Eight years of homicide evolution in Monterrey, Mexico: a network approach
Rodrigo Dorantes-Gilardi, Diana García-Cortés, Hiram Hernández-Ramos, Jesús Espinal-Enriquez
EE IGHT YEARS OF HOMICIDE EVOLUTION IN M ONTERREY ,M EXICO : A NETWORK APPROACH
A P
REPRINT
Rodrigo Dorantes-Gilardi
Centro de Estudios InternacionalesEl Colegio de MéxicoMexico City [email protected]
Diana García Cortés
Computational Genomics DepartmentNational Institute of Genomic MedicineMexico City [email protected]
Hiram Hernández Ramos
Facultad de CienciasUniversidad Nacional Autónoma de MéxicoMexico City [email protected]
Jesús Espinal-Enriquez
Computational Genomics DepartmentNational Institute of Genomic MedicineMexico City [email protected]
September 11, 2020 A BSTRACT
Homicide is without doubt one of Mexico’s most important security problems, with data showingthat this dismal kind of violence sky-rocketed shortly after the war on drugs was declared in 2007.Since then, violent war-like zones have appeared and disappeared throughout Mexico, causingunfathomable human, social and economic losses. One of the most emblematic of these zones is thecity of Monterrey, a central scenario in the narco-war. Being an important metropolitan area in Mexicoand a business hub, Monterrey has counted hundreds to thousands of casualties. In spite of severalapproaches being developed to understand and analyze crime in general, and homicide in particular,the lack of accurate spatio-temporal homicide data results in incomplete descriptions. To betterunderstand the underlying mechanisms by which violence has evolved and spread through the city,here we propose a network-based approach. For this purpose, we define a homicide network wherenodes are geographical entities that are connected through spatial proximity and crime similarity. Datais taken from a crime database spanning 86 months in the Monterrey metropolitan area, containingmanually curated geo-located and dated homicides, as well as from Open Street Map for urbanenvironment. Under this approach, we first identify independent crime sectors corresponding todifferent connected components. Each of these clusters of crime presents crime evolution similar tothe one at state and national levels. We then show how crime spread from neighborhood to adjacentneighborhoods when violence was mainly cartel-related and how it was chiefly static at a differenttime. Finally, we show a relation between homicidal crime and urban landscape by studying thedistance of safe and violent neighborhoods to the closest highway and by studying the evolution ofhighway and crime distance over the cartel-related years and the following period. With this approach,we are able to describe more accurately the evolution of homicidal crime in a metropolitan area.
Keywords crime networks · homicide dynamics · spatial networks · Mexico’s drug war
Introduction
Violence in Mexico during the drug war
In an attempt to legitimate his questionable victory on the 2006 presidential elections, former Mexican president,Felipe Calderon Hinojosa (FCH), launched an intervention of the Mexican Army and Federal Police into the State of a r X i v : . [ phy s i c s . s o c - ph ] S e p rXiv Template
A P
REPRINT
Michoacán, to face the drug cartels that operated in that region [1]. “Operación Michoacán” was the starting point ofFCH’s Drug War [2]. This battle between Mexican security forces and drug cartels spread the violence throughout thewhole country and homicide rates scaled to levels never seen before in Mexico [3].The dramatic increase of drug-related homicides during FCH’s drug war, brought a wave of violence as a collateraleffect to the aforementioned war. During the administration of the next president, Enrique Peña Nieto (EPN), there wasan apparent period of lower homicide rates. That period corresponded to his first three years of government. However,at the beginning of 2015 a renewed wave of violence emerged at a national level and has not decreased since [4].FCH’s drug war also generated battles between drug cartels to control broad territories. In the year of 2010, there was aschism between the Gulf Cartel, one of the most important criminal groups in Mexico, and the “Zetas” band, the formerarmed force of the Gulf Cartel. This struggle took place in the northern states of Tamaulipas and Nuevo León. Theviolence triggered by this separation caused the highest homicide rates in the history of several cities in the region suchas Monterrey [3], and, more generally, in the Monterrey Metropolitan Area (MMA), the second urban area in Mexico,and an important industrial hub.The rise of violence during the second part of EPN’s administration, was not only due to the drug war, it has beendemonstrated that drug-band related homicides have a strong influence in general homicides [5]. The increasinghomicides rates have thus permeated in Nuevo León and the MMA.
Geo-social MMA Background
MMA is composed by 19 municipalities, namely, Monterrey, Guadalupe, San Nicolás de los Garza, Apodaca, San PedroGarza García, Santa Catarina, General Escobedo, Juárez, Hidalgo, Santiago, Cadereyta Jiménez, García, Salinas Victoria,Pesquería, Ciénega de Flores, General Zuazua, Marín, Carmen, y Doctor González. The number of neighborhoodswithin these municipalities comes to , , where more than 4.5 Million people live (Figure 1).Neighborhoods can tile up a city in the way that each tile is a village of its own. Studies about cities in the United Stateshave shown that crime is highly concentrated in some neighborhoods within cities [6]. However, a typical urban tilingcomprises hotspot neighborhoods bordering low crime-rate ones [7]. It is not uncommon in Latin America to haveextreme cases of slums next to luxurious neighborhoods separated typically by urban infrastructure. MMA is not theexception; while San Pedro Garza García is the second municipality in terms of human development index (HDI=0.87in 2015 [8]), the municipality of Santa Catarina (Northern border of San Pedro Garza García) has neighborhoods withimportant levels of poverty. Moreover, neighborhoods have been shown to be interdependent in terms of what happensin one affects the other [9]. In particular, crime shows diffusion processes across neighborhoods [10, 11]. Approaches to understand violence
There have been many attempts to describe, understand, predict or control the dynamics and spread of conflict andgang-related violence: from literature-based approaches [3], data-mining-based network inference [12, 13], reaction-diffusion equations [11], to combined methods [14, 15]. Understanding temporal and spatial evolution of homicides ina metropolitan area is of utmost importance to alleviate and diminish said violence.A crucial step in the analysis and development of accurate models of homicide dynamics is data collection, in particular,availability of geolocated data and precise dating of events is required. In this sense, works such as the one developed byOliveira [16] have collected crime data of robbery and burglaries. Other approaches have used homicide data, however,the level of granularity for these data is in the best cases by municipality [3, 17], or by country [18]. In both cases: lackof precision in geolocated data and large time-steps limit the description of a complex phenomenon such as the spreadof crime-related homicides throughout a metropolitan area.Newspaper
El Norte has documented the homicide violence in the MMA from the year of 2011 until February of 2018.Its database reports daily homicides with number of casualties for each event and the associated longitude and latitudecoordinates. To our knowledge, this is the most accurate and comprehensive geo-located homicide database for anyplace in Mexico.
Outline
In this work, we used geo-located homicide data to analyze the structure and dynamics of homicide-related violencein the MMA. Data was taken from
El Norte newspaper daily-updated database (ENDB). We developed a spatial andtemporal analysis of the homicides in the MMA. 2 rXiv
Template
A P
REPRINT
Figure 1:
Political division of MMA.
This map contains the names of municipalities in the MMA and their locationinside the state of Nuevo León. The location of Nuevo León within Mexico is also displayed.We first analyzed the time series of homicides at municipality (city), locality, and neighborhood level. By means ofneighborhood-derived analysis, we assembled a network where neighborhoods are connected if they are adjacent and ifthere was at least one homicide during a given time window. Said networks were intersected by periods.Second, to detect whether or not violence was correlated in a temporal fashion, neighborhood networks were build iftwo places had at least one event during the same week, or if events were separated by one week. Finally, we correlatedthe places in which homicides took place and Open Street Maps (OSM) locations. We observed the most frequentplaces close to homicides, to assess whether or not a tendency appeared.The approaches used here allow us to answer different questions related to the dynamics and structure of homicidalviolence: Is violence in MMA spreading through time? Is violence spreading through space? Do municipalities in theMMA present a concerted pattern of violence or is there a correlation among them? Is violence bursting simultaneouslyor a time-window appears between violence in different places? Are there features of the urban environment related tothe location of homicides?.The analysis of close neighborhoods showed that during 6-month periods there is only links between MonterreyDowntown Neighborhood and its closest neighbors. The within-week networks and 1-week-separated correlationnetworks showed that the most correlated places are geographically apart, contrary to the already known fact thatphysical proximity determines the spread of violence. Finally, the OSM data shows that homicides correlate with3 rXiv
Template
A P
REPRINT locations close to highways and freeways, but more importantly, main roads in MMA segregate homicidal neighborhoodsfrom the non-violent ones. Additionally, the political division of some municipalities, and the roads appear to constrainthe areas in which homicides occur.With this approach, we conclude that the dynamics of homicides in MMA is not similar to other places in which dataexist. We argue that the evolution of those dynamics were strongly influenced by the FCH’s drug war. Here we have anaccurate description of the structure and dynamics of homicide crime in a large metropolitan area.
Results
Figure 2:
Timeseries of homicides in Mexico, Nuevo León State, and the Monterrey Metropolitan Area (MMA).
In this plot, the monthly number of homicides for those three levels of granularity are described. It can be observed atthe beginning of the FCH’s presidency, the substantial increase in homicide rate in the three levels. The second increasein homicides in Mexico did not follow the same trend in Nuevo León nor MMA, however, in both places violenceincreased.Figure 2 shows the monthly official data of homicides in Mexico, the Mexican State of Nuevo León, and the area givenby the Monterrey Metropolitan Area (MMA) from 1990 until 2019. As it can be observed there is a steady increase inthe number of homicides starting in 2007 until 2012 at national level, and a second one from 2016 to 2018. The year ofthe first increase in homicides (2007) coincides with the start of the war on drugs started by president Felipe Calderón.In the case of Nuevo León, the first steady increase in homicides happens almost three years later, at the end of 2009and reaches almost 10% of all homicides in Mexico in 2011, the peak of the narco war in Monterrey. The permeation ofthe drug war reached MMA at the end of 2009 and directed the fluctuation of homicides at a State level, as it can beseen in Figure 2. We also observe a second smaller increase in Nuevo León after 2014, posterior to a steady decrease inhomicides during the period 2012–2014.Figure 3 shows the casualties by municipality of the MMA in the form of a lollipop plot. Each point represents anaggregation of the monthly casualties per one-thousand inhabitants in each municipality. There is no homogeneity inhomicide rates across municipalities, revealing crime focal points located within some municipalities during this period.This is mainly the case of Apodaca, Monterrey, Cadereyta Jiménez, San Nicolás de los Garza and Guadalupe. Thedistribution of homicide events is denser around the 2011–2012 period, decreasing around 2014 and increasing againafter 2016, coinciding with the homicide trend at a State level from Figure 2.4 rXiv
Template
A P
REPRINT
Figure 3:
Timeseries of homicides by week in the cities of MMA . This lollipop plot shows the number of casualtiesfor each municipality of MMA during the period under study as well as the population of each municipality (next toname). The upper right panel shows the cumulative number of homicides in the “El Norte Data Base”. As it can beobserved, the trend of ENDB data coincides with MMA homicides reported by INEGI in Figure 2.Of particular interest is the case of Santa Catarina municipality: Despite it not being one of the top places in terms ofnumber of casualties in MMA, it consistently presents homicides during the whole period, even during the last monthsof 2017 and beginning of 2018.The homicide time-series were used to correlate violence across municipalities across the 86-month period on a weeklybasis (Figure 4). The highest correlated city-pair is Monterrey and San Nicolás de los Garza, meanwhile García and5 rXiv
Template
A P
REPRINT
Figure 4:
Correlation of homicides in the cities of MMA . This heatmap represents the Pearson correlation coefficientof the homicide time series of the 19 municipalities that compose the MMA. Pink squares show positive correlation,meanwhile negative correlations are depicted in blue.Pesquería are the highest anti-correlated municipalities. We assigned a significance score via Z-scores to the correlationvalues by means of a null-model built from reshuffled iterations of the casualties by municipality. Significantly correlated(Z-score > ) municipality pairs are displayed in Table 1. Spatial distribution of homicides
In order to have a more accurate description of the dynamics of homicidal violence, we observed the neighborhoods inwhich events occurred. In Figure 5, we depicted the number of casualties per neighborhood for all years under study. Asit can be appreciated, the patterns of violence change over the period of time, and space. Importantly, we can observethe above-mentioned decrease in homicides after 2016.
Spatial evolution of homicides
In order to better understand the change in homicide patterns across time, we use a finer level of granularity of MMA:neighborhoods. We counted a total of , neighborhoods in municipalities, where of these neighborhoodshave at least one homicide in the January 2011–February 2018 time-window. In other words 28% of all neighborhoodsin MMA suffered a homicide within their boundaries. This contrasts with the known property of focalized crime on asmall number of neighborhoods within a city [6]. 6 rXiv Template
A P
REPRINT
Table 1: Pearson correlation coefficient and Z-score of weekly aggregated casualties at a municipality levelMunicipality 1 Municipality 2 PCC Z-scoreMONTERREY SAN NICOLÁS DE LOS GARZA 0.3581 6.4388GUADALUPE MONTERREY 0.3557 6.3556APODACA MONTERREY 0.317 5.6961CIÉNEGA DE FLORES SALINAS VICTORIA 0.2893 5.4055MONTERREY SANTIAGO 0.2839 5.0035JUÁREZ MONTERREY 0.2788 4.8732APODACA SAN NICOLÁS DE LOS GARZA 0.2547 4.6693CADEREYTA JIMÉNEZ MONTERREY 0.2422 4.2779CIÉNEGA DE FLORES HIDALGO 0.2263 4.143GENERAL ESCOBEDO MONTERREY 0.2291 4.0746CADEREYTA JIMÉNEZ HIDALGO 0.2053 3.8278APODACA CADEREYTA JIMÉNEZ 0.2149 3.8185PESQUERÍA SANTIAGO 0.2027 3.7373GUADALUPE SAN NICOLÁS DE LOS GARZA 0.2014 3.6547CARMEN SAN PEDRO GARZA GARCÍA 0.1992 3.4842JUÁREZ SANTIAGO 0.1829 3.1747MONTERREY PESQUERÍA 0.1756 3.0951CIÉNEGA DE FLORES SANTIAGO 0.1714 3.0139To look at the evolution of homicides across space, we constructed a homicide network representing the spatialconfiguration of violence within a given year, for each year of our time-window. In the homicide network nodes areneighborhoods and two nodes are connected by an edge if (i) they are spatially adjacent neighborhoods (their respectivepolygons are adjacent), and (ii) there was at least one homicide in the two neighborhoods.The analysis of each yearly crime network shows the spatial dynamics of adjacent neighborhoods in terms of homicide(Figure 6). The global spread of violence in distant focal points is seen across all years (scattered red points, neigh-borhoods with at least one homicide that year), suggesting a division in crime sectors (blue points are neighborhoodswithout homicides). The structure of crime sectors however, varies greatly between years. Years 2011 and 2012 havecrime networks with many connected components (45 and 50, respectively) suggesting a spread of homicides between ausually violent neighborhood to its adjacent neighborhoods. The number of connected components decreased steadilyuntil 2014 onwards (18), where a pattern of isolated violent neighborhoods emerges. Considering that the war on drugsclimaxed in the MMA in 2011 and decreased in late 2012, we can expect that the spreading of crime during these yearsis due to the unique properties of a war-like state. After 2014 the number of connected components stabilized until2017, where we see a 2-fold increase.In order to have a global view of the homicidal crime, we constructed a network using the whole-period window. There,nodes are connected if they are adjacent and have at least one homicide during the 86 months. Figure 7 represents sucha network, where nodes are colored by connected component with more than 10 edges (main connected components),white nodes are neighborhoods without crime (not in crime network), and black nodes belong to connected componentswith less than 10 edges.Looking at the composition of the nodes, each one of the eight main connected components adheres to a mainmunicipality strongly. For instance, yellow nodes belong to a main component with principal municipality Monterrey,containing 72% of the nodes. Red connected component has all its nodes in the municipality of Guadalupe, to the east.Green component in the west has 55 of its nodes in Santa Catarina.It is worth noticing cases such as the border between Santa Catarina and San Pedro Garza García municipalities(Green and white nodes in the west). Despite the fact that several neighborhoods of both cities share a border, noneof these share an edge. Interestingly, safe neighborhoods seem to be located around the municipality boundaries thusdisconnecting the main connected components. 7 rXiv
Template
A P
REPRINT
Figure 5:
Geospatial yearly distribution of homicides in MMA . In these maps, neighborhoods are depicted accordingto the number of homicides that took place there. light colors represent lower number of casualties, meanwhile darkcolors take account for higher homicide numbers.
Homicides are related to urban environment
As we saw in the previous result, crime connected components or crime sectors are usually located within municipalities.Although interesting by itself, the fact that municipality borders can act as a separator between crime sectors leads toask whether there are other barriers to crime in the MMA.8 rXiv
Template
A P
REPRINT
Figure 6:
Geospatial distribution of homicides in MMA during the eigth years . This network representation showsthe clusters formed by all neigborhoods that have at least one homicide and are geographycally together. Each colorrepresents a network component. The bottom left table shows the municipalities of the MMA that belong to each cluster,and the number of neighborhoods belonging to that city. 9 rXiv
Template
A P
REPRINT
Figure 7:
Network composed by adjacent violent neighborhoods during the entire period under study . Eachcolor represents a connected component with more than 10 edges. The bottom left table indicates the municipalitiesthat belong to each component. Black dots are neighborhoods with at least one homicide but not connected to largecomponents.The relevance of high-speed roads or highways in crime has been previously shown to be an important factor [19].Specifically in the context of a time window comprising part of a drug war, highways are of high relevance: they areusually the spots where bodies were abandoned or displayed, persecutions between criminals and police/military takeplace on highways, etc. Moreover, highways crossing a city act as an urban boundary. They separate neighborhoods,and municipalities, and can separate social and urban landscapes.In order to include highways in our data, specifically in order to observe the proximity of crime events near a highwayand also analyze the distance of violent neighborhoods from them, we obtained all Open Street Map (OSM) data pointsrelated to highways in the MMA.As we see in Figure 8, distance from crime to highway varies between years. In order to ignore outlier points weconsider the 99th percentile (distance such that 99% of crimes are closer than or equal to a highway) over the eightyears. The two years related to the climax of the war on drugs (2011–2012) show the shortest distances to highways(4379 and 4475 meters, respectively). The time period between 2013–2016 shows longer distances to highways, and adecrease in distance is observed again in 2017.In general, highways appear to be part of a backbone of violence as many of the homicides lay directly on the highwaypaths (Figure 9). It can also be appreciated how the political division together with the highways could be formingthe patterns between violent and non-violent neighborhoods (red and blue neighborhoods, respectively). Indeed, thepolygons resulting from the intersection between municipalities and highways appears to unveil crime patterns andshould be further investigated. One of the most evident effects of this separation is again San Pedro Garza García andSanta Catarina municipalities. 10 rXiv
Template
A P
REPRINT
Figure 8: .Figure 9:
Geospatial distribution of homicides in MMA with highways . Discussion
Spatiotemporal dynamics of homicide
Regarding the temporal evolution of crime, an interesting result came from data of Santa Catarina municipality. Asmentioned in the results section, Santa Catarina is not the most violent municipality in MMA, but this place has amonotonous homicide rate, even during the years of low homicide rates in MMA. This behavior contrasts with the restof municipalities of the MMA: It can indeed be appreciated in Figure 4, where Santa Catarina’s correlations are thelowest for the entire set.On the other hand, the other municipalities with high homicide rates have a similar behavior, with an important increasein 2011–2012, short after decreasing until 2017, where violence raised again, but at lower levels.11 rXiv
Template
A P
REPRINT
Socio-economic segregation of homicide violence
Opposite to Santa Catarina, San Pedro Garza García municipality has a very low homicide rate (150 and 44 homicides,respectively). These two cities are separated by one avenue: Cromo Street. In the side of San Pedro there were nohomicides, as opposite to Santa Catarina, where several neighborhoods suffered from homicidal violence. This contrastcan be attributed to the difference in the GDP between both places. San Pedro Garza García is the second city withthe highest GDP in Mexico [20], meanwhile some neighborhoods in Santa Catarina have important development lag.According to the last economical survey (2015), the GDP per capita in San Pedro Garza García was $25,636 USD,meanwhile that of Santa Catarina was $10,783 USD.Similar to this behavior related to revenue per capita, other municipalities such as San Nicolás de los Garza, have asmall number of neighborhoods with homicides compared to the rest of municipalities within the city. However, in thecase of San Nicolás de los Garza, homicides occurred chiefly in its periphery. This can be observed from Figure 9,several neighborhoods on its central area are in blue, meanwhile the borders are in red (north-east of MMA).Concomitant with the political borders of municipalities, highways also appear to depict lines of segregation betweensecure and violent neighborhoods. The remarkable difference between neighborhoods at opposite sides of a highwaymay reflect the presence of security forces in only one of both sides. Further investigation regarding this topic is neededto reach a solid conclusion, however, we discuss here the empirical observation.
Networks and geopolitical division of homicide
The crime network constructed from joining adjacent neighborhoods if they had at least one homicide during the timewindow, clearly shows that homicide events are linked to the municipality in which the homicide occurred. Similarto the aforementioned relationship between GDP and violence, here we observe that homicide regions are stronglysegregated by the political division of the MMA: network components belong to practically just one municipality.Table 2: Proportion of main municipality within a main connected component.Connected component Number of neighborhoods Main municipality Proportion Ratio1 305 Monterrey 220 0.722 80 Guadalupe 34 0.423 65 Apodaca 65 14 55 Guadalupe 55 15 50 Santa Catarina 48 0.966 23 Cadereyta 23 17 19 Juárez 19 18 16 General Escobedo 15 0.94This last result is reinforced with Figure 7, where each component belongs to one municipality with a high proportion(Table2). In this sense, a very interesting case is the one corresponding to San Nicolás de los Garza and Guadalupemunicipalities, as they share neighborhoods with other violent municipalities in the crime network. On the one hand,Guadalupe has two main connected components associated with it: a cluster exclusively composed by Guadalupe’sneighborhoods, and another with neighborhoods shared with other two municipalities, i.e. Monterrey, San Nicolas,Apodaca and Juarez. Finally, there is no main connected component in San Nicolás de los Garza holding a substantialnumber of neighborhoods.Probably, both phenomena should be explained by the fact that they are disputed territories or even transit-only places.The second hypothesis may apply better to the case of San Nicolás, as it does not have a main connected component.The other fact that emerges from the crime network, is that several neighborhoods are in the middle of clearly violentareas but they do not have any casualty during the eight years of data. As an example of this, we point to thoseneighborhoods between the green, yellow and red clusters in Figure 7: All these neighborhoods belong to Guadalupemunicipality. The gray nodes inside the yellow component, which belong to Monterrey, are also remarkable, but themost dense set of neighborhoods with no main connected component is the one between the yellow and cyan clusters.This set is part of San Nicolás and despite the fact that (i) this area is surrounded by three crime connected components,and (ii) it also contains non-clustered homicide neighborhoods (black dots), hundreds of neighborhoods are not touchedby the homicidal violence.This last observation coincides with the hypothesis that San Nicolás de los Garza municipality is not a disputed territory,but instead a transit one. This fact is fully inline with the urban environment and its relation with homicides in MMA.12 rXiv
Template
A P
REPRINT
Urban environment and homicidal behavior
By observing Figure 9, we may corroborate that the urban environment also shapes the broader distribution of homicidesin MMA during the measured period. The central polygon of Monterrey’s Downtown encompasses the most violentarea of MMA during the whole period (Centro de Monterrey neighborhood). Similarly, the 85th freeway that crossesMonterrey and reaches the southern municipality of Santiago is also a hub in terms of how many homicides occurred onthis road.By observing Figure 9 one may notice how many homicidal events occurred on highways. As previously mentioned,it is possible to observe how roads separate places with or without homicides: Apparently the combined political,socio-economical and transportation factors may separate relatively large areas of homicidal violence.
Concluding remarks
In this work, by means of a network approach, using a unique and carefully curated database of geolocated homicidesduring a period of eight years, together with map data from Open Street Maps, we have been able to describe thespatiotemporal dynamics of homicidal violence in one of the most important urban areas in Mexico, the MonterreyMetropolitan Area (MMA).To our knowledge, this is the first time that such an amount of manually curated data is used to construct spatial andtemporal networks, to provide insights of how violence increases and decreases as relative to the underlying differentsocial turmoil in Mexico during this time window.We have been capable to study at different levels of granularity of temporal (8-year, year, and month), and spatialcomponents (country-level, state-level, municipality, and neighborhood), in order to provide a multi-scale approach thatallows to dissect possible explanations behind the violent behavior in urban metropolitan areas, specifically in the caseof the war on drugs in Mexico.Further steps in this regard may focus in performing null models to corroborate whether the geo-socio-political divisionsegregates homicidal regions in MMA. A large effort of several groups should also be made on the data collection andcuration.
Methods
Data acquisition
Events mapping
Entries from the ENDB were mapped into the shapefiles using latitude and longitude coordinates to locate the neighbor-hood and municipality where each event took place. Entries were kept if they were located inside a neighborhood tofilter for urban areas. Mapping was performed using pandas and geopandas packages in Python. After the filter, 2114entries in the ENDB remained.
Neighborhood networks
Adjacent polygons for each neighborhood were determined using geopandas . Each node represents a neighborhoodand nodes were linked if both nodes had an event during a determined time window in the ENDB and if they wereadjacent polygons in the geospatial data. A network was build using the entire period and yearly networks were also13 rXiv
Template
A P
REPRINT generated. For the spatial evolution of homicides, networks were assembled by counting the number of events that wereregistered in the same week for two adjacent neighborhoods and by using a one week shifted window. Networks werecreated using networkx and visualization and analysis of their structural features were performed using Cytoscape.
Municipalities correlation
The entire dataset of causalities aggregated by week was used to calculate the Pearson Correlation Coefficient (PCC)between each pair of municipalities in the MMA. A null-distribution was obtained from a thousand iterations of PCCcalculation of the resshufled causalities by week for each municipality. A z-score was assigned to each correlation valueby placing the observed PCC in the null-distribution. Correlation heatmap was generated using the
ComplexHeatmap package in R.
Open Street Map
Data from Open Street Map (OSM) was obtained using the Overpass API for the Monterrey Metropolitan Area (bbox = ( − . , . , ( − . , . ). Nodes and ways were parsed from the data using the pyosmium library,where ways were fined-grained to retrieve only their nodes coordinates. List of abbreviations • Monterrey Metropolitan Area: MMA • El Norte Data Base: ENDB • Enrique Peña Nieto: EPN • Felipe Calderón Hinojosa: FCH • Open Street Map: OSM
References [1] John Mill Ackerman Rose. The limits of transparency: The case of mexico’s electoral ballots.
Mexican LawReview , (8):1, 2007.[2] Salvador Maldonado Aranda. Stories of drug trafficking in rural mexico: territories, drugs and cartels in michoacán.
European Review of Latin American and Caribbean Studies/Revista Europea de Estudios Latinoamericanos y delCaribe , pages 43–66, 2013.[3] Jesús Espinal-Enríquez and Hernán Larralde. Analysis of mexico’s narco-war network (2007–2011).
PLoS One ,10(5):e0126503, 2015.[4] Jonathan D Rosen and Roberto Zepeda.
Organized Crime, Drug Trafficking, and Violence in Mexico: TheTransition from Felipe Calderón to Enrique Peña Nieto . Lexington Books, 2016.[5] Jacqueline Cohen, Daniel Cork, John Engberg, and George Tita. The role of drug markets and gangs in localhomicide rates.
Homicide Studies , 2(3):241–262, 1998.[6] Richard Rosenfeld, Robert Fornango, and Andres F Rengifo. The impact of order-maintenance policing on newyork city homicide and robbery rates: 1988-2001.
Criminology , 45(2):355–384, 2007.[7] Harvey Warren Zorbaugh.
The gold coast and the slum: A sociological study of Chicago’s near north side .University of Chicago Press, 1983.[8] Sistema nacional de información municipal.[9] George E Tita and Robert T Greenbaum. Crime, neighborhoods, and units of analysis: putting space in its place.In
Putting crime in its place , pages 145–170. Springer, 2009.[10] Robert J Sampson.
Great American city: Chicago and the enduring neighborhood effect . University of ChicagoPress, 2012.[11] April M Zeoli, Jesenia M Pizarro, Sue C Grady, and Christopher Melde. Homicide as infectious disease: Usingpublic health methods to investigate the diffusion of homicide.
Justice quarterly , 31(3):609–632, 2014.[12] Patricia L Brantingham, Martin Ester, Richard Frank, Uwe Glässer, and Mohammad A Tayebi. Co-offendingnetwork mining. In
Counterterrorism and Open Source Intelligence , pages 73–102. Springer, 2011.14 rXiv
Template
A P
REPRINT [13] James A Kitts. Beyond networks in structural theories of exchange: Promises from computational social science.In
Advances in group processes , pages 263–298. Emerald Group Publishing Limited, 2014.[14] Andrew V Papachristos, David M Hureau, and Anthony A Braga. The corner and the crew: The influence ofgeography and social networks on gang violence.
American sociological review , 78(3):417–447, 2013.[15] Andrew V Papachristos. Murder by structure: Dominance relations and the social structure of gang homicide.
American Journal of Sociology , 115(1):74–128, 2009.[16] Marcos Oliveira, Eraldo Ribeiro, Carmelo Bastos-Filho, and Ronaldo Menezes. Spatio-temporal variations in theurban rhythm: the travelling waves of crime.
EPJ Data Science , 7(1):29, 2018.[17] CA Piña-García and Leticia Ramírez-Ramírez. Exploring crime patterns in mexico city.
Journal of Big Data ,6(1):65, 2019.[18] Lewis F Richardson. Frequency of occurrence of wars and other fatal quarrels.
Nature , 148(3759):598, 1941.[19] Young-An Kim and John R Hipp. Physical boundaries and city boundaries: consequences for crime patterns onstreet segments?