Clusters of Violence in Mexico: An Analysis of Homicide

Rates from 2000 to 2012

María del Pilar Fuerte-Celis, María de los Dolores Sánchez-Castañeda

Journal of Latin American Geography, Volume 20, Number 1, April 2021,

pp. 99-133 (Article)

Published by University of Texas Press

DOI: https://doi.org/10.1353/lag.2021.0004

For additional information about this article

https://muse.jhu.edu/article/787928

[ Access provided at 22 May 2021 01:30 GMT from the University of Connecticut ]
 Clusters of Violence in Mexico: An Analysis
of Homicide Rates from 2000 to 2012
María del Pilar Fuerte-Celis
Cátedra CONACyT, Centro Geo, Aguascalientes
María de los Dolores Sánchez-Castañeda
Facultad de Ciencias, UNAM, México

ABSTRACT
In this article we offer a spatial analysis of the homicide rates registered in the municipalities
of Mexico from 2000 to 2012. We seek to identify the variables that can be associated from
local spaces with the occurrence of homicides, adopting a geospatial perspective based on
the study and understanding of violence. Our goal is to observe spatial patterns by which
municipalities can be grouped according to variables linked to the evolution of homicide
rates recorded before and during the period known as the beginning of the war on drugs.
We use geostatistical techniques, in particular Moran’s I Index and Getis-Ord G*i tests, and
validate the results using the random forest technique. We offer a classification proposal in
which we identify the presence of organized crime, population density patterns, and socio-
economic inequality as the main variables associated with the spatial patterns shown by
homicide rates in Mexico.

Keywords: clusters of violence, homicide rates, spatial analysis, Tobler, Mexico

Resumen
En este artículo ofrecemos un análisis espacial de las tasas de homicidios registradas en los
municipios de México durante el período de 2000 a 2012. Buscamos identificar aquellas
variables que desde los espacios locales pueden asociarse con la ocurrencia de homicidios,
adoptando una perspectiva geoespacial que abona en el estudio y comprensión de la violencia.
Nuestro objetivo es observar patrones espaciales que permitan agrupar a los municipios de
acuerdo a variables comunes y a la evolución de las tasas de homicidio registradas durante el
periodo 2000 al 2012, periodo que se conoce como el comienzo de la guerra contra las drogas.
Para ello, recurrimos al uso de técnicas geoestadísticas, particularmente usamos el índice I
de Moran y las pruebas de G*i de Getis-Ord, y validamos los resultados usando la técnica de
bosque aleatorio o random forest. Ofrecemos una propuesta de clasificación en la que iden-
tificamos la presencia del crimen organizado, los patrones de poblamiento y la desigualdad

JOURNAL OF LATIN AMERICAN GEOGRAPHY 20(1), 99–133

 Journal of Latin American Geography

socioeconómica como las principales variables asociadas a los patrones espaciales que presen-
tan las tasas de homicidio en México.

Palabras claves: aglomeración de violencia, tasas de homicidio, análisis espacial, Tobler,

México

Introduction According to Tobler’s first law of geog-

Since 2000, there has been an increase in raphy (Tobler, 1970; Westlund, 2013), in
violence at the hands of criminal groups in geographic space everything is related to
different regions of Mexico (Ríos, 2013). In everything else, but closer spaces are more
2006, the new government decided to adopt related to each other than to spaces farther
a different strategy to fight organized crime, away. In this vein, we present an analysis of
a war on drugs that has been blamed for a homicide rates in Mexican municipalities to
remarkable increase in levels of violence find out how similar they are to neighboring
(Calderón et al., 2015). In 2012, the strategy municipalities, and to identify which vari-
changed again, but high levels of violence ables are significant in explaining the differ-
continued, so it is useful to revisit this period ences (Goodchild, 1987; 2004).
in search of new explanations for the situa- Our working hypothesis is that the distri-
tion. bution of homicide rates is not random. We
Violence is a complex phenomenon where assume that there are spatial patterns and that
cultural, social, economic, and political certain variables are linked to the evolution
factors come into play. An empirical analy- of these rates. In other words, homicide rate
sis of the variables recorded can help us to values are partially a function of the value
find patterns and build indicators for a better of this same variable in neighboring areas
understanding and interpretation of it. In (Flint et al., 2000). To test this hypothesis,
particular, an analysis of the distribution of we use a geostatistical analysis and study the
homicide rates allows us to recognize spatial link between other variables that have been
patterns and to identify variables linked to identified as important in the literature on
the phenomenon, which is the main objec- violence.
tive of this article. We built a database from records of the
Are homicide rates distributed randomly following official sources: homicides in
in time and space? Are there any patterns Mexico from the National System of Vital
that might denote the presence of variables Statistics (INEGI, 2020a); demographic
that influence these rates? What variables are data from the Population Censuses of 2000,
linked to the evolution of homicide rates in 2005, and 2010 (INEGI, 2020b, 2020c, 2020d)
Mexico? Is there a spatial correlation (Good- and from CONAPO (2010) for inter-census
child, 1987; 2004; Westlund, 2013)? Answers years; socio-economic characteristics of the
to these questions could help us to under- municipalities (CONEVAL, 2010); data on
stand this period in Mexico’s history. the eradication of illicit crops (SEDENA,

100
 Clusters of Violence in Mexico

2010); and non-official records developed more violent the society (Folino, 2015). From
by trustworthy scholars, including a record this perspective, Latin America is one of the
of migration routes (Casillas, 2008) and a most violent regions in the world because it
record of violence caused by organized crime has the highest homicide rates, and unlike
(Atuesta et al., 2019). A significant finding is other regions of the world where the rates
that patterns in homicide rates during the are trending downward, in Latin America
period of study do show clusters of violence, they are on the rise (United Nations Office
which are groups of municipalities close to on Drugs and Crime [UNODC], 2019; Rett-
each other that have similar rates of violence berg, 2020).
and other variables associated with the A society’s homicide rate can be consid-
phenomenon. We identified eighteen clus- ered as a proxy variable to measure the
ters of violence that are different from each magnitude of violence (Sánchez Munguía,
other but similar on the inside. 2009; Merino, 2011; Loftin et al, 2015). Yet
In the first section of this text, we offer a we must also consider that we are reducing
literature review focused on studies about violence to its most extreme expression--ho-
Mexico and Latin America that we use to micides--setting aside the analysis of other
approach the topic of violence and the use of types of violence. Another caveat using
homicide records at the administrative level. this indicator is that homicide rates cannot
In the second section, we describe in detail explain why the homicides were committed
the methodology used to build the geosta- or their impact on other areas of political,
tistical models used in our analysis, and we economic, or social life.
present the results. In the third section, we In other words, even though it is a very
analyze the results in relation to the literature, powerful indicator, homicide rates do not
and we conclude with comments on how a measure violence directly, nor do they say
statistical and spatial analysis can contribute anything about the causes or effects of the
to understanding complex phenomena such homicides. Therefore, we need other data
as violence. to overcome these limitations. Recently, an
academic approach to homicide rates from
different angles has taken on strength, over-
Violence and Homicide coming the mainly journalistic perspective
Rates used for many years (Shirk & Wallamn, 2015;
A homicide rate is used to express the ratio Dávila-Cervantes & Pardo-Montaño, 2016;
of the population to homicides recorded in 2018; Arteaga- Botero et al., 2019).
a certain place during a certain period; it Some recent studies on Latin America link
refers to the number of annual homicides per homicide rates to the society’s political deter-
100,000 inhabitants. As a summarized indi- minant factors, placing special emphasis on
cator, the homicide rate does not measure power relations (Sanchez, 2006; Ceobanu et
violence specifically, but we can reasonably al., 2011). Some studies on Mexico consider
assume that the higher the homicide rate, the the strength of this indicator (Snyder &

101
 Journal of Latin American Geography

Durán- Martínez, 2009; Durán-Martínez, is the work by Pan et al. (2012), who analyze
2015; Ríos, 2013; Dell, 2015; Ponce, 2019; the impact on variables such as investment,
Ponce et al., 2019; Trejo & Ley, 2020); others growth, and the development of local state
focus on identifying the effects of law enforce- economies. In the same vein, the study by
ment on homicide rates (Guerrero-Gutiérrez, Liu et al. (2012) shows that there is a link
2011); and still others link homicide rates to between unemployment rates and homi-
arms trafficking and trade (García-Ponce et cide rates. Meanwhile, Márquez et al. (2015)
al., 2013; Castillo et al., 2020). find that the correlation between school
We found a set of articles about homicide enrollment and homicide rates is signifi-
(Spinelli et al, 2010; Zunino et al., 2012) that cant—there is an enrollment migration from
study the territorial evolution, distribution, municipalities with the highest rates toward
and association between municipalities and municipalities with the lowest rates.
regions. For instance, in 2012 Sánchez-Jabba From a spatial perspective, Osorio (2015)
et al. developed a study of the incidence and proposes a model based on the intensity
evolution of homicides at a municipal level of violence through turf wars among drug
in Colombia between 2003 to 2010. We see a cartels. Espinosa and Rubin (2015) use a
trend in the grouping of violent and peaceful statistical model to analyze the effects of mili-
municipalities. The authors use spatial meth- tary interventions on homicide rates. Along
odologies, in particular spatial correlation the same lines, Flores and Rodríguez-Oreg-
indexes (Sánchez-Jabba et al., 2012). A similar gia (2014) look for patterns of different levels
study was developed by Andrade and Diniz of violence in Mexican municipalities based
(2013), who give a detailed description of the on whether or not they had interventions
regional behavior of homicides in Brazil from against organized crime.
the 1980s to 2010. These authors study two In both Mexico and Latin America, there
processes of violent deaths. They call the first is a growing interest in understanding the
one internalization, which is characterized phenomenon of violence based on homicide
by the downturn in mortality by homicide rates since there is a serious problem with
within capital cities and metropolitan areas insecurity, and sorely needed public policies
but an expansion of it in other municipalities. must be effective if they are to have a consis-
The second process is diffusion, by which tent impact (Rettberg, 2020). In this sense,
Federal Units without tradition of violent spatial models have been used to detect clus-
deaths experience significant increases in ters with a high incidence of homicides as
cases. The authors reveal the existence of tools to create targeted preventative and
a reorganization of violence in Brazil that early-response campaigns.
obeys the logic of clustering of violent areas In their seminal research of the United
and clear contagion effects in neighboring States in the 1960s through 1980s, Land et
places (Andrade & Diniz, 2013). al. (1990) show which covariates have signif-
Another approach considers the impact icant effects on homicide rates from a list of
of homicide rates on society. One example different social, demographic, ecological, and

102
 Clusters of Violence in Mexico

economic variables. They use their results to explained by economic and social condi-
build a baseline model to estimate homicide tions than by individualist or psychological
rates consistently by different levels of aggre- aspects. For example, high homicide rates
gation, such as cities, metropolitan areas, tend to go hand in hand with low levels of
and more. Baller et al. (2001) use this model income and development (Burraston et al.,
to detect clusters with high homicide rates 2019); in other cases, an increase in inequal-
based on the Moran’s I Index and the Local ity and poverty levels would better explain an
Indicators of Spatial Autocorrelation (LISA) increase in violence (Fajnzylber et al., 2001;
test, and they include in their research a 2002).
consideration of differentiated spatial effects. Camara and Salama (2004) and Briceño-
Our literature review shows that import- León (2012) state that the link between
ant efforts have been undertaken to analyze poverty and homicide rates is not direct, and
homicide rates in the Americas from differ- we may even expect moderate homicide rates
ent angles (Imgram & Marchesini da Costa, in societies with low income levels and high
2014; Dell, 2015). In particular, the spatial levels of social inequality (Cámara & Salama,
approach developed by Osorio (2015), Flores 2004). Yet it is a fact that a change in socio-
and Rodríguez-Oreggia (2014), Guerrero- economic variables, such as a loss of purchas-
Gutiérrez (2011), Molzahn et al. (2012), ing power, affects homicide rates (Botello,
and Espinal-Enriquez and Larralde, (2015) 2014; Martínez-Cruz & Rodríguez-Castelán,
, among others, has proven to be useful but 2016).
needs more research to complement the Studies about our continent have focused
seminal work by Land et al. (1990) and on the analysis of variables such as lack of
Baller et al. (2001). We wish to contribute employment opportunities, income inequal-
to this approach, taking as the closest point ity, and expected profits for participating in
of reference the work by Espinal-Enríquez criminal activities and found those variables
and Larralde (2015), who study the distribu- to be associated with changes in homicide
tion of homicides in Mexico with a network rates (Liu et al., 2012). A general consen-
approach that allows them to find munic- sus on the link between the economy and
ipality nodes with the highest number of violence posits that the most significant vari-
interactions. able is inequality (Cámara & Salama, 2004),
although in the global literature high homi-
cide rates in Latin America are linked to the
Variables Linked to availability of weapons and the existence
Homicides and Violence of illegal activities (Sanjurjo, 2016), such as
Besides patterns of violence based on the general trafficking (of people and of differ-
spatial distribution of homicide rates, we ent products) and drug trafficking (Hopkins
need to look thoroughly at the causes and et al., 2013; Valdés, 2013). In these activities
effects of violence in society. One of the most violence is necessary for operations, so the
plausible hypotheses is that crime is better willingness to use weapons and to commit

103
 Journal of Latin American Geography

homicides increases or can be used to main- Meanwhile, Mora Bayo (2014) shows
tain control, reach a better position, affirm that for communities located in the moun-
personal status through vengeance killings tains of the state of Guerrero in Mexico, an
or punishments, or seek respect from other increase in violence responds to institutional
groups (Bailey & Taylor, 2009; Boyce et al., policies geared toward devaluing the life of
2015; Briscoe & Kalkman; 2016). anyone who is expendable to the neoliberal
From a demographic perspective, it is project, which also organizes the division
assumed that people who become victims or of labor and the extraction of legal and ille-
victimizers contribute to the growth in homi- gal resources. In an analysis of vulnerability,
cide rates—mostly men (Valdivia & Castro, Nygren (2018) states that apparent segmen-
2013). Moreover, they tend to be young men tation (or variability) may simply respond to
because, among other reasons, young people a rational organization that power promotes,
participate more in high-risk activities such as do Calderón (2015), who points out that,
as gang membership (UNODC, 2019). In although the presence of organized crime
other words, homicides tend to be concen- increases the levels of violence when it suits
trated in the demographic of men aged 20 to them, peace can also benefit them when they
39 (Gamlin, 2015). see that economic gains increase.
From a geographic perspective, the prox- These socio-anthropological approaches
imity to illegal markets and to places where show the need to look at violence carefully, so
merchandise is moved, such as ports and that even as we observe regularities through
especially borders, has been identified as quite descriptive statistical and spatial analy-
significant. Literature on civil wars identifies sis, we should not forget about the systemic
these areas as places full of violence linked and micro-sociological phenomena that may
to the control of highways, borders between explain the regularities observed.
towns, and essential zones of influence for
the strategies of armed groups (Cohen &
Tita, 1999; Haggett, 2000; Zeoli et al., 2014). Methodology to
More precise analyses note the effects of Integrate and Analyze
neoliberal policies on the increase in homi- the Data
cide rates; however, we did not use them Based on the literature review, and consider-
in this geostatistical model. For example, ing our goal to contribute to the use of geosta-
Estévez (2017) develops the idea that necrop- tistical methods to understand violence, we
olitical wars are a type of conflict that explains proceeded to create a database that would
both criminal and gender violence, character- include the records of the variables identified
ized by a dysfunctional socio-lawful space in the literature as significant for understand-
that is permanently corrupt and deliberately ing violence. The first input we used was the
lethal, and that guarantees the impunity of set of homicide rates aggregated by munic-
necropower’s technologies of death: massa- ipality and by year for the period between
cres, femicide, and forced disappearances. 2000 and 2012.

104
 Clusters of Violence in Mexico

Our main source of data was the National ence or absence of the trait to be observed for
Institute of Statistics and Geography each year—deaths linked to narcotrafficking,
(INEGI), which periodically brings together and the inclusion or not on migration routes.
and publishes different official records. We Also, we identified if each municipality is
used the national record of annual vital statis- located on the northern or southern border,
tics, the 2000 and 2010 population censuses, since this factor is also related to high levels
and the 2005 intercensal survey to obtain the of violence in the literature on violence.
sociodemographic, economic, and demo- Once our database was built, we did an
graphic variables such as population density, exploratory analysis of homicide rates and of
urban or rural classification, and economi- the rest of the variables to validate their inclu-
cally active population, among others. For sion in the study. Next, using spatial analysis
the intercensal years 2001 to 2004, 2006 to we built clusters of violence; finally, we used
2009, 2011, and 2012, we used the popula- the random forest technique to validate the
tion projections of the National Population clusters of violence we had proposed, and
Council (CONAPO). to identify the variables linked to homicide
We also used estimates done by the rates in Mexico.
National Council for the Evaluation of Social We chose the clusters-of-violence concept
Development Policy (CONEVAL), specifi- for two reasons: firstly, it is derived from
cally the social gap index for 2000, 2005, and the field of statistics (Duque et al., 2011),
2010, and the 2010 Migration Intensity Index, and secondly, it considers contributions of
in addition to the number of hectares of economic geography to the study of clusters.
cannabis destroyed by military forces during We define a cluster of violence as a group of
2006 to 2012, as published by the Ministry of municipalities that, besides spatial proxim-
National Defense (SEDENA). ity, share certain characteristics distributed
One of the non-official sources we used homogenously within but can be considered
was the database published by the Drug different from other clusters.
Policy Program (PPD–CIDE), which In statistics, the concept of clusters refers
includes the number of deaths allegedly asso- precisely to population groups that have
ciated with drug trafficking between 2006 homogeneity within but heterogeneity
and 2012. We also drew on the work done between them, which makes the analysis
by Casillas (2008), which gives a record easier through techniques developed for it,
of the routes followed by Central Ameri- such as sampling (Li et al., 2013). In certain
can migrants when crossing the country— conditions regarding the availability of tech-
since the migrant population has also been nical, human, time and material resources, it
linked to different levels of violence—and would be enough to study the differences
mainly those that overlap with the distribu- between groups and assume that they repre-
tion routes used by organized-crime groups. sent differences among the entire population.
With the non-official sources, we identify From a geographic perspective, research
whether each municipality shows the pres- on technological innovation and develop-

105
 Journal of Latin American Geography

ment has shown that physical proximity us to identify at least three periods, illus-
creates a “knowledge spillover” that also trated in Figure 1. The first trends down-
acts as a pole of attraction for other actors, ward, carried over from previous decades
thus producing a synergy between space and (Escalante, 2010) and including the period
knowledge (Fischer et al., 2006; Caragliu 2000 to 2004, with slight increases in 2005
& Nijkamp, 2013). On the other hand, it is and 2006, but continuing downward through
assumed that this knowledge spillover, also 2007. The second period is from 2008 to 2011,
called social capital, can sustain and explain when the trend reverses and the rates begin
cultural and economic situations associated to rise systematically until reaching their
with crime and violence. peak in 2011. The third period (which is not
In this sense, the concept of clusters of analyzed in this article) begins in 2012, corre-
violence plays a dual heuristic role; on the sponding to the change of government and
one hand, it allows us to identify groups of the election of a new president, and is char-
municipalities and makes it easier to design acterized by a slight drop in homicides but far
and implement targeted policies, and on the from the levels seen at the beginning of 2000.
other hand, it invites us to pay closer atten- Regarding demographic structure illus-
tion to the variables linked to homicide rates trated in Figure 2, we observe that close to
in each group. 70 percent of victims recorded were between
fifteen and forty-four years of age, putting
this age group at the highest risk of being
Spatial Analysis of a victim of homicide. More specifically, the
Homicide Rates 25-39 age group is at the highest risk of being
Overall, homicide rates surged 159 percent a victim, since almost 45 percent of homi-
from 2001 to 2012. The national average of cides registered in the period belong to that
those twelve years was 18.30 per 100,000 group.
inhabitants, and the most violent year was Taking into account the distribution of
2010 with a national average rate of 38.38. The homicides by federal entity shown in Figure
most peaceful year in the entire territory was 3, we find that the five states with the high-
2003, with an average rate of 10.27 homicides est percentage of homicides are the State
per 100,000 inhabitants. Compared to other of Mexico (13.4 percent), Chihuahua (12.8
countries in the region, Mexico is above percent), Guerrero (7.7 percent), Mexico
Latin American average but below levels seen City (6.5 percent), and Sinaloa (6 percent).
in the most violent countries such as El Salva- Together, these five federal entities make
dor and Venezuela, yet far above the least up 46.6 percent of all homicides recorded
violent Uruguay and Costa Rica (Sanjurjo, during the period of the study. At the other
2016; Muggah, 2017; Rettberg, 2020). end of the spectrum we find the states of
The exploratory analysis of annual percent- Aguascalientes (0.3 percent), Baja California
age of homicides in Mexico from the total Sur (0.2 percent), Campeche (0.4 percent),
observed during the whole period allows Tlaxcala (0.4 percent), and Yucatán (0.3

106
 Clusters of Violence in Mexico

Figure 1. Number
and percentage of
homicides by year.

Figure 2. Pyramid of
homicides by group
of age and sex, male
(left) and female
(right).

percent), which together make up 1.6 percent Culiacan, Chihuahua, Ecatepec de More-
of recorded homicides. los, Monterrey, Iztapalapa, Guadalajara, and
Finally, in the distribution of homicides Gustavo A. Madero recorded the highest
at the municipal level illustrated in Figure percentage of homicides in the country. The
4, we find that Juarez, Tijuana, Acapulco, twenty most violent municipalities make up

107
 Journal of Latin American Geography

Figure 3. Number
and percentage of
homicides by state.

Figure 4. Number
and percentage
of homicides by
municipality.

slightly more than 20 percent of the homi- For the annual spatial municipal distribu-
cides recorded during the period, which is tion of homicide rates, we do not present the
a very high number if we consider that they maps due to space restrictions; however, we
make up less than 1 percent of all municipali- describe our findings here. We observe that
ties in the country. the values of the rates are not constant, and

108
 Clusters of Violence in Mexico

that the change is associated with the place index, while the southern states like Guerrero,
and year the homicides were recorded. Some Oaxaca, and Chiapas have a very high social
states and municipalities are more violent gap. In the north, the social gap is medium
than others, and in some years violence to very low, but the municipalities in states
decreased while in others it increased. It is such as Tamaulipas and San Luis Potosi have
intuitive to note that the variation in homi- medium to high levels. And a strip of munic-
cide rates is linked to other variables. We thus ipalities running south from Sonora through
tested empirically the coincidences among the southwest of Chihuahua, the mountains
the municipal rates with spatially disaggre- of Sinaloa, western Durango, northeastern
gated sociodemographic variables. Nayarit, and northern Jalisco has a very high
We find that the states with the highest social gap index.
population density, illustrated in Figures 5 (a), Finally, in this exploratory analysis we
(b), and (c), have the highest percentage of were not able to identify any link between
homicides. The middle and southern part of homicide rates and migrant routes, but there
the country are the most densely populated is an overlap with indices of migration inten-
and the northern municipalities the least. Yet sity illustrated in Figure 6. Those municipal-
in the same state we can find a wide range of ities that attract more migrants have seen an
different values, especially in the north, since increase in homicide rates.
capital cities are very densely populated and
the rest of the municipalities have a very low identification of clusters of violence
population density. Spatial autocorrelation shows the degree to
As for the social gap index, which is a which objects or activities in a geographic
summarized measure of a set of variables unit are similar to other objects or activ-
that show the poverty level, access to basic ities in neighboring geographic units
services, and quality of life, and which is (Goodchild, 1987). A positive spatial autocor-
estimated with principal component anal- relation supports or confirms the existence
ysis (PCA), we find that its spatial distri- of Tobler’s principle, which considers that,
bution shown in Figures 5 (d), (e), and (f) in a geographic space, everything is related
is correlated with the presence of criminal to everything, but closer spaces are more
groups, perhaps because the higher the level related to each other than to spaces farther
of marginalization of the population, the away. In this sense, we attempt to identify
more vulnerable it is to illegal activities. This clusters or hotspots with a high spatial auto-
also calls into question the alleged social correlation of municipal homicide rates, and
policy of narcotrafficking, particularly in to assess the presence of patterns in order to
the area under the influence of the so-called group the municipalities by means of these
Sinaloa Cartel, the oldest in the country homicide rates.
(Váldez, 2013). The Moran’s I Index is one of the most
Those municipalities in the middle part of frequently used measurements of global
the country have a medium to high social gap spatial autocorrelation. Applying this tech-

109
 Journal of Latin American Geography

(a)

(b)

(c)

Figure 5. Municipal spatial distribution of: (a) Population density 2000; (b) Population density 2005;
(c) Population density 2010; (d) Social gap index 2000; (e) Social gap index 2005; (f) Social gap
index 2010.

110
Clusters of Violence in Mexico

(d)

(e)

(f)

111
 Journal of Latin American Geography

Figure 6. Municipal spatial distribution of Migration intensity index 2010.

nique to a variable allows us to assess the value and be surrounded by other units with
existence of Tobler’s principle. However, a high values. In general terms, local indicators
problem that may arise is that, even though make the most sense when analyzing why
the Moran’s I Index allows us to measure there is a global autocorrelation or not. (López
the global association of the characteristics Hernández & Palacios Sánchez, 2000).
analyzed, it does not show where the groups Positive Moran’s I values suggest a posi-
are. Therefore, we complement the analysis tive spatial association, meaning that there
with the Getis-Ord Index, which defines the is a group of neighboring units with similar
degree of grouping for high or low values, values; negative values suggest a negative
meaning that it measures the concentration spatial association, meaning that there are
of high or low values in a particular area of groups of values that differ from those in
the study. neighboring units. Meanwhile, positive or
The Getis and Ord Gi (Gi*) tests analyze negative non-significant values suggest that
each geographical unit within the context of there is a lack of spatial association, meaning
neighboring units. A unit with a high value is that the data are distributed randomly across
relevant, but it is possible that it is not a statis- geographical units.
tically significant hotspot. To be a statistically Table 1 shows the results of global auto-
significant hotspot, a unit must have a high correlation tests. The Z Moran’s I and the Y

112
 Clusters of Violence in Mexico

and E(G*) Getis-Ord values are positive for from the data is to predict a result based on
the entire period, hence the null hypothesis a series of previous characteristics from a
(random spatial distribution of homicide series of training data sets where we observe
rates) is rejected, suggesting that there are the result and the test data set, to which the
some high values of homicide rates that are learning model is fitted to predict the results.
spatially concentrated. Random forests entail an ensemble of
To find out where these clusters are, we use training samples made up of different start-
the local Moran’s I and the local Getis-Ord tt* ing-points in the decision-tree samples that
tests. The results for the local Moran’s I show are later applied to the test set. Each tree
the randomization of the clusters and maps of classifies an “individual” uniquely. At the
atypical values, while the Getis-Ord tt* results same time, different trees can classify the
are represented as hotspots maps for homi- same individual in different classes. There-
cide rates illustrated in Figures 7, 8, and 9. fore, each tree represents a vote for an indi-
Considering the spatial contiguity and vidual in favor of belonging to a class. Once
discriminating by the levels and type of several trees are planted, the forest aggregates
autocorrelation, we established as a group- the votes and classifies each member in the
ing criterion that geographically connected most popular class, the one with the greatest
municipalities that are part of a hotspot in at number of votes (Breiman, 2001).
least one year—meaning, with a significant This technique also helps to find the most
Local tt*I (d)—are part of the same cluster. significant variables that are strongly related
We grouped all municipalities to obtain eigh- to the response variable in this case, the
teen clusters of violence resulting in the map homicide rate, thus adding interpretation
of Figure 10. Municipalities with non-signifi- value. Also, it helps to find a smaller number
cant high autocorrelation scores are grouped of variables that are sufficient to make a parsi-
by geographic contiguity and INEGI’s clas- monious prediction of the response variable
sification of the economic dynamics and (Genuer et al., 2010). In other words, in the
geographic proximity at a state cluster level. case of this work, we can validate our classi-
The constitution and diversity of the clus- fication theory by testing if it is possible to
ters is noteworthy. We were able to iden- predict the cluster to which each municipal-
tify a total of eighteen geographic regions ity is assigned. Hence, we are able to identify
called clusters of violence, characterized by a reduced number of significant variables
being similar within but different from each related to the spatial and temporal evolution
other. In the following stage of our study, we of homicide rates.
proceeded to validate these results. We applied the random forest algorithm
using the randomForest R library twice under
validation of the clusters of violence supervised mode, once with eighty-seven
Random forest is a learning statistical tech- variables and the second time with thir-
nique based on the aggregation of multiple ty-seven of the most important variables from
decision-tree results. The goal of learning the first time it was executed. Table 3 shows

113
 Journal of Latin American Geography

Figure 7. Getis-Ord hotspots of homicide rates by year, 2000 to 2005.

114
Clusters of Violence in Mexico

115
 Journal of Latin American Geography

Figure 8. Getis-Ord hotspots of homicide rates by year, 2006 to 2011.

116
Clusters of Violence in Mexico

117
 Journal of Latin American Geography

the comparison of parameter values and icant variables linked to homicide rates, we
summary results of the algorithm with eighty- found that, although each time we ran the
seven and thirty-seven variables, respectively. random forest the order of importance for
Classification error rates by class decrease as the variables changed because the selection
the number of trees grown increases for all of variables to build the tree is random, the
classes; also, for this case the error rates are same variables are identified consistently:
smaller and stabilize quicker after reducing centroids in each municipality, presence of
the number of predictor variables while grow- organized crime, intensity of migration in
ing the same number of trees, which results in 2010, population density, the social gap index,
a smaller overall error rate (OBB estimated and unemployed population in 2010.
error rate), as shown in Table 3.
Only for class or cluster 10 the error rate distribution of the clusters of violence
remains above 60 percent, but it stabilizes Our geostatistical analysis shows that homi-
after approximately 120 trees are built in cide rates are not homogeneous and do not
both cases, with eighty-six and thirty-seven have variation patterns that we can attribute
variables. Yet we must keep in mind that, entirely to chance. Following Tobler’s princi-
in general, classes with a small number of ple, we have found that it is possible to group
members show a higher systematic classifi- Mexican municipalities into eighteen clus-
cation error. The results of the random forest ter patterns that describe a behavior similar
allow us to validate the usefulness of our clus- within each cluster but different from behav-
ters to predict classes according to homicide ior of other clusters also considering other
rates in Mexican municipalities. socio-demographic variables.
According to the random forest with thir- In the twelve-year period covered in the
ty-seven variables, 271 municipalities should study, the most violent cluster was number 6,
be re-classified; however, almost half of them consisting of thirty-one municipalities in the
have no geographic adjacency, which is a very north of Chihuahua. This cluster had an aver-
important criterion for classification. There- age homicide rate of 49.22 per every 100,000
fore, after a new verification we re-classi- inhabitants. At the other end of the spectrum,
fied only 149 municipalities according to the the most peaceful cluster was number eigh-
random forest to our clusters proposal. The teen, consisting of 142 municipalities in the
map in Figure 10 already considers this reclas- states of Tabasco, Campeche, Yucatán, and
sification. Finally, we also identify atypical Quintana Roo, with an average rate of 5.51
municipalities listed in Table 4, which means homicides per 100,000 inhabitants.
that they are observations whose outlier We should point out that in 2010, even the
measure indicates that these observations do most peaceful cluster saw a jump in homi-
not fit well in the established classes. Hence, cide rates to 16.60, compared to 2.85 in 2001.
outliers are special situations that should be Between 2000 and 2010, violence increased in
observed in more detail. the entire country, even in the most peaceful
With respect to identifying the most signif- clusters.

118
 Clusters of Violence in Mexico

Figure 9. Getis-Ord hotspots of homicide rates, 2012.

Global Moran’s I Global Getis-Ord G*

Year Moran’s I E(I) Var(I) Z-score P-value G* E(G*) Var(G*) Z-score P-value
2000 0.1775 -0.0004 0.0001 14.585 0 0.0007 0.0004 0 13.828 0
2001 0.2139 -0.0004 0.0002 17.512 0 0.0008 0.0004 0 16.883 0
2002 0.1448 -0.0004 0.0001 12.440 0 0.0008 0.0004 0 11.988 0
2003 0.2126 -0.0004 0.0001 17.452 0 0.0008 0.0004 0 16.753 0
2004 0.1130 -0.0004 0.0001 9.709 0 0.0007 0.0004 0 9.656 0
2005 0.1724 -0.0004 0.0001 14.630 0 0.0008 0.0004 0 13.772 0
2006 0.2350 -0.0004 0.0001 19.583 0 0.0009 0.0004 0 18.733 0
2007 0.0880 -0.0004 0.0001 7.649 0 0.0007 0.0004 0 8.112 0
2008 0.2462 -0.0004 0.0001 20.455 0 0.0009 0.0004 0 19.711 0
2009 0.4012 -0.0004 0.0001 33.615 0 0.0013 0.0004 0 32.722 0
2010 0.3691 -0.0004 0.0001 32.104 0 0.0025 0.0004 0 31.579 0
2011 0.3162 -0.0004 0.0001 30.001 0 0.0015 0.0004 0 28.943 0
2012 0.2727 -0.0004 0.0001 22.855 0 0.0010 0.0004 0 21.765 0

Table 1. Spatial correlation tests for annual homicide rates.

Rates of violence escalated first in cluster 8, homicide rates in this group surged abruptly,
made up of municipalities in Tamaulipas, the reaching 878 percent over the previous year,
northeast of Coahuila, and the middle and then remained relatively stable before seeing
northern parts of Nuevo Leon. In 2001-2002, another exponential rise in 2009-2010, with

119
 Journal of Latin American Geography

Figure 10. Clusters of violence.

an increase of 671 percent over the previous rates remained steady but reached the high-
year. This is the cluster where homicide rates est levels throughout the first years of the war
have increased the most. on drugs, even increasing slightly. From 2001
Meanwhile, the most stable cluster was to 2012, they rose by 636.32 percent, but it
number 1, made up of municipalities in should be noted that from 2010 to 2011, they
Baja California, for which homicide rates declined by 41 percent. Still, this cluster was
remained at around 5.8 per 100,000 inhab- the third most violent in the country, with an
itants during the whole period. The lowest average rate of 27.28 homicides per 100,000
point was in 2005 at 4.56, and the highest inhabitants for the entire period analyzed
was two years earlier at 7.11. Unlike in cluster (2000–2012).
8, the peak in violence occurred before pres- From this analysis we can infer that
ident Felipe Calderon’s administration, even there is a negative contagion effect, where
though on average the rates were higher after regions with high homicide rates have an
2006. effect on close-by clusters so that homicides
Cluster 3, made up of municipalities in increase in them as well. Such is the case in
the northern part of Sonora, shows the same areas where borders of the following states
behavior, since the largest increase in homi- converge: Durango, Sinaloa, and Chihua-
cide rates occurred before 2006. Later, the hua; the Gulf coast of Veracruz and Tamau-

120
 Clusters of Violence in Mexico

Number of
Cluster States Included
Municipalities
1 Baja California 5
2 Baja California Norte 6
3 Northern Sonora 29
4 Mid-coast and central Sonora 25
Southwestern Sonora, southern Chihuahua, western and central Durango, Sinaloa, northern
5 98
Nayarit and western Zacatecas
6 Northern Chihuahua 31
7 Coahuila, eastern Durango, southern Nuevo Leon 64
8 Tamaulipas, northeastern Coahuila, central and northern Nuevo Leon 74
9 Western Nayarit, Jalisco, northern Colima, Aguascalientes, Guanajuato, and Queretaro 204
10 Eastern Nayarit, northern and Valles Jalisco, and southeastern Zacatecas 17
11 Northern Jalisco, Zacatecas, San Luis Potosi 110
12 Southern Tamaulipas, Veracruz and Puebla 394
Southern Colima, southeastern Jalisco, central and southern Michoacán, Guerrero (Tierra
13 Caliente, Great Coast, North, Central, and Acapulco), Mexico (Valle de Bravo, Tejupilco, 104
Ixtapan), southern Morelos
Northern Michoacán (Morelia and north of Zamora, Patzcuaro, and Zitacuaro clusters),
14 370
México (rest), México City, Morelos, Hidalgo, and Tlaxcala
Guerrero (Costa Chica and Mountains), Puebla (Atlixco Valley and Matamoros, and part of
15 Mixteca), and Oaxaca (Coast, Southern Mountains, Isthmus, Central Valley and southern 322
Mixteca)
16 Oaxaca (Can˜ada, Papaloapam and Northern Mountains), and southern Puebla 330
17 Chiapas 118
18 Tabasco, Campeche, Yucatan, and Quintana Roo 142

Table 2. Clusters.

lipas, and some municipalities in the state of Quintana Roo, encompassing the area from
Nuevo Leon; and the Pacific Coast, including Isla Mujeres to the border with Belize, as well
the states of Jalisco, Michoacan, Guerrero, as in municipalities on the border between
and part of Oaxaca. Tabasco, Chiapas, and Guatemala.
There are also atypical cases, where certain One possible explanation for these cases
municipalities stand out for being radically is their location on borders with other
different from their neighbors. Such is the countries, but we need more specific stud-
case of Loreto, in cluster 1, which suddenly ies of their characteristics, populations, and
reported much higher rates than the cluster economic activities to better understand the
average, even though in some years it shows variables that affect homicide rates in each
a rate of zero, which makes us think that specific context. Nevertheless, our proposal
there were specific events when these atyp- of eighteen clusters is a starting point to
ical homicides took place, or that popula- facilitate a differentiated understanding of
tion is so low that one case makes for a high violence, its causes, and its associated vari-
mortality rate. Another atypical case is in ables.
the southern part of the Caribbean Coast of Overall, for the entire country, we found

121
 Journal of Latin American Geography

Parameters of the model and

First run Second run
summary results
Number of variables 86 37
Number of tress 500 500
Number of variables in each cut 9 6
OOB estimated error rate 13.79% 11.32%
Processing time 1 minute <1 minute
Municipalities classified incorrectly 337 271

Table 3. Parameters of random trees.

Municipality State Measure

Isla Mujeres Quintana Roo 16.64022
Solidaridad Quintana Roo 15.16223
Balancan Tabasco 14.90781
Benito Juarez Quintana Roo 14.26400
Cozumel Quintana Roo 14.18144
Tenosique Tabasco 13.76766
Huimanguillo Tabasco 12.69184
Lazaro Cardenas Quintana Roo 12.27230
Cantamayec Yucatan 12.04884
La Libertad Chiapas 12.01606
Marques de Comillas Chiapas 11.63377
Emiliano Zapata Tabasco 11.46953
Othon P. Blanco Quintana Roo 11.41765
Akil Yucatan 11.12367
Reforma Chiapas 10.18863
Felipe Carrillo Puerto Quintana Roo 10.10572

Table 4. Atypical municipalities.

spatial patterns that confirm Tobler’s prin- are consistent across clusters, it is important
ciple. That is, that the level of violence in a to note that the relative impact of each factor
municipality is partly explained by the levels varies between clusters. While in some clus-
of violence of its neighbors. Furthermore, we ters unemployment emerged as the most
found that—even accounting for geographic significant factor, in others, poverty emerged
patterns—factors like population density, as the most powerful factor. A specific anal-
poverty, migration, unemployment, and the ysis of each cluster would be necessary if we
presence of organized criminal actors have a are to understand them, and if public-policy
significant influence on the levels of violence. tools to target priorities based on the local
Although the results mentioned before characteristics of violence are to be created.

122
 Clusters of Violence in Mexico

Overall, as social gaps lessen, employment The findings urge us to consider doing
increases, and organized crime is fought, it more studies that will allow us to measure
is possible to expect a reduction in levels of the explanatory levels of each of the variables
violence. It will also be necessary to design used to build the clusters and thus to prior-
policies that reverse population concen- itize those variables that might have more
tration and preemptively address issues weight econometrically. In this way, public
in regions close to violent hotspots in an policies for the prevention of violence could
attempt to prevent contagion. These would be created based on a territorial perspec-
appear to be the most adequate measures, tive of the distribution of the phenomenon.
according to what we have learned from our Although, there are some atypical cases that
analysis of events between 2000 and 2012. challenge our results (i.e. Tenosique, Isla
Mujeres, Felipe Carrillo Puerto, etc.), these
cases should not be dismissed as unexplain-
Conclusions able anomalies but should be incorporated in
In our purpose to understand the space-time other in-depth qualitative studies seeking to
patterns of violence and variables associ- understand and complement the local differ-
ated with homicides to advocate for targeted ences that could condition our finding.
prevention and combat policies, we have By knowing the spatial and temporal
offered an analysis of the vicinities of homi- distribution of homicide rates, we can iden-
cide rates in Mexico. We have built a classifica- tify the most violent areas as well as the most
tion of clusters in which we identify eighteen stable, peaceful, and changeable ones. Our
zones with characteristics that vary among cluster proposal is a tool for creating public
each other but are similar within. We have policy that allows for a more efficient diagno-
called this classification clusters of violence. sis in the different Mexican municipalities--a
Clusters of violence allow us to consider tool that should be used with others to move
homicide as a phenomenon that varies in beyond the simplistic outlook calling for
space and time, and that is associated with more iron-fist policies and harsher punish-
the location of municipalities and with other ments to dissuade criminals (Coimbra &
socioeconomic variables distributed accord- Briones, 2019).
ing to Tobler’s principle (municipalities that Reducing violence is a priority for Mexico;
are the closest geographically are the most there has been much discussion of its causes
similar to each other). In summary, we can and many proposed solutions, and we hope
differentially understand a complex phenom- to influence this discussion by highlighting
enon that is extremely important in Mexico the usefulness of currently available spatial
as well as see how congruent the establish- analysis models, which allow us to somehow
ment of these zones is with variables that reduce the complexity of the phenomena
weigh on the phenomenon of death, such as and to carry out empirical analyses that put
the intensity of migration, age, the presence several different variables simultaneously
of organized crime, and demographic density. into play.

123
 Journal of Latin American Geography

The random forest model, Moran’s I Index, collaboratively can become more effective.
and Getis-Ord tests are effective tools to Moreover, by defining the most significant
classify and predict the spatial and temporal variables in the classification, we can target
behavior of homicide rates and violence. A prevention measures precisely to the vari-
multidimensional classification and variable ables that may have the greatest impact, thus
selection model allows us to conduct a better contributing to the design of better preven-
time-space analysis of homicides and it is a tion and combat programs.
useful tool for designing public policy. In our In terms of comparative public policy, this
case, by classifying all the municipalities of article also opens an entirely new line of
Mexico into eighteen different clusters we observations about Latin America regard-
can identify local problems; plan regional ing social and cultural comparability and the
strategies; and target programs, follow-up, behavior of organized crime. We can make
and evaluation. comparisons using these methodological
Understanding that violence does not tools and then contrast the results in differ-
end at the borders within municipalities or ent contexts to develop common dialogues
between states leads us to a different perspec- for the region.
tive, one where plans created and applied

References
Andrade, L. T. de, & Diniz, A. M. A. (2013). A reorganização espacial dos homicídios no
Brasil e a tese da interiorização. Revista Brasileira de Estudos de População, 30 (Supl.), 171-
191. https://doi.org/10.1590/S0102-30982013000400011

Arteaga-Botello, N., Dávila-Cervantes, C. A., & Pardo-Montaño, A. M. (2019). Necro-spaces

and Violent Homicides in Mexico. International Journal of Conflict and Violence (IJCV), 13,
a660-a660. https://doi.org/10.4119/ijcv-3125

Atuesta, L. H., & Ponce, A. F. (2017). Meet the Narco: increased competition among
criminal organisations and the explosion of violence in Mexico. Global Crime, 18(4), 375–
402. https://doi.org/10.1080/17440572.2017.1354520

Atuesta, L. H., Siordia, O. S., & Lajous, A. M. (2019). The “war on drugs” in
Mexico:(official) database of events between December 2006 and November 2011. Journal
of conflict resolution, 63(7), 1765-1789. https://doi.org/10.1177/0022002718817093

Bailey, J., & Taylor, M. M (2009). Evade, corrupt, or confront? Organized crime and the
state in Brazil and Mexico. Journal of Politics in Latin America, 1(2), 3–29. https://doi.
org/10.1177/1866802X0900100201

124
 Clusters of Violence in Mexico

Baller, R. D., Anselin, L., Messner, S. F., & Deane, G. (2001). Structural covariates of US
county homicide rates: Incorporating spatial effects. Criminology, 39(3), 561–90. https://
doi.org/10.1111/j.1745-9125.2001.tb00933.x

Botello, H. A. (2014). Inequidad, violencia y pobreza en los municipios de Colombia. Perfil

de Coyuntura Económica, (24), 33–48. DOI: 10.17533/udea.pece.n24a02

Boyce, G., Banister, J., & Slack, J. (2015). You and what army? Violence, the state, and
Mexico’s war on drugs. Territory, Politics, Governance, 3(4), 446–68. https://doi.org/10.108
0/21622671.2015.1058723

Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32. https://doi.

org/10.1023/A:1010933404324

Briceño-León, R. (2012). La Comprensión de los Homicidios en América Latina: ¿Pobreza

o Institucionalidad? Ciência & Saúde Coletiva, 17(12), 3159–3170. https://www.scielosp.
org/article/csc/2012.v17n12/3159-3170/es/

Briscoe, I., & Kalkman, P. (2016). The new criminal powers: The spread of illicit links to
politics across the world and how it can be tackled. Netherlands Institute of International
Relations, Clingendael. https://www.clingendael.org/sites/default/files/pdfs/the_new_
criminal_powers.pdf

Burraston, B., Watts, S. J., McCutcheon, J. C., & Province, K. (2019). Relative
Deprivation, Absolute Deprivation, and Homicide: Testing an Interaction Between
Income Inequality and Disadvantage. Homicide Studies, 23(1), 3–19. https://doi.
org/10.1177/1088767918782938

Camara, M., & Salama, P. (2004). Homicidios en América del sur: ¿los pobres son peligrosos?
Revista de Economía Institucional, 6(10), 159–81. https://www.redalyc.org/articulo.
oa?id=41901007

Calderón, G., Robles, G., Díaz-Cayeros, A., & Magaloni, B. (2015). The Beheading of
Criminal Organizations and the Dynamics of Violence in Mexico. Journal of Conflict
Resolution, 59(8), 1455–1485. https://doi.org/10.1177/0022002715587053

Caragliu, A., & Nijkamp, P. (2013). From islands to hubs of innovation. In H. Lawton Smith
& U. Hilpert (Eds.), Networking Regionalized Innovative Labour Markets (pp. 119-135).
Routledge.

125
 Journal of Latin American Geography

Castillo, J. C., Mejía, D., & Restrepo, P. (2020). Scarcity without leviathan: The violent
effects of cocaine supply shortages in the mexican drug war. Review of Economics and
Statistics, 102(2), 269-286. https://doi.org/10.1162/rest_a_00801

Casillas, R. (2008). Las rutas de los centroamericanos por México, un ejercicio de

caracterización actores principales y complejidades. Migración y Desarrollo, (10), 157–174.
http://www.scielo.org.mx/pdf/myd/n10/n10a7.pdf

Ceobanu, A. M., Wood, C. H., & Ribeiro, L. (2011). Crime victimization and public support
for democracy: Evidence from Latin America. International Journal of Public Opinion
Research, 23(1), 56–78. https://doi.org/10.1093/ijpor/edq040

Cohen, J., & Tita, G. (1999). Diffusion in homicide: Exploring a general method for
detecting spatial diffusion processes. Journal of Quantitative Criminology, 15(4), 451–493.
https://doi.org/10.1023/A:1007596225550

Coimbra, L. O., & Briones, Á. (2019). Crimen y castigo. Una reflexión desde América
Latina. URVIO. Revista Latinoamericana De Estudios De Seguridad, (24), 26–41. https://
doi.org/10.17141/urvio.24.2019.3779

CONAPO. (2010). Proyecciones de la Población de México y de las Entidades Federativas,

2016-2050 y Conciliación Demográfica de México, 1950 -2015. https://www.gob.mx/conapo/
acciones-y-programas/conciliacion-demografica-de-mexico-1950-2015-y-proyecciones-
de-la-poblacion-de-mexico-y-de-las-entidades-federativas-2016-2050

CONEVAL. (2010). Consejo Nacional de Evaluación de la Políica de Desarrollo Social 2015.

Evolución de las dimensiones de la pobrez 1990 – 2014. https://www.coneval.org.mx/
Medicion/EDP/Paginas/Evolucion-de-las-dimensiones-de-la-pobreza-1990-2014-.aspx

Dávila-Cervantes, C. A., & Pardo-Montaño, A. M. (2018). Violence in Colombia and

Mexico: trend and impact on life expectancy of homicide mortality between 1998 and
2015. Public Health, 163, 1–8. https://doi.org/10.1016/j.puhe.2018.06.001

Dávila Cervantes, C. A., & Pardo Montaño, A. M. (2016). Análisis de la tendencia e impacto
de la mortalidad por causas externas: México, 2000–2013. Salud colectiva, 12(2), 251–
264002E. https://doi.org/10.18294/sc.2016.743

Dell, M. (2015). Trafficking network and the Mexican drug war. The American Economic
Review, 105(6), 1738–1779. DOI: 10.1257/aer.20121637

126
 Clusters of Violence in Mexico

Duque, J. C., Aldstadt, J., Velasquez, E., Franco, J. L., & Betancourt, A. (2011). A
computationally efficient method for delineating irregularly shaped spatial clusters.
Journal of Geographical Systems, 13(4), 355-372. DOI 10.1007/s10109-010-0137-1

Durán-Martínez, A. (2015). To kill and tell? State power, criminal competition,

and drug violence. Journal of Conflict Resolution, 59(8), 1377-1402. https://doi.
org/10.1177/0022002715587047

Enciso, F. (2010). Los fracasos del chantaje. Régimen de prohibición de drogas y

narcotráfico. In A. Alvarado & M. Serrano (Eds.), Los grandes problemas de México:
Seguridad Nacional y Seguridad Interior (pp. 61–104). El Colegio de México.

Escalante, F. (2010). Panorama del homicidio en México. Esquema de análisis territorial

1990-2007. In A. Alvarado & M. Serrano (Coord.), Los grandes problemas de México.
Capítulo XV: Seguridad nacional y seguridad interior (pp. 301-330). El Colegio de México.

Espinal-Enríquez, J., & Larralde, H. (2015). Analysis of Mexico’s narco–war network (2007–
2011). PLoS ONE, 10(5), 1–15. https://doi.org/10.1371/journal.pone.0126503

Espinosa, V., & Rubin, D. (2015). Did the military interventions in the Mexican drug war
increase violence? The American Statistician, 69(1), 17–27. https://doi.org/10.1080/000313
05.2014.965796

Estévez, A. (2017). La violencia contra las mujeres y la crisis de derechos humanos: de la

narcoguerra a las guerras necropolíticas. Revista interdisciplinaria de estudios de género de El
Colegio de México, 3(6), 69–100. https://dx.doi.org/10.24201/eg.v3i6.142

Fajnzylber, P., Lederman, D., & Loayza, N. (2001). Crimen y Violencia en América Latina.
World Bank. http://documents.worldbank.org/curated/en/522791468749792478/
Crimen-y-Violencia-en-America-Latina.

Fajnzylber, P., Lederman, D., & Loayza, N. (2002). Inequality and violent crime. The
journal of Law and Economics, 45(1), 1-39. https://www.journals.uchicago.edu/doi/
abs/10.1086/338347

Fischer, M., Scherngell, T., & Jansenberger, E. (2006). The Geography of Knowledge
Spillovers Between High-Technology Firms in Europe: Evidence from a Spatial
Interaction Modeling Perspective. Geographical Analysis, 38(3), 288–309. https://doi.
org/10.1111/j.1538-4632.2006.00687.x

127
 Journal of Latin American Geography

Flint, C., Harrower, M., & Edsall, R. (2000). But How Does Place Matter? Using
Bayesian Networks to Explore a Structural Definition of Place. In proceedings of New
Methodologies for the Social Sciences: The Development and Application of Spatial Analysis
for Political Methodology. Pennsylvania State University. https://www.geovista.psu.edu/
publications/others/FHE/fhepaper.htm

Flores, M., & Rodríguez-Oreggia, E. (2014). Spillover effect on homicides across Mexican
municipalities: A spatial regime model approach. The Review of Regional Studies, 44(3),
241–262. https://rrs.scholasticahq.com/article/8072-spillover-effects-on-homicides-

Folino, J. O. (2015). Predictive efficacy of violence risk assessment instruments in Latin-

America. The European Journal of Psychology Applied to Legal Context, 7(2), 51–58. https://
doi.org/10.1016/j.ejpal.2014.11.006

Gamlin, J. (2015). Violence and homicide in Mexico: a global health issue. The Lancet,
385(9968), 605–606. https://www.thelancet.com/pdfs/journals/lancet/PIIS0140-
6736(15), 60234-3.pdf

García-Ponce, O., Dube, A., & Dube, O. (2013). Cross-Border Spillover: U.S. Gun Laws and
Violence in Mexico. American Political Science Review, 107(3), 397–417. https://www2.
nber.org/conferences/2011/SI2011/CRI/Dube_Dube_Garcia_Ponce.pdf

Genuer, R., Poggi, J. M., & Tuleau-Malot, C. (2010). Variable selection using random forests.
Pattern Recognition Letters, 31(14), 2225–2236. https://doi.org/10.1016/j.patrec.2010.03.014

Getis, A., & Ord, J. (1992). The analysis of spatial association by use of distance statistics.
Geographical Analysis, 24(3), 189–206. https://doi.org/10.1111/j.1538-4632.1992.tb00261.x

Guerrero-Gutiérrez, E. (2011). Security, drugs, and violence in Mexico: A survey. 7th North
American Forum, Washington DC. Lantia Consultores, S.C.

Goodchild, M. F. (1987). A spatial analytical perspective on geographical information

systems. International Journal of Geographical Information Systems, 1(4), 327–334. https://
doi.org/10.1080/02693798708927820

Goodchild, M. F. (2004). GIScience, geography, form, and process. Annals of

the Association of American Geographers, 94(4), 709-714. Doi:10.1111/j.1467-
8306.2004.00424.x

128
 Clusters of Violence in Mexico

Haggett, P. (2000). The geographical structure of epidemics. Oxford University.

Hopkins, M., Tilley, N., & Gibson, K. (2013). Homicide and Organized Crime in England.
Homicide Studies, 17(3), 291–313. https://doi.org/10.1177/1088767912461786

Instituto Nacional de Estadística y Geografía [INEGI]. (2020a). Sistema Nacional de

Estadísticas Vitales. Estadísticas de Mortalidad. https://www.inegi.org.mx/temas/
mortalidad/

Instituto Nacional de Estadística y Geografía [INEGI]. (2020b). CENSO GENERAL DE

POBLACIÓN Y VIVIENDA 2000. https://www.inegi.org.mx/programas/ccpv/2000/
default.html?init=2

Instituto Nacional de Estadística y Geografía [INEGI]. (2020c). II CONTEO DE

POBLACIÓN Y VIVIENDA DEL AÑO 2005. https://www.inegi.org.mx/programas/
ccpv/2000/default.html?init=2

Instituto Nacional de Estadística y Geografía [INEGI]. (2020d). CENSO GENERAL DE

POBLACIÓN Y VIVIENDA 2010. https://www.inegi.org.mx/programas/ccpv/2000/
default.html?init=2

Ingram, M. C., & Marchesini da Costa, M. (2017). A Spatial Analysis of Homicide

Across Brazil’s Municipalities. Homicide Studies, 21(2), 87–110. https://doi.
org/10.1177/1088767916666603

Land, K. C., McCall, P. L., & Cohen, L. E. (1990). Structural Covariates of Homicide Rates:
Are There Any Invariances Across Time and Social Space? American Journal of Sociology,
95(4), 922–963. https://doi.org/10.1086/229381

Liu, Y., Fullerton, T. M., & Ashby, N. J. (2012). Assessing the Impacts of Labor Market and
Deterrence Variables on Crime Rates in México. Contemporary Economic Policy, 31(4),
669-690. https://doi.org/10.1111/j.1465-7287.2012.00339.x

Li, W., Goodchild, F., & Church, R. (2013). An efficient measure of compactness for two-
dimensional shapes and its application in regionalization problems. International Journal
Inof Geographical Information Science, 27(6), 1227–1250. https://doi.org/10.1080/13658816
.2012.752093

129
 Journal of Latin American Geography

Loftin, C., McDowall, D., Curtis, K., & Fetzer, M. D. (2015). The Accuracy of Supplementary
Homicide Report Rates for Large U.S. Cities. Homicide Studies, 19(1), 6–27. https://doi.
org/10.1177/1088767914551984

López Hernández, F. A., & Palacios Sánchez, M. A. (2000). Distintos modelos de

dependencia espacial. Análisis de autocorrelación. In Anales de Economía Aplicada. XIV
Reunión ASEPELT. Universidad de Oviedo. España.

Márquez, F., Pérez, F., & Rodríguez, C. (2015). The (Non-) Effect of Violence on Education
Evidence from the “War on Drugs” in Mexico. RAND Labor and Population.

Martinez-Cruz, A. L., & Rodríguez-Castelán, C. (2016). Crime and persistent punishment:

a long-run perspective on the links between violence and chronic poverty in Mexico. Policy
Research Working Paper no. WPS 7706. World Bank Group.

Merino, J. (2011). Los operativos conjuntos y la tasa de homicidios: Una medición. Nexos.
http://www.nexos.com.mx/?p=14319

Molzahn, C., Ríos, V., & Shirk, D. A. (2012). Drug violence in Mexico: Data and analysis
through 2011. Trans-Border Institute, University of San Diego.

Mora Bayo, M. (2014). A criminalização da pobreza e os efeitos do estado de segurança

neoliberal: reflexões a partir de Montanha, Guerrero. Revista De Estudos E Pesquisas
Sobre As Américas, 7(2), 174–208. https://periodicos.unb.br/index.php/repam/article/
view/20226

Muggah, R. (2017). El auge de la seguridad ciudadana en América Latina y el Caribe.

International Development Policy| Revue internationale de politique de développement, 9(9).
https://doi.org/10.4000/poldev.2512

Nygren, A. (2018). Inequality and interconnectivity: Urban spaces of justice in Mexico.

Geoforum, 89, 145-154. http://dx.doi.org/10.1016/j.geoforum.2017.06.015

Osorio, J. (2015). The Contagion of Drug Violence: Spatiotemporal Dynamics of the

Mexican War on Drugs. Journal of Conflict Resolution, 59(8), 1403–1432. https://doi.
org/10.1177/0022002715587048

130
 Clusters of Violence in Mexico

Pan, M., Widner, B., & Enomoto, C. E. (2012). Spillover Effects of Crimes in Neighboring
States of Mexico. International Journal of Business and Social Science, 3(14), 13–32. https://
corescholar.libraries.wright.edu/econ/248

Ponce, A. F, (2019). Violence and electoral competition: criminal organizations and

municipal candidates in Mexico. Trends in Organized Crime, 22(2), 231–254. https://doi.
org/10.1007/s12117-018-9344-9

Ponce, A. F., Velarde, R. V. L., & Santamaría, J. S. (2019). Do local elections increase
violence? Electoral cycles and organized crime in Mexico. Trends in Organized Crime.
https://doi.org/10.1007/s12117-019-09373-8

Rettberg, A. (2020). Violencia en América Latina hoy: manifestaciones e impactos. Revista

de Estudios Sociales, (73), 2–17. https://doi.org/10.7440/res73.2020.01

Ríos, V. (2013). Why did Mexico become so violent? A self-reinforcing violent equilibrium
caused by competition and enforcement. Trends in organized crime, 16(2), 138–155. DOI
10.1007/s12117-012-9175-z

Sánchez-Jabba, A. M., Castelblanco, O., Díaz, A. M., Tautiva, J., Ángel, L., Peláez, A., &
Sánchez-Jabba, A. (2012). Evolución geográfica del homicidio en Colombia. Documentos
de Trabajo Sobre Economía Regional y Urban. Institutional Repository of the Banco
de la República. Working paper No. 169. https://repositorio.banrep.gov.co/
handle/20.500.12134/3082

Sánchez Munguía, V. (2009). La actual lucha del gobierno mexicano contra la delincuencia
en la frontera con Estados Unidos. Frontera Norte, 23(45), 97–130. http://www.scielo.org.
mx/scielo.php?script=sci_arttext&pid=S0187-73722011000100004

Sanchez, R. M. (2006). Insecurity and violence as a new power relation in Latin America.
ANNALS-American Academy of Political and Social Science, 606, 178 -198. http://www.jstor.
org/stable/25097823 .

Sanjurjo, D. (2016). La influencia de la posesión civil de armas de fuego en las tasas de homicidio
de América Latina y el Caribe. Documento de trabajo 1/2016. Real Instituto Elcano.
https://www.almendron.com/tribuna/la-influencia-de-la-posesion-civil-de-armas-de-
fuego-en-las-tasas-de-homicidio-de-america-latina-y-el-caribe/

131
 Journal of Latin American Geography

Secretaría de la Defensa Nacional [SEDENA]. (2010). “Solicitud de Información Pública

0000700061918”. Número de Hectáreas Destruidas y Erradicadas por el Ejército Nacional y la
Marina Nacional de Amapola y Mariguana en el periodo 2000- 2017.

Shirk, D., & Wallman, J. (2015). Understanding Mexico’s drug violence. Journal of Conflict
Resolution, 59(8), 1348–1376. https://doi.org/10.1177/0022002715587049

Spinelli, H., Zunino, G., Alazraqui, M., Guevel, C., & Darraidou, V. (2010). Mortalidad
por armas de fuego en Argentina: 1990-2008. Representación OPS/OMS Argentina; 68.
https://iris.paho.org/bitstream/handle/10665.2/5526/2010-ARG-mortalidad-armas-
fuego.pdf?sequence=1&isAllowed=y

Snyder, R., & Durán-Martínez, A. (2009). Drugs, Violence, and State-Sponsored Protection
Rackets in Mexico and Colombia. Colombia Internacional, (70), pp 61–91. https://doi.
org/10.7440/colombiaint70.2009.03

Trejo, G., & Ley, S. (2020). Votes, Drugs, and Violence: The Political Logic of Criminal Wars in
Mexico. Cambridge University.

Tobler, W. R. (1970). A Computer Movie Simulating Urban Growth in the Detroit Region.
Economic Geography, 46(2), 234–240. https://doi.org/10.2307/143141

United Nations Office on Drugs and Crime [UNODC]. (2019). Global Study On Homicide.
Homicide: extent, patterns, trends and criminal justice response. United Nations Office on
Drugs and Crime. https://www.unodc.org/documents/data-and-analysis/gsh/Booklet2.
pdf

Valdés, G. (2013). Historia del narcotráfico en México. Aguilar.

Valdivia, M., & Castro, R. (2013). Gender bias in the convergence dynamics of the regional
homicide rates in Mexico. Applied Geography, 45, 280–291. https://doi.org/10.1016/j.
apgeog.2013.09.015

Westlund, H. (2013). A brief history of time, space, and growth: Waldo Tobler’s first law of
geography revisited. The Annals of Regional Science, 51(3), 917-924. DOI:10.1007/s00168-
013-0571-3

132
 Clusters of Violence in Mexico

Zeoli, A. M., Pizarro, J. M., Grady, S. C., & Melde, C. (2014). Homicide as infectious disease:
Using public health methods to investigate the diffusion of homicide. Justice quarterly,
31(3), 609–632. https://doi.org/10.1080/07418825.2012.732100

Zunino, M. G., Diez Roux, A. V., & Souza, E. R. D. (2012). Homicidios por armas de fuego
en Argentina, 1991-2006: un análisis de niveles múltiples. Ciência & Saúde Coletiva, 17,
3219-3232. https://www.scielosp.org/article/csc/2012.v17n12/3219-3232/es/

133