English (pdf)
Article in xml format
Article references
Automatic translation
Send this article by e-mail
Cited by SciELO 
Cited by Google 
Similars in
SciELO 
Similars in Google 
Permalink
Introduction
Homicidal violence in Latin American cities has established itself as one of the most persistent, complex and territorially structured phenomena of the 21st century. In multiple urban contexts in the global South, homicides not only represent an extreme expression of criminality, but also reveal deep social, institutional and territorial tensions. Among the multiple factors that explain this lethal violence, the presence of illegal drug markets has been identified as a key determinant, both for their capacity to generate disputes between criminal actors and for their implications for the forms of informal territorial control that emerge where the state is weak or absent.
Numerous studies have documented the relationship between drug trafficking and violence, and have highlighted that this relationship is neither linear nor uniform, but depends on contextual, organisational and institutional variables (Goldstein, 1986; Jiménez-García et al., 2023; Lessing & Willis, 2019). In particular, systemic violence linked to the day-to-day functioning of drug markets has been proposed as a robust theoretical framework to explain the spatial concentration of homicides in certain urban areas. However, despite the extensive literature on crime and drugs in Latin America, few studies explicitly incorporate the spatial dimension of these phenomena, as well as the mechanisms by which they are structured and spread territorially.
This gap is particularly problematic if one takes into account that homicidal violence linked to drug trafficking is not only concentrated territorially, but also spreads, influences and is reconfigured in urban spaces according to criminal dynamics and institutional responses. Understanding how and why violence is concentrated in certain neighbourhoods, and what role drug seizures play as indications of criminal activity or state intervention, is fundamental for the design of effective and territorially sensitive public policies In this context, this article addresses the question: is there a spatial dependency relationship between homicides and drug seizures in Latin American urban contexts? To answer this question, a comparative analysis is developed between Cali (historical centre of Colombian drug trafficking), Curitiba (Brazilian logistical node dominated by the First Capital Command (PCC using the acronym in Portuguese) and Monterrey (Mexican border city vied for by fragmented cartels). This selection allows us to examine how different configurations of illegal markets and territorial governance structures influence patterns of urban lethal violence.
Methodologically, the study applies spatial econometric models to identify the magnitude and direction of the effects that seizure activity has on the distribution of homicides by neighbourhood. This approach reveals a counter-intuitive finding: seizures do not always reduce violence; in certain contexts, they can exacerbate it by destabilising pre-existing criminal equilibriums, with important implications for the urban security strategies currently implemented in the region.
This article is structured in five main sections: a review of the relevant literature, the presentation of the comparative methodology employed, the empirical results of the spatial analysis, the discussion of its conceptual and practical implications for the design of public policies with a territorial approach, and a final annex that includes the official statistics on crime in Colombia corresponding to 2024.
Literature review
Understanding the spatial structure of the relationship between homicides and drug seizures requires a theoretical and methodological approach that integrates multiple perspectives on urban violence, illicit markets and their territorial manifestations. This state of the art articulates the main conceptual and empirical contributions around three complementary dimensions: the theoretical frameworks that explain the mechanisms underlying drug- related violence; the spatial methodological approaches that allow us to model and analyse these territorial relationships; and the specific findings of previous research in Latin American contexts similar to the cities studied. This critical review will allow us to identify not only the established consensus in the literature, but also the knowledge gaps that justify the South-South comparative approach adopted in this research.
Goldstein’s theory of systemic violence
One of the main references in the study of the relationship between drugs and violence is Paul Goldstein (1985), whose tripartite conceptual model has been widely tested and debated by scholars around the world (Jacques & Wright, 2011; Ousey & Lee, 2007; Sarrica, 2008). This model explains the connection between drugs and violence through three non-mutually exclusive mechanisms: psycho-pharmacological violence, economic-compulsive violence and systemic violence (Goldstein, 1985).
This study focuses particularly on the third type: systemic violence. Goldstein defines systemic violence as traditionally aggressive patterns of interaction within the drug distribution and use system (Goldstein, 1985). This type of violence arises from the demands of working or doing business in an illicit market, where actors have no recourse to the legal system to resolve their conflicts or to promote their product (Goldstein et al., 1997). Systemic violence functions as a mechanism of informal social control within drug markets, replacing formal institutions of dispute resolution (Liem & Moeller, 2025). The territorial character of systemic violence is particularly relevant to understanding the spatial distribution of homicide in urban contexts. As Liem and Moeller (2025) indicate, systemic violence manifests itself in disputes over the control of strategic areas for the distribution, processing or transit of illicit substances. Territoriality emerges as a fundamental element of drug-related violence, generating patterns of spatial concentration of homicides in certain urban areas (Jiménez-García et al., 2023).
Goldstein (1985) provided specific examples of systemic violence that include: territorial disputes, sanctions for non-compliance with internal and external drug market codes, elimination of competition, punishments for disobedience, restructuring the power structures of criminal organisations, and response to state force. These diverse mechanisms explain why systemic violence constitutes a substantial proportion of drug-related violence in Latin American urban contexts (Jiménez-García et al., 2023).
In Latin American contexts, several researchers have adapted and applied Goldstein’s theory. Jiménez-García et al. (2023) conducted an empirical study in Pereira (Colombia), in which they applied both Goldstein’s systemic violence theory and contingent causality theory, and demonstrated through structural equation analysis (PLS-SEM) that there is a significant relationship between violence, drug trafficking and socio-economic disadvantage. This work pioneered the empirical testing of Goldstein’s framework in a Colombian urban context and concluded that systemic violence is behind a substantial proportion of homicides in cities with strategic drug markets.
In another significant study, Jiménez-García and Rentería-Ramos (2020) argued that drug trafficking acts as a catalyst for homicidal violence in Colombian cities through systemic mechanisms operating at different territorial scales. In Mexico, Valdés (2013) documented how the fragmentation of the trafficking system following the militarisation of the drug war intensified systemic violence in border cities. Similarly, Rubio-Ramos (2024) in Colombia has examined the spatial relationship between institutional trust, violence and cocaine markets, and found that perceptions of state legitimacy moderate the impact of seizures on homicide levels.
However, as Reuter (2009) points out, Goldstein’s theory has limitations in explaining spatial and temporal variations in the levels of violence associated with drug markets. Systemic violence is not homogeneously distributed across urban space, even when the presence of illicit markets is constant. This suggests the need to complement this perspective with approaches that address specific territorial determinants and local configurations of criminal power.
Blumstein (1995) built on Goldstein’s tripartite framework and added a fourth, broader connection between drugs and violence: the community disruption effect of the drug industry. Norms and behaviours within the drug industry influence the behaviour of others who have no direct connection to the drug industry. For example, the influence of the widespread presence of guns among drug dealers may encourage others in the community to similarly arm themselves for self-defence, or use those firearms to resolve their own disputes that have nothing to do with drugs.
This effect of community disorganisation is especially visible in Latin American contexts, where the dynamics of the illegal drug market profoundly transform urban social fabrics. Systemic violence, in this context, not only responds to the economic logic of drug trafficking, but also operates as a mechanism of territorial and social control, shaping forms of informal regulation and parallel power structures. These forms of order, although illegal, contribute to shaping everyday practices, local hierarchies and norms of coexistence, transcending drug transactions to influence multiple dimensions of urban life (Misse, 2019).
Goldstein’s theory of systemic violence thus offers a robust conceptual framework for understanding the spatial distribution of homicides in Latin American cities with a strong presence of drug markets. However, its application must take into account the contextual specificities of each territory, the organisational characteristics of criminal groups and the particular configurations of local drug markets.
Adaptive criminal spaces: An emerging theory
Contemporary approaches to illicit markets have begun to conceptualise them not as static entities, but as complex adaptive systems with the capacity to reorganise, learn and evolve (Emmerich, 2015). This perspective, developed by authors such as Williams (2001) and Kenney (2007), argues that criminal organisations possess an evolving intelligence that allows them to respond strategically to institutional pressures, reconfiguring their structures, territories and operating methods in accordance to the threats and opportunities presented in the environment. In contrast to traditional approaches that conceive of illegal markets as passive objects of state intervention, this theory emphasises their dynamic nature and morphological adaptability. As Bouchard (2007) points out, criminal networks exhibit resilience properties that allow them to absorb significant disruptions without collapsing, redistributing resources, redirecting routes and modifying hierarchies. These structural adjustments can manifest themselves spatially, generating patterns of concentration, dispersion or territorial displacement in response to external pressures.
Ayling’s (2009) empirical studies on criminal adaptability in contexts of high institutional pressure have shown that illicit organisations develop homeostatic mechanisms similar to those of biological organisms to maintain their functionality under adverse conditions. This adaptive capacity manifests itself at three levels: tactical, evident in changes in operational methods; strategic (for example, when there is a reconfiguration of alliances and value chains); and systemic, when there is a fundamental transformation of the organisation and its relationship with the environment. Violence, in this framework, operates as a multifunctional adaptive mechanism that allows for defence, expansion or reorganisation depending on the circumstances.
In particular, Bright and Delaney’s (2013) proposal on the territorial plasticity of criminal markets is relevant for this study: in the face of localised interventions, they do not disappear, but migrate, fragment or mutate into new spatial configurations. In their research on the evolution of a drug trafficking network, these authors documented how the density of the network remained relatively stable despite external interventions, albeit with significant changes in its decentralisation and in the roles played by key members. This phenomenon explains why institutional actions focused on specific areas often generate displacement effects, diffusion or intensification of violence in adjacent or functionally connected territories.
Empirical evidence suggests that these adaptive dynamics vary significantly according to institutional context, criminal competition structure and specific territorial history. For example, Snyder and Durán- Martínez (2009) documented how, faced with similar state pressures, illicit markets in Mexico responded with strategies of violent confrontation, while in Colombia many organisations opted for strategies of mimicry and invisibilisation. This variability in responses is conditioned not only by internal factors within criminal organisations, but also by the coherence of state institutions and the degree of competition in illegal markets.
These contextual differences in adaptive responses explain why state interventions produce heterogeneous effects on patterns of violence. As Morselli and Petit (2007) have argued, the effectiveness of institutional actions depends less on their intensity than on their ability to anticipate and counteract subsequent criminal adaptations. In contexts where interventions disrupt criminal equilibria without creating viable alternatives, illegal organisations may intensify the use of violence as a compensatory mechanism to re-establish territorial control, discipline emerging competitors or signal their capacity to outsiders.
This complex adaptive system’s perspective reveals how criminal organisations operate under their own logic of survival and expansion, regardless of the institutional strategies implemented to combat them. Criminal behaviour does not respond to random patterns, but to an adaptive rationality that seeks to maximise benefits and minimise risks in the face of any change in their environment. In this context, it is criminal organisations that use violence as a mechanism to relate to the territory where they operate, and use it as a strategic resource to maintain their social control and resolve internal disputes. As Bright et al. (2017) specify, the effectiveness of security policies increases when they manage to anticipate and neutralise these criminal adaptive capacities, generating institutional environments where violence loses its instrumental value for illegal actors.
The ecology of crime and the hotspots theory
The ecology of crime and the theory of dangerous places bring a complementary dimension to the analysis of the spatial distribution of drug-related violence. This approach, developed mainly by Weisburd and Telep (2014), argues that crime is not randomly distributed in urban space, but tends to concentrate in specific micro- spaces that maintain their temporal stability as crime hotspots. The so-called “law of crime concentration” maintains that nearly 50% of crimes are committed in just 5% of urban space (Weisburd, 2015). This spatial concentration is even more pronounced in crimes related to illicit markets, as demonstrated by Weisburd and Green (1995) when documenting that drug transactions present higher levels of spatial concentration than other types of crime.
The ecological approach emphasises the role of environmental and situational factors in shaping spatial patterns of criminality. According to Manzano et al. (2020), places that concentrate criminal activity tend to have specific socio-spatial characteristics: high population transience, low informal social control, low natural surveillance and optimal accessibility for illicit activities. These characteristics explain why certain neighbourhoods become epicentres of drug markets and associated violence.
In Latin American contexts, several studies have applied this ecological perspective. Cerda et al. (2012) documented in Medellín how specific interventions in urban space modified patterns of violence associated with illicit markets, while Morenoff et al. (2001) found that the spatial concentration of socio-economic disadvantage predicts the formation of hot spots of homicidal violence.
Meanwhile, in a pioneering study on the concentration of crime in micro-spaces in Latin America, Chainey et al. (2019) analysed data from 42 cities in Argentina, Brazil, Colombia, Mexico, Uruguay and Venezuela. Their findings revealed that crime is concentrated at significantly higher levels than in Western industrialised contexts; in fact, 50% of homicides occur in only 1.4% of the street segments of all cities studied. This spatial dependence, in which criminal events in one place influence the likelihood of similar events occurring in adjacent areas, manifests itself through contagion and diffusion effects that are particularly strong in drug market contexts such as those studied in this article.
The ecological perspective offers solid theoretical foundations for the application of spatial econometric techniques, such as the spatial lag model (SLM) and the spatial error model (SEM), as they explicitly recognise that criminal phenomena are not spatially independent and their distribution responds to structured processes of territorial dependence. These models not only confirm the association between seizures and homicides, but they also allow us to identify direct effects (in the same neighbourhood) and indirect effects (in adjoining neighbourhoods), reflecting precisely the spatial diffusion mechanisms postulated by the ecology of crime.
Spatial models for understanding patterns of crime and violence
The spatial analysis of crime has undergone a significant methodological transformation in recent decades, evolving from simple hotspot mapping to sophisticated spatial econometric models capable of capturing complex relationships of territorial dependency. This evolution responds to the progressive recognition that criminal phenomena, particularly those related to illicit markets, present; structured spatial patterns that cannot be adequately captured by conventional statistical techniques (Anselin et al., 2000).
Early modern approaches to spatial analysis in criminology focused on the identification and visualisation of geographical concentrations of crime (hotspots) using techniques such as quadrant analysis, Kernel density estimation and Getis-Ord statistics (Eck et al., 2005). These approaches, while useful for the initial description of patterns, were insufficient to model complex interactions between adjacent spatial units and to control for the influence of contextual factors.
The development of spatial econometrics, mainly driven by Anselin (1988), provided a more rigorous methodological framework to address these challenges. The main contribution of this approach lies in its ability to explicitly model three types of spatial effects: spatial autocorrelation (dependence between values of a variable in nearby spatial units), spatial heterogeneity (systematic variation of relationships across space) and spillover effects (transmission of influences between adjacent areas) (LeSage & Pace, 2009).
Spatial econometric models are a key tool for analysing spatially dependent phenomena, such as homicidal violence in urban contexts marked by drug trafficking and other illegal economies. In particular, spatial lag models (SLM) and spatial error models (SEM) allow us to capture, respectively, the effects of contagion or diffusion of violence between neighbouring territorial units and the processes of spatial autocorrelation not directly explained by the independent variables.
In theoretical terms, these models are grounded in the ecology of crime and, in the case of the present study, in systemic violence theory, which posit that crime is spatially distributed in a non-random fashion due to the interaction between criminal opportunities, the absence of effective gatekeepers, and the presence of motivated offenders and criminal markets (Brantingham & Brantingham, 1993; Cohen & Felson, 1979; Goldstein, 1985). This logic has been empirically confirmed in recent studies on urban violence (Lee & Lee, 2020; Lin et al., 2025).
The SLM model is particularly useful when it is assumed that the level of violence in a territorial unit (such as a neighbourhood or a commune) is directly influenced by the levels of violence in neighbouring units. This approach captures a spatial feedback effect, characteristic of contexts in which disputes over the control of illegal markets or the presence of armed groups generate patterns of territorial diffusion of crime (Biagi & Detotto, 2014; Lee & Lee, 2020). The SEM model, on the other hand, is appropriate when the error is expected to be spatially correlated, i.e. there are unobserved factors that simultaneously affect contiguous units.
Both Biagi and Detotto (2014) and Fitterer et al. (2018) highlight the importance of accounting for spatial effects in their analyses of crime. Biagi and Detotto’s study on tourism and crime in Italy found that spatial dependence was statistically significant, demonstrating that ignoring these effects can lead to biased estimates. Similarly, Fitterer et al. (2018) used a spatially lagged Poisson model to analyze the link between violent crime and bar concentration in Victoria, Canada. Their research revealed that crime is not just correlated with local neighborhood conditions but also with the conditions of surrounding neighborhoods, further confirming the need to consider spatial dependencies.
In South Korea, a study by Lee and Lee (2020) on violent crime confirms that variables such as the density of dangerous facilities or family disorganisation have significant effects in spatial models, highlighting the need to consider both local conditions and spillovers from neighbouring areas. In a similar vein, Lin et al. (2025) analyse wildlife crime in China’s Yunnan province using an SLM model, and find spatial concentration patterns that could not be explained without including spatial structure in the analysis.
This type of modelling has also been used to study other urban phenomena, such as graffiti. In particular, Alattar’s (2024) study of graffiti in San Francisco applies both lag and spatial error models, and shows that streets with a high degree of centrality and connectivity in the road network have a higher incidence of graffiti. Furthermore, it finds that the spatial error of a street significantly influences the amount of graffiti reported on adjacent streets. This finding supports the hypothesis that criminal phenomena or urban disorder are spatially diffused not only by local characteristics but according to the structure of territorial connectivity.
In the framework of this study, the use of SLM and SEM models allows not only to improve the accuracy of estimates of the relationship between homicides and drug seizures, but also to identify the logistics of concentration and diffusion of violence, which could be relevant for the design of targeted territorial intervention policies.
Methodology
This study employs a case comparison approach to analyse the spatial relationship between homicides and drug seizures in Monterrey, Cali and Curitiba. Combining spatial analysis and spatial econometric modelling, we seek to identify patterns of geographic concentration and spatial dependence. The selection of cities maximises the variability of urban and criminal contexts, allowing us to explore how the dynamics of illicit markets shape homicidal violence in different Latin American realities.
Type of study
This study adopts a comparative case study approach (Bartlett & Vavrus, 2017), focusing on analysing the spatial structure of the relationship between homicides and drug seizures in three Latin American cities: Monterrey (Mexico), Cali (Colombia) and Curitiba (Brazil). The comparative case strategy allows us to explore the variability of spatial patterns of violence and criminal markets in heterogeneous urban contexts, simultaneously controlling for relevant geographical, political and criminal differences (Stake, 2006).
The comparative design allows for the observation of how adaptive criminal systems respond in a differentiated manner to specific institutional and territorial contexts (Bright et al., 2017), integrating both the structural dimension and the dynamic aspects of urban criminality. It is framed within a South-South comparative perspective, which privileges cities in the global South with shared historical processes of rapid urbanisation, structural inequality and the evolution of illicit markets. Unlike traditional approaches that contrast cities in the global South with metropolises in the North, this research prioritises internal Latin American diversity and recognises the convergences and local specificities of these urban contexts.
The selection of cases follows the criterion of maximum variation (Gerring, 2006; Stake, 2006), allowing us to contrast political systems, security strategies, trajectories of violence and diverse criminal configurations. In this way, the aim is not only to describe spatial patterns of homicides and seizures, but also to contribute theoretically to the understanding of the spatial organisation of crime in Latin American cities.
Selection of cities
The choice of Monterrey, Cali and Curitiba as units of analysis is based on their strategic relevance at the national level and in Latin American criminal dynamics. Monterrey, capital of the state of Nuevo León, is located in the northeast of Mexico, less than 300 kilometres from the border with the United States. This geographical position makes it a crucial logistical node for transnational drug trafficking routes (Beittel, 2018; Villarreal, 2024). In addition to its economic importance, Monterrey has been the epicentre of violent disputes between cartels, which positions it as a key case to study the relationship between criminal markets and urban violence.
Cali, located in south-western Colombia, is the third most populated city in Colombia and a fundamental axis of connection between the Andean and coastal regions. Historically associated with the Cali Cartel in the 1990s (Chepesiuk, 2003), today it faces phenomena of micro- trafficking and the reconfiguration of criminal groups. Its status as an intermediate city in post-conflict transition makes it particularly relevant for examining the new forms of violence associated with drug markets (Rubio- Ramos, 2024).
The selection of Curitiba as a comparative case responds to its methodological value as a critical case (Flyvbjerg, 2006), which broadens the analytical spectrum under the maximum variation design (Seawright & Gerring, 2008), allowing for the examination of the relationship between drug trafficking and violence under distinctive institutional and criminal conditions. Unlike Monterrey, located on the US-Mexico border and affected by cartel fragmentation (Beittel, 2018), and Cali, with its historical legacy of criminal organisations structured in the Colombian context (Chepesiuk, 2003), Curitiba represents a distinct paradigm characterised by its strategic position in southern Brazil as a logistical node between the Triple Frontier and the Atlantic ports that integrate it into international cocaine routes (Misse, 2019); its criminal configuration dominated by the territorial hegemony of the Primeiro Comando da Capital, which has implemented a low-profile criminal governance model (Ferreira, 2019), and the implementation of public security policies with a non- militarised police emphasis (Cano & Ribeiro, 2016). This systematic variation in key contextual factors, including geopolitical location, criminal organisational structure and form of state response, along with the constant significant exposure to drug trafficking, makes Curitiba a methodologically essential component for assessing the hypothesis of territory-specific effects in the drug- violence relationship (Lessing & Willis, 2019) and for developing a comparative framework that transcends the limitations of single-case studies prevalent in the Latin American literature on organised crime (Jiménez- García et al., 2023).
The selection of these three cities allows us to contrast different trajectories of violence, strategies of state control and spatial configurations of crime. The comparison is also based on the heuristic value of approaching cities in the global South under the criterion of maximum variation (Marques, 2024) and on the use of optimal urban mesoanalytic units to analyse criminal governance.
Units of analysis and variables
This study adopts the neighbourhood level as the unit of analysis in each of the selected cities: Monterrey, Cali and Curitiba. The choice of the neighbourhood as the scale of observation responds to the need to capture micro- territorial spatial dynamics, where the manifestations of homicidal violence and the activity of criminal markets tend to be concentrated and spatially structured.
In the case of Cali, neighbourhoods correspond to official administrative divisions recognised by the Administrative Department of Municipal Planning, constituting reference areas for the provision of public services, urban planning and the monitoring of security dynamics. In Curitiba, neighbourhoods fulfil an equivalent function: they are official administrative divisions recognised by the Instituto de Pesquisa e Planejamento Urbano de Curitiba (IPPUC), used for the management of local public policies. In Monterrey, the analogous term is colonia, which refers to urban units officially recognised as legalised settlements, functioning as a basis for territorial administration and the registration of urban and criminal statistics.
The choice of these units responds to their appropriate meso-analytical scale: they are small enough to capture phenomena of spatial concentration of violence and illicit market activity, but also have the structure and administrative recognition necessary to obtain reliable official data. This scale allows for accurate observation of both localised criminal governance processes and the spatial spread of violent phenomena into neighbouring territories.
We worked with two main variables: the dependent variable corresponds to the number of homicides registered in each neighbourhood during the period analysed, which allows us to capture the most severe expression of criminal violence and observe patterns of spatial concentration; the independent variable is defined as the number of drug seizure events in each neighbourhood, used as a proxy for the presence and activity of illicit markets in the urban space.
The analytical approach explicitly considers the spatial relationship between neighbourhoods and recognises that the phenomena of violence are not distributed in isolation, but show patterns of spatial dependency that need to be modelled appropriately.
Analytical strategy and spatial technique
To identify the spatial structure of the relationship between homicides and drug seizures, an analytical strategy based on spatial econometric models (AnseIin, 1988; LeSage & Pace, 2009) was employed. The analysis was carried out in three consecutive stages: the first stage assessed the existence of spatial dependence and the presence of autocorrelation; the second stage involved fitting the models and choosing the most appropriate model; and the third stage involved interpreting the estimated effects.
First, the existence of spatial dependence in the data was assessed by calculating the overall Moran index, a widely used measure of spatial autocorrelation to determine whether high or low values of a variable tend to cluster spatially.
Once the presence of significant spatial autocorrelation was confirmed, spatial regression models were fitted to estimate the relationship between seizures and homicides (second stage), controlling for the influence of space. Two main models were applied: the spatial lag model (SLM), which incorporates the value of the dependent variable in neighbouring units as a predictor and thus captures the possible spatial diffusion or contagion of violence; and the spatial error model (SEM), which introduces spatial dependence in the error terms, allowing us to identify whether there are unobserved factors that are distributed in a spatially correlated manner.
The choice of the most appropriate model for each city was based on the Akaike information criterion (AIC) and the application of LM (Lagrange multiplier) tests, which allow us to assess whether it is more appropriate to model spatial dependence as a direct relationship between neighbouring units (SLM) or as a spatially correlated error structure (SEM). Finally, in the third stage, to interpret the estimated effects, the direct, indirect and total spatial impacts associated with the seizure variable were calculated. These impacts make it possible to distinguish whether the effect of seizures is concentrated in the same neighbourhood (direct impact), spreads to adjacent neighbourhoods (indirect impact) or is distributed in a combined manner (total impact) (Elhorst, 2014). This methodological approach allows addressing the research question from a perspective that recognises the spatially interdependent nature of criminal phenomena in complex urban settings.
The spatial models were applied exclusively on the homicide variable for the three cities, due to its empirical and theoretical relevance, as is also observed in its centrality within Colombian national statistics (see the final Tables of this article).
Results
Comparative descriptive statistics
Table 1. Descriptive statistics of homicides and events/seizures by city
Cali has the highest average number of homicides per neighbourhood (3.51), reflecting a particularly intense dynamic of lethal violence, followed by Curitiba (2.34) and, at a greater distance, Monterrey (0.96) (Table 1). This order reveals stark contrasts in the logistics of urban violence amongst the cities. While Cali combines high levels of homicides with a considerable presence of drug seizure events (3.91 per neighbourhood), Curitiba presents a particular situation: despite registering an intermediate level of homicides, it concentrates an average of 118.09 drug seizure events per neighbourhood, a very high figure compared to Cali and Monterrey (2.05). This divergence suggests that in Curitiba institutional pressure on drug markets may be greater or more focused, while in Cali lethal violence and illicit market activity appear to be more directly intertwined. Monterrey, on the other hand, shows a comparatively contained situation of violence, both in terms of homicides and seizure events.
These preliminary differences in homicide and event averages highlight the diversity of the selected urban contexts, which will be further explored through subsequent spatial analyses. Although absolute levels of crime vary, the main interest lies in identifying patterns of spatial dependence rather than directly comparing rates across cities.
Global spatial autocorrelation
The overall spatial autocorrelation of homicides was assessed using the Moran index, the results of which are presented in Table 2. This analysis allows us to determine whether the distribution of homicides in space is random or whether, on the contrary, there are patterns of geographical concentration.
The detection of patterns of geographic concentration implies that homicides do not occur randomly, but tend to cluster spatially. This concentration suggests the existence of territorial dynamics of violence, one of the main explanations being the presence of criminal markets, in particular drug trafficking, whose activity is reflected in the recorded seizures. While this study focuses on the relationship between homicides and drug seizures, it also recognises that other factors such as socio- economic inequalities, criminal governance dynamics or failures in territorial control may influence the observed spatial concentration. The finding justifies the need to apply spatial models that explicitly incorporate spatial dependency in the analysis of homicides.
The positive and significant values of the Moran index (p < 0.05) in the three cities indicate the presence of spatial autocorrelation in the distribution of homicides, although its intensity varies. Cali has the highest level of spatial dependence (followed by Curitiba and, lastly, Monterrey) (Figure 1 and 2). These findings suggest that homicides tend to be concentrated in certain urban areas, an aspect that will be further explored through local cluster analyses.
Distribution of spatial clusters
The spatial distribution of homicides reveals the existence of clusters of high concentration (High-High), where neighbourhoods with high homicide levels are surrounded by equally violent neighbourhoods, as well as clusters of low relative concentration (Low-High), where neighbourhoods with low homicide levels are close to areas of high violence. Table 3 shows the frequency of the different types of clusters identified, allowing us to observe relevant differences in the spatial concentration of violence.
In Monterrey, 12 neighbourhoods were identified as High-High and four as Low-High clusters (Figure 3), revealing specific concentrations of homicidal violence in a city where, in general, most neighbourhoods do not present significant clusters. In Curitiba, four neighbourhoods were identified as High-High clusters, representing 4.9 % of the total number of neighbourhoods analysed (Figure 3), indicating the presence of localised clusters of violence. In Cali, 12 neighbourhoods were identified as High-High (4.5 %) and seven as Low- High (2.6 %) (Figure 3), reflecting a denser and more widespread spatial configuration of violence than in the other cities analysed.
These results indicate that, despite differences in homicide volume between cities, there are relevant patterns of spatial concentrations that warrant the incorporation of spatial regression models for further analysis.
Optimal spatial models
After identifying spatial concentration patterns, spatial regression models were fitted for each city to capture the spatial dependence detected. Three specifications were compared: the ordinary regression model (OLS), the spatial lag model (SLM) and the spatial error model (SEM), and the best model was selected according to the Akaike information criterion (AIC) and the diagnosis of autocorrelation in the residuals (Table 4).
In Monterrey and Cali, the best fit using the spatial error model (SEM) suggests that homicidal violence presents a spatial organisation that cannot be explained by observed variables alone, which is consistent with complex territorial processes, such as the presence of criminal markets and illegal governance dynamics (Figure 4). These results confirm the need to employ spatial techniques to capture invisible externalities and dependencies in urban processes of violence, advancing the response to the research question posed. In Curitiba, the spatial lag model (SLM) indicates that homicides in one neighbourhood are not only explained by its own characteristics (Figure 4), but also by homicide levels in surrounding neighbourhoods, revealing a process of spatial diffusion or contagion of violence.
In all three cities, the event variable (drug seizures) showed a positive and statistically significant association with homicides. This implies that the presence or intensity of drug trafficking activities is spatially related to levels of lethal violence, although differences in the magnitude of the coefficients suggest that the weight of these dynamics varies across urban contexts.
These results do not imply that one type of model is intrinsically superior to the other, but rather that both capture different mechanisms of spatial dependence. While the SEM reflects unobserved structural influences, the SLM shows direct interactions between neighbouring spatial units. In this way, the models complement each other and enrich the understanding of how the relationship between homicides and drug seizures is spatially structured, contributing to answering the research question posed.
Spatial impacts
Spatial impact analysis complements the evaluation of optimal models by estimating the direct, indirect and total effect that drug seizures have on homicides. The direct impact measures how seizures affect homicides in the same neighbourhood where they occur, while the indirect impact reflects the effect on other neighbourhoods and reveals the dynamics of the spatial spread of violence. The sum of both defines the total impact, allowing us to understand whether the effects of criminal markets are local, by neighbourhood or combined in each urban context analysed (Table 5).
In Monterrey, where a spatial error model (SEM) was selected, drug seizure events generate a total positive impact of 0.2465 homicides, with the direct impact (0.1877) being considerably larger than the indirect impact (0.0588). This suggests that the dynamics of homicidal violence are mostly determined by unobserved structural factors within the same neighbourhood, rather than by the influence of nearby neighbourhoods.
In Cali, also modelled using SEM, the direct and indirect impacts are almost equivalent (0.0978 and 0.0904, respectively), indicating that both the neighbourhood itself and the surrounding area experience similar increases in homicides associated with drug events. This reflects a more homogeneous spatial distribution of criminal dynamics.
In Curitiba, where the spatial lag model (SLM) proved to be the most optimal, the impacts are smaller in magnitude, although still positive: the direct impact is 0.0119 and the indirect impact is 0.0060. Here, homicidal violence shows a more evident pattern of spatial spread, where homicides in one neighbourhood are related to the levels of violence in adjoining neighbourhoods.
The significance of these impacts was verified by tests associated with the calculation of the impacts in the spatial models. Statistical significance (p < 0.05) in all cases confirms that drug seizures are consistently associated with spatial variations in homicides.
Discussion of results
The results of this study allow us to advance our understanding of a fundamental question in the analysis of urban violence in Latin America: why is homicidal violence territorially concentrated in certain urban spaces and how is this concentration related to the presence of criminal markets, specifically drug markets? The empirical evidence presented confirms that homicide is not a random or homogenous phenomenon but is conditioned by territorialised criminal organisational structures. In the three cities analysed, statistically significant patterns of spatial dependence between homicides and drug seizures were identified, suggesting that illegal markets exert a structuring influence on the spatial configuration of lethal violence. These findings are part of a critical line of research that has documented how drug trafficking operates as a key explanatory variable in urban contexts marked by inequality, informality and weak state presence (Goldstein, 1986; Jiménez-García, 2024; Jiménez-García et al., 2024; Misse, 2019).
However, one of the main contributions of this study is to reveal that this relationship between homicides and criminal markets is neither linear nor uniform. On the contrary, it is expressed in different ways depending on the institutional, geographical and criminal context of each city. In Curitiba, for example, homicidal violence shows a pattern of spatial diffusion (SLM model), in which the levels of violence in one neighbourhood are significantly associated with those of its neighbours. This finding is consistent with the idea of a territorial contagion effect (Weisburd & Telep, 2014) but also suggests that forms of criminal governance - in this case, the hegemony of the Primeiro Comando da Capital - could be containing absolute homicide levels by rationalising their use as a control mechanism (Ferreira, 2019). In contrast, in Cali and Monterrey, where the optimal model was the SEM, homicides appear to be determined by structural factors not directly observed, which could include informal arrangements, power vacuums or market reorganisation dynamics following state interventions (Johnson, 2016; Reuter, 2009; Snyder & Durán-Martínez, 2009).
The results of this study reveal how criminal spaces operate as complex adaptive systems with the capacity to reorganise, adapt and evolve in the face of external pressures. As demonstrated by the different spatial patterns in the three cities analysed, criminal organisations do not remain passive in the face of interventions, but rather they develop strategic territorial responses specific to each urban context. In Curitiba, for example, the SLM model indicates a spatial contagion effect that reflects adaptive mechanisms of territorial diffusion, probably facilitated by the hegemony of the Primeiro Comando da Capital, which has implemented forms of criminal governance that rationalise the use of violence (Ferreira, 2019). In contrast, the SEM models identified in Cali and Monterrey suggest different adaptive processes, where non-directly observed structural factors such as informal arrangements, historical rivalries or market reconfigurations determine the spatial distribution of violence (Bright et al., 2017). These differences in adaptive mechanisms explain why similar interventions may generate divergent outcomes in terms of violence, depending on the specific organisational and territorial characteristics of each criminal system.
In this sense, the use of spatial econometric techniques such as SLM and SEM models is essential. These models allow us to observe not only local associations, but also indirect effects and latent spatial dependency structures (Anselin, 1988). The measurement of direct and indirect impacts in this study showed that drug seizures generate significant effects both within the neighbourhoods where they occur and in adjacent neighbourhoods, which reinforces the need to think about security policy from an interdependent territorial logic.
In parallel, intra-urban heterogeneity in the distribution of homicides and their relationship to seizures is evident. Even within the same city, the results show that there are neighbourhoods with highly concentrated dynamics of violence and others with marginal levels of lethality. This variability confirms the validity of the micro-territorial approach adopted in the study and reinforces the idea that the phenomena of violence should be analysed with analytical tools that recognise the multiple scales at which they manifest themselves (Chainey et al., 2019). The neighbourhood unit, in this case, proves to be an optimal scale for capturing both the logistics of concentration and the processes of spatial contagion of violence. However, a future agenda should consider multi-scalar analyses that integrate the effects of regional, metropolitan and transnational variables on local criminal markets (Jiménez-García et al., 2021; Manzano et al., 2020).
Moreover, policy design must consider the adaptive nature of criminal systems. As this study demonstrates, criminal organisations respond to interventions through spatial and functional reorganisation, not disappearance. In contexts where there is a stable form of criminal governance, such as in Curitiba, organisations may rationalise the use of violence as an adaptive strategy for survival (Ferreira, 2019); while in more fragmented settings, such as in Cali, adaptation may be expressed through the intensification of competitive violence (Bright et al., 2017). This suggests that, when designing intervention strategies from the national to the local level, the potential adaptive responses of organised crime should be anticipated, incorporating resilience analysis and the vulnerability of criminal networks.
However, it is important to recognise that, despite these results, there remains a structural limitation to conducting studies to share further findings and design new hypotheses: the scarcity of robust instruments to directly measure the presence and functioning of criminal markets in the urban space. In the absence of systematic data on actors, volumes, prices and territorial control, seizures function as an imperfect, albeit available, proxy that needs to be complemented with other sources of information to achieve a more complete understanding of the phenomenon (Jiménez-García et al., 2023). This methodological shortcoming reinforces the urgency of constructing more precise indicators to operationalise the concept of the criminal market in empirical studies. This is the only way to advance a research agenda that goes beyond the mere association between violence and illegality and focuses on the concrete mechanisms of production, regulation and diffusion of homicidal violence.
The findings of this study open up a line of strategic reflection: how to harness institutional arrangements to discourage the use of homicidal violence as a form of dispute resolution or territorial control. Evidence suggests that certain configurations of criminal governance can reduce the need to resort to overt violence, even if this means accepting logistics of informal order (Lessing & Willis, 2019). In contrast, state interventions without territorial coordination, social intelligence and institutional reconstruction mechanisms tend to amplify conflict. The challenge, therefore, is not to choose between state presence or absence, but to build forms of intervention that transform the structural conditions that make homicidal violence a useful and available tool in many Latin American cities.
Finally, as the annual crime reports show, homicide represents a relatively small fraction of total recorded crime in Colombia. Nevertheless, its lethal character, its strong symbolic charge and, above all, its spatial concentration pattern distinguish it from other more diffuse or distributed forms of crime (such as computer crime or theft). The Tables at the end of this article provide an overview that contrasts with the logic of territorial targeting analysed in this study.
Conclusions
This study has empirically demonstrated that urban homicides in the cities of Monterrey, Cali and Curitiba are not randomly distributed, but rather present highly structured spatial patterns depending on the presence and activity of criminal markets, particularly drug trafficking. The comparative analysis between cities showed that drug seizures, understood as a proxy for the intensity of the illicit market, are significantly associated with the concentration of homicides, although with contextual variations that reveal the importance of institutional, criminal and territorial conditions in each city.
The research confirms that illegal markets do not operate in a vacuum, but rather shape informal governance dynamics with direct implications for the production and diffusion of lethal violence. In this sense, homicidal violence appears as a resource differentially used by criminal actors according to their control structures, rivalries, and state intervention. The finding that optimal patterns of spatial dependency differ between cities -SLM in Curitiba and SEM in Cali and Monterrey- reinforces the idea that the mechanisms of contagion, feedback or structural disruption are not homogeneous, but specific to the type of criminal governance and the degree of institutional fragmentation that exists.
We can also analyse the behaviour of criminal spaces through an analogy with resilient biological systems. Just as certain organisms develop immunity to antibiotics, criminal organisations generate social and logistical adaptations that allow them to adapt to external disturbances (Ayling, 2009). Homicidal violence, from this perspective, could be interpreted as a homeostatic mechanism that these organisations employ to maintain their operations when they perceive threats to their territories or markets. This criminal mimesis does not occur randomly, but follows identifiable patterns of structural transformation, role rotation and selection of strategies optimised for each specific context. Understanding this adaptive intelligence of organised crime is essential to developing security strategies that, in addition to controlling immediate manifestations of violence, address the structural conditions that these organisations systematically exploit to their advantage.
From a methodological point of view, the application of spatial econometric models overcame the limitations of conventional analysis and provided a solid tool to observe both the direct and indirect effects of illicit markets on violence. This approach contributes to a more complex understanding of urban crime patterns and represents a significant advance in the study of the link between illegality and lethality.
However, the study also acknowledges the limitations imposed by the paucity of data on criminal markets per se. The use of seizures as a proxy variable, while reasonable, does not capture the diversity of actors, repertoires of action or forms of territorial control that characterise urban criminal environments. Therefore, the development of more specific indicators and mixed methodologies that integrate spatial analysis with ethnography, fieldwork and analysis of judicial and intelligence sources is proposed as a line of future research.
Finally, it can be argued that homicidal violence in Latin America cannot be effectively addressed without understanding its territorial, organisational and structural underpinnings. Public policies must embrace this complexity and abandon one-dimensional approaches, opting instead for inter-institutional territorial strategies that recognise the spatial logic of criminality, strengthen state legitimacy and discourage the use of violence as an organisational resource. Although homicide represents only a fraction of all recorded crimes, as shown in the statistical appendix at the end of this article, its symbolic intensity, lethality and strong spatial concentration make it a privileged expression for understanding urban criminal ecologies. Only with this type of approach will it be possible to build safer, more equitable cities that are less trapped in the cycles of violence that accompany expanding illegal markets.
