo-
jadas UAE-GRTD

Universidad del Rosario and


Massacre victims: guerilla Victims of massacres by guerilla groups 2000- Restrepo, Spagat, and Var-
2012 gas (2004)

Universidad del Rosario and


Massacre victims: paramili- Victims of massacres by paramilitary groups
Restrepo, Spagat, and Var-
tary 2000-2012 gas (2004)

Universidad del Rosario and


Victims of massacres by criminal emergent
Massacre victims: Bacrim Restrepo, Spagat, and Var-
groups 2000-2012
gas (2004)

Notes: This table shows the source of variables we used at the municipality level (the minor level of
disaggregation of the data). In our analysis, we aggregated these variables at the national level.
PRETENDING TO BE THE LAW ix

Table A.II. Descriptive Statistics: Time-invariant variables

Mean Std. Dev. Min Max

Geographic

Altitude (km) 0.994 0.948 0.0 2.9


Province area (100 km2) 64.896 123.335 0.9 1023.9
Distance to main city (km) 78.444 59.860 0.0 325.5
Density (inhabitants per km) 149.411 492.627 0.2 3670.0

Basic socioeconomic

Log populations 11.961 1.027 9.0 15.2


Share of women 0.498 0.015 0.4 0.5
Rural share 0.473 0.203 0.0 0.9

Fiscal and state presence

Total income per capita (Thousand CLP) 1384.380 352.285 427.0 2497.3
Total expenditure per capita (Thousand CLP) 1450.031 376.034 400.9 2633.1
Justice ineÿciency index 0.076 0.044 0.0 0.2
Total number of institutions 0.233 0.193 0.0 1.5

Illegal resources

Share of area suitable for coca 0.375 0.314 0.0 1.0


Share area with gold exploration 0.027 0.056 0.0 0.3
Share PNIS municipalities 0.056 0.178 0.0 1.0
Illegal traÿcking routes 0.009 0.025 0.0 0.2

Violence and victimisation

Farc presence 0.013 0.073 0.0 0.6


Other illegal group presence 0.445 0.398 0.0 1.0
Share expelled population 0.173 0.215 0.0 1.1
Lands taking 0.352 0.640 0.0 4.0
Lands abandoned 1.759 2.296 0.0 11.1
Massacre victims: Guerilla 0.067 0.180 0.0 1.3
Massacre victims: Paramilitary 0.026 0.033 0.0 0.2
Massacre victims: Bacrim 0.034 0.076 0.0 0.5
Notes: The occurrence of massacres is measured as March 31st 2021. COVID-19 variables measured as
September 30th 2021. Control variables measured before 2018.
PRETENDING TO BE THE LAW x

Table A.III. Descriptive Statistics by massacres

No Massacre Massacre

Geographic

Altitude (km) 1118.163 721.357


(1001.701) (763.309)
Province area (100 km2) 66.296 61.824
(147.297) (34.344)
Distance to main city (km) 76.325 83.093
(55.607) (68.905)
Density (inhabitants per km) 106.654 243.237
(262.271) (789.918)

Basic socioeconomic

Log populations 11.730 12.468


(0.957) (1.004)
Share of women 0.497 0.500
(0.015) (0.013)
Rural share 0.486 0.444
(0.195) (0.220)

Fiscal and state presence

Total income per capita (Thousand CLP) 1390.932 1370.002


(362.670) (332.857)
Total expenditure per capita (Thousand CLP) 1461.974 1423.823
(385.277) (358.789)
Justice ineÿciency index 0.075 0.077
(0.045) (0.040)
Total number of institutions 0.228 0.244
(0.156) (0.258)

Illegal resources

Share of area suitable for coca 0.340 0.452


(0.314) (0.306)
Share area with gold exploration 0.018 0.048
(0.052) (0.058)
Share PNIS municipalities 0.034 0.102
(0.157) (0.212)
Illegal traÿcking routes 0.010 0.008
(0.026) (0.022)
Continue...
PRETENDING TO BE THE LAW xi

No Massacre Massacre

Violence and victimisation

Farc presence 0.005 0.029


(0.047) (0.110)
Other illegal group presence 0.406 0.531
(0.399) (0.387)
Share expelled population 0.124 0.278
(0.194) (0.223)
Lands taking 0.265 0.543
(0.518) (0.825)
Lands abandoned 1.051 3.312
(1.457) (2.970)
Massacre victims: Guerilla 0.055 0.094
(0.189) (0.156)
Massacre victims: Paramilitary 0.018 0.045
(0.022) (0.044)
Massacre victims: Bacrim 0.024 0.055
(0.072) (0.082)
Notes: The occurrence of massacres is measured as March 31st 2021. COVID-19 variables measured as
September 30th 2021. Control variables measured before 2018.
PRETENDING TO BE THE LAW xii

Table A.IV. Correlation test per 100 inhabitants with province char-
acteristics

All Small Big


Provinces Provinces Provinces
(1) (2) (3)

Geographic

Altitude (km) 0.000 −0.001 0.003


(0.002) (0.002) (0.002)
Province area (100 km2) −0.001 −0.001 −0.002
(0.000) (0.000) (0.001)
Distance to main city (km) −0.107 −0.078 −0.090
(0.018) (0.026) (0.050)
Density (inhabitants per km) 0.006 0.038 0.004
(0.001) (0.020) (0.002)

Basic socioeconomic

Log populations 7.176 3.910 8.378


(0.854) (1.038) (2.488)
Share of women 561.076 293.077 898.481
(85.016) (103.296) (251.904)
Rural share −36.669 −26.850 −46.092
(5.536) (7.986) (18.927)

Fiscal and state presence

Total income per capita (Thousand CLP) 0.010 0.008 0.015


(0.005) (0.006) (0.007)
Total expenditure per capita (Thousand CLP) 0.007 0.004 0.014
(0.004) (0.004) (0.006)
Justice ineÿciency index 37.670 5.595 235.297
(35.128) (34.614) (89.886)
Total number of institutions 8.219 −4.013 34.731
(7.387) (3.256) (26.616)

Illegal resources

Share of area suitable for coca −1.410 −0.944 −9.574


(3.050) (2.492) (14.758)
Share area with gold exploration 5.651 10.059 −114.983
(22.607) (15.551) (131.628)
Share PNIS municipalities −3.582 −1.329 −48.085
(4.530) (2.498) (24.087)
Illegal traÿcking routes 45.237 17.355 36.040
(39.152) (28.637) (82.550)

Observations 116 87 29
Continue...
PRETENDING TO BE THE LAW xiii

All Small Big


Provinces Provinces Provinces
(1) (2) (3)

Violence and victimisation

Farc presence −4.440 −3.423 −294.940


(2.567) (0.686) (309.648)
Other illegal group presence 7.273 −3.801 6.907
(3.187) (2.008) (9.407)
Share expelled population −13.380 −2.531 −43.222
(4.686) (2.177) (22.663)
Lands taking 3.539 −1.328 3.646
(2.675) (1.184) (3.638)
Lands abandoned −0.262 −0.075 −2.721
(0.440) (0.253) (2.274)
Massacre victims: Guerilla −7.795 −3.430 −34.414
(3.163) (1.897) (75.072)
Massacre victims: Paramilitary −7.698 6.972 −152.490
(23.712) (14.346) (143.954)
Massacre victims: Bacrim 3.947 3.259 74.135
(11.361) (6.917) (239.581)

Observations 116 87 29
Notes: This table presents univariate regressions based on province characteristics controlling by de-
partment fxed e ects. The occurrence of massacres is measured as March 31st 2021. Column 1 presents
estimated coeÿcient and standard errors from a regression for the number of test per capita in all provinces.
Column 2 presents the same regression for small provinces and column 3 presents the regression for big
provinces. * is signifcant at the 10% level, ** is signifcant at the 5% level, *** is signifcant at the 1%
level.
PRETENDING TO BE THE LAW xiv

Appendix B. Qualitative evidence: Massacres and COVID-19


This section presents qualitative evidence of event descriptions in Raleigh et al.
(2010). José Miguel Vivanco, Americas director at Human Rights Watch, stated that
armed groups in di erent communities in Colombia have used violence to enforce their
own regulations to curb the spread of Covid-19. He also highlighted that this harsh
enforcement of control demonstrates the longstanding failures of the state to provide
adequate protection to vulnerable communities in remote regions of the country.
• On April 27th , 2020, armed men from a FARC Dissident group threw a grenade,
shot and killed three men, and injured four others in Cauca. According to
reports, a fyer signed by the FARC circulated in the previous days in which
they threatened people who did not respect the quarantine imposed to prevent
the spread of coronavirus.
• On May 14th , 2020, unidentifed armed individuals killed three men inside a
house in Cauca. The attack’s motive is unknown, but in previous days, neigh-
bors received death threats for those who broke the obligatory isolation imple-
mented during the coronavirus outbreak.
• On August 10th , 2020, two students, aged 12 and 17, were killed in the town
of Leiva (Nariño) as they were on their way to school to hand in homework, as
there were no in-person classes due to the Covid-19 quarantine.

You might also like