Media (in)visibility of violence in Manguinhos, Rio de Janeiro, Brazil (2020–2024): a statistical modeling approach

Abstract

Introduction:

This study investigates how the visibility of different types of violent events in Manguinhos (Rio de Janeiro) evolved in online media between 2020 and 2024, and compares these patterns with community-reported data from Fogo Cruzado.

Methods:

Two datasets were analyzed: online news articles collected via the Meltwater^® platform (through Arquimedes Social^®) and classified into four categories, police confrontation, homicide/death, drug trafficking, and generic violence, and verified incident records from Fogo Cruzado. Analyses included log–log transformations with power-law assessment to explore long-term patterns, Negative Binomial Regression (NB) to test temporal trends, and Interquartile Range (IQR) smoothing to detect anomalous peaks in coverage. Only the “police confrontation” category showed a statistically significant upward trend in media visibility, with an average growth of approximately 3.6% per month, consistent with power-law dynamics. The other categories exhibited no significant trends, indicating more stable or sporadic reporting. IQR analysis identified five peaks in coverage, likely linked to large-scale police operations or high-profile incidents.

Results:

Findings suggest that police confrontation coverage follows a self-reinforcing cycle shaped by state interventions and exceptional events, while other violence types remain comparatively underreported. This visibility asymmetry underscores how journalistic narratives selectively amplify certain forms of violence, potentially shaping public perceptions, legitimizing specific security approaches, and obscuring structural drivers.

Conclusion:

By integrating quantitative modeling with anomaly detection, this study provides a replicable framework for monitoring media cycles, assessing visibility disparities, and informing debates on representation, accountability, and policy responses in contexts of chronic urban violence.

Introduction

Urban violence is a persistent challenge to social well-being and public policy in Brazil, with particularly acute effects in socioeconomically vulnerable territories such as Manguinhos, in the northern zone of Rio de Janeiro. In these contexts, violence is not only a public safety issue but also a factor that shapes how events are framed in the media, influencing public perceptions, political agendas, and resource allocation. Media narratives can amplify certain types of incidents, such as large-scale police operations, while downplaying others, creating an uneven visibility of violence across different sources of information.

Understanding the dynamics of media coverage is critical for assessing how urban violence is represented, perceived, and acted upon. Prior research shows that the frequency and framing of violent events in the press can affect the political salience of security issues, the legitimacy of policing strategies, and the prioritization of interventions in marginalized areas. Yet, there is limited empirical evidence on how this visibility evolves over time and how it aligns—or diverges—from records produced by grassroots monitoring initiatives.

The choice of Manguinhos as a case study is far from haphazard. A former study by us (Cunha et al., in press) has shown that Manguinhos, located in Rio de Janeiro Administrative Area (AP) 3.2, is a deprived community where violence became structural over decades. As described in detail elsewhere, PA 3.2's monthly shootouts plateaued in 2022 at a relatively lower level than in 2017. The gains in violence reduction during that period were real but bounded. It is possible that the “easy” gains from aggressive policing or gang consolidation had already been realized by 2021, and what remained in 2022 was a resistance to intervention, with persistent drug market activity and community distrust sparked by events like the Jacarezinho massacre nearby. After 2022, data yet to be comprehensively analyzed documented renewed plateauing at unacceptably high levels. Essentially, structural violence became the bread and butter of the local community.

A second, distinctive feature of this territory is that the Oswaldo Cruz Foundation (Fiocruz)—the largest biomedical facility in Latin America—sits within the Manguinhos perimeter. Over the years, various initiatives have been launched from this proximity, with mixed outcomes. Without being exhaustive, two emblematic examples stand out: the innovative sanitation methods implemented by Polish-Brazilian sanitary engineer Szachna E. Cynamon (1925–2007) and community development projects led by American-Brazilian missionary Victor V. Valla (1937–2009). The fact that two expatriates spearheaded major community improvement initiatives during Brazil's military dictatorship is far from coincidental: this was a time when local leadership was dismantled through imprisonment, assassination, exile, and displacement, in a broader context of state neglect and prejudice.

With the return to democracy, new attempts at community development emerged, yet structural violence continued to act as a major deterrent. This negative feedback loop has tangible public health consequences: armed violence intermittently forces the closure of primary health posts—including the Dr. Valla unit located within Manguinhos—and disrupts the work of health professionals (Mota et al., 2025). In a cruel paradox, the very facilities and staff tasked with mitigating the consequences of violence and other vulnerabilities (e.g., hunger, infectious and chronic diseases) find their work undermined by the very conditions they are meant to address.

Manguinhos offers a compelling case study for examining these dynamics. It is a territory marked by recurrent armed confrontations, criminal disputes, and state security interventions, all of which generate fluctuating patterns of journalistic attention. This study draws on two complementary data sources: (i) online news articles collected via Meltwater, a media intelligence platform, and (ii) verified records from Fogo Cruzado, a collaborative platform that monitors armed violence in real time. By analyzing these datasets, we seek to understand how different types of violent events gain or lose visibility over time and how this media attention interacts with the broader context of structural violence in the community.

Despite the extensive literature on urban violence in Brazil, relatively little attention has been paid to how different forms of violence are selectively represented in the media over time. While prior studies have focused on the occurrence, structure, and governance of violence, fewer have examined the dynamics of media visibility as an independent process. This study addresses this gap by analyzing how journalistic attention evolves across different categories of violence, and how these patterns relate to broader cycles of public attention and state intervention.

Guided by these considerations, this study examines the following question: how has the visibility of different types of violence in Manguinhos evolved in online media between 2020 and 2024? We argue that media narratives disproportionately emphasize police confrontations, showing a sustained upward trend over the study period, while other forms of violence appear more sporadically and receive comparatively less attention. Grassroots monitoring data from Fogo Cruzado are used to contextualize these patterns, providing reference points to identify moments of convergence and divergence between media reporting and on-the-ground realities. To asses such question, the remaining of this text will be organized as follows: The next Section describes the construction of the two datasets, including the Meltwater media dataset and the Fogo Cruzado grassroots dataset, and details the classification criteria for four violence categories.Then, we explain the statistical modeling strategy, including a exploratory log–log transformation to examine scaling patterns; power-law regression to detect long-term growth trends; Negative Binomial regression to model overdispersed count data; and Interquartile Range (IQR) outlier detection to identify short-term visibility spikes. Results section presents the temporal trends by category, interprets model coefficients, and examines the contextual role of grassroots data in interpreting media visibility patterns. Finally, the paper discusses the findings in light of agenda-setting and media framing theories, integrating conclusions and outlining implications for policy and future research.

Literature review

The study of urban violence in Brazil has been extensively developed within sociology, criminology, and public security studies, offering a robust body of empirical and theoretical contributions. A central strand of this literature examines the structuring of violence in urban peripheries, highlighting the interplay between state action, criminal organizations, and territorial control.

Michel Misse's work on illegal markets and the social construction of crime provides a foundational framework for understanding how categories of violence are produced, negotiated, and institutionalized within both state and non-state arenas (Misse, 2019). Similarly, Sergio Adorno's research on violence and the criminal justice system emphasizes the persistence of inequality and selective enforcement, demonstrating how patterns of violence are embedded in broader social and institutional structures (Adorno, 2002).

More recent contributions have focused on the dynamics of organized crime and urban conflict. Camila Nunes Dias and Bruno Paes Manso analyze the internal organization of criminal factions and their interaction with state policies, particularly in the context of prison systems and urban territories (Dias, 2013; Manso, 2020). Daniel Hirata's ethnographic work further explores the everyday functioning of illicit markets and the governance of violence in marginalized areas, emphasizing the coexistence of formal and informal regulatory regimes (Hirata, 2021).

Together, these studies show that violence in Brazilian cities is not random, but structured, historically embedded, and deeply connected to both state practices and criminal economies. They also highlight how categories such as “violence” and “confrontation” are not neutral descriptors, but are shaped by institutional, political, and discursive processes.

While this literature provides a comprehensive understanding of the production and organization of violence, less attention has been given to how violence is selectively represented in the media over time. Existing studies suggest that media narratives play a key role in shaping public perceptions of violence, often amplifying specific types of events—such as large-scale police operations—while rendering others less visible.

This paper builds on these insights by shifting the analytical focus from the occurrence of violence to its visibility. Rather than examining violence as a social phenomenon per se, we investigate how different types of violent events gain or lose prominence in journalistic coverage. In doing so, we bridge sociological analyses of violence with approaches from media studies and complex systems, proposing that media attention follows nonlinear and burst-driven dynamics (Barabási, 2010).

By combining statistical modeling with anomaly detection, this study contributes to the literature by offering a quantitative framework to analyze how media visibility evolves over time, complementing existing qualitative and ethnographic approaches.

Data

This study is based on two independent but complementary datasets that capture violence in Manguinhos (Rio de Janeiro) between January 2020 and March 2024. The first dataset comprises online news coverage obtained via the Meltwater^® platform, focusing on journalistic visibility. The second is drawn from Fogo Cruzado, a collaborative platform that monitors and geolocates armed violence events. Together, these sources allow us to investigate both the visibility and frequency of violent incidents in the region.

It is important to distinguish between data and methodology in this study. The data consist of two independent sources: (i) media coverage collected through the Meltwater^® platform, and (ii) verified incident records from Fogo Cruzado. The methodology refers to the analytical procedures applied to these datasets, including statistical modeling and anomaly detection techniques designed to identify both long-term trends and short-term fluctuations in media visibility.

Online media: meltwater^®

The online news dataset was collected using Meltwater^®, a media intelligence platform that continuously crawls and indexes digital content from thousands of news websites, blogs, and social media sources. The system identifies relevant materials through automated keyword searches, using Boolean logic to refine the scope of results.

For this study, Meltwater was configured to search for articles that mentioned terms related to violence (e.g., “shooting,” “homicide,” “drug trafficking”) in combination with geographical references to Manguinhos. The platform's search engine identifies articles in real time as they are published and stores both article metadata and full content in a structured database.

Through a partnership with Arquimedes Social^®, which has licensed access to the Meltwater platform, we obtained a curated dataset of articles published between January 2020 and March 2024. This dataset includes the title, body text, publication date, and source URL of each article, all of which were preprocessed for classification and modeling in later stages of the analysis.

Fogo Cruzado

(Fogo Cruzado 2024) (in plain English, crossfire) is a civil society initiative and open-data platform that monitors gun violence in real time across major Brazilian metropolitan areas. It aggregates data from user-generated reports, media monitoring, and institutional sources. Submitted events are verified through triangulation before being included in the database. Each entry typically contains the date, time, geographic location, and a brief description of the incident.

Although the platform previously maintained a public-access interface for querying and visualizing events, its website was offline during the preparation of this study. Data used here were provided directly by two of the study's co-authors (JCM and APC), who had access to the internal dataset through prior collaboration with the Fogo Cruzado team. The dataset includes all verified incidents that occurred within the Manguinhos perimeter between January 2020 and March 2024.

Methodology

Understanding which types of violence are more prominently covered and how this visibility evolves over time is essential to analyzing media attention dynamics.

Prior to applying modeling techniques, we classified news articles into four categories of violence: police confrontation, death/homicide (not originating from major intervention by police special forces), drug trafficking, and generic violence (i.e., not resulting in deaths, but rather violent threats, wounds etc.). An exploratory data analysis was then conducted using log–log transformations and power law distributions, which provided insight into the underlying mathematical and statistical properties of the data. Based on these findings, we applied and compared two statistical trend modeling approaches: Negative Binomial Regression (NB) and Interquartile Smoothing (IQR). Statistical findings, modeling results, and interpretative discussion will be presented in the following sections of this article.

Data extraction

The data extraction process was conducted separately for each of the two sources—online media (via Meltwater^®) and Fogo Cruzado—using tailored criteria and filtering procedures to ensure relevance, precision, and consistency across the datasets.

Online media: Meltwater^®

News articles were retrieved using Meltwater's Boolean search engine, which enables complex keyword queries to identify relevant content from thousands of digital outlets. To capture media coverage of violent events in Manguinhos, the following query was used in Portuguese:

Q1 – News

(“Manguinhos” OR “Complexo de Manguinhos”) AND (“violência” OR “segurança pública” OR “tiroteio” OR “confronto” OR “polícia” OR “milícia” OR “tráfico” OR “bala perdida” OR “operações policiais” OR “UPP” OR “crime organizado”)

This query targeted articles that mentioned either “Manguinhos” or “Complexo de Manguinhos” in combination with at least one term related to violence or public security. The search was restricted to the Portuguese language and to the time period between January 1, 2020, and March 31, 2024.

The resulting Meltwater dataset contains structured metadata for each article, including: publication date, headline, news outlet, article URL, opening text, matched sentence (hit sentence), estimated reach, and automated sentiment score. Additionally, each article was manually classified into one of four violence categories—police confrontation, homicide/death, drug trafficking, or generic violence—based on its content. This structure enabled longitudinal visibility analysis and comparative modeling in later stages of the study.

Figure 1, illustrates the distribution of news articles by violence category over time, aggregated by month. The most frequently represented category throughout the period is police confrontation, followed by generic violence. Periodic spikes in coverage, particularly visible in early 2021 and late 2022, correspond to clusters of events that received high media attention. In contrast, homicide/death and drug trafficking appear less consistently, often with isolated peaks. It highlights fluctuations in journalistic visibility and reinforces the dominance of certain event types in the public narrative around violence in Manguinhos.

Figure 1

Fogo Cruzado

The Fogo Cruzado dataset used in this study was obtained directly from two co-authors, who previously collaborated with the organization and had access to its verified internal records. These records were extracted from the broader Rio de Janeiro metropolitan dataset and filtered to include only those incidents whose geographic coordinates or textual descriptions referenced the neighborhood of Manguinhos. The final dataset spans the period from January 2020 to March 2024.

Each entry in the dataset corresponds to a unique incident of armed violence and includes a wide array of structured variables. Key fields include:

• Date and time of occurrence.

• Location (both free-text and georeferenced).

• Presence of police or security agents.

• Number of civilians and security agents killed or injured.

• Contextual tags (e.g., stray bullet, residence invasion, school/transport hub proximity).

• Victim demographics, including children, adolescents, and the elderly.

The dataset also classifies each event by a motivo principal (primary motive), such as police operation, criminal dispute, or mass shooting, among others.

Figure 2, presents a stacked bar chart of Fogo Cruzado events in Manguinhos, disaggregated by year and primary event type. Although the absolute number of events remains relatively low across all years compared to metropolitan totals, 2022 and 2023 show a notable increase in the number and diversity of recorded motives. Police operations remain the most frequently tagged category, followed by criminal disputes between different factions and events with unspecified or unclassified motives. The year 2020 displays the fewest entries, likely reflecting underreporting or more limited coverage in the early phase of the dataset, which roughly coincides with the emergence of the COVID-19 epidemic, with devastating effects in Brazil and around the world. As happens everywhere in the world, the COVID-19 epidemic (declared by the World Health Organization a pandemic on 11 March 2020) had major impacts on the most different information systems (Ågerfalk et al., 2020).

Figure 2

This localized dataset provides a robust benchmark for comparing journalistic coverage patterns and enables an analysis of visibility gaps across different sources of violence reporting.

Statistical modeling and findings

To analyze how different forms of violence are portrayed over time in online media coverage of Manguinhos, we adopt a two-phase analytical strategy:

• Log–log transformation and exploratory analysis of monthly media visibility;

• Statistical modeling based on power-law underlying distribution (or lack of), followed by time series modeling to detect long-term trends and fluctuations.

• Tables 1–3 summarize descriptive statistics and analytical. procedures

This approach allows us to investigate both the distributional properties and the temporal dynamics of journalistic attention to violence.

Logarithmic transformation and monthly distribution

We begin by aggregating all articles from the Meltwater database into monthly counts by violence category. A logarithmic transformation [log₁₀(x + 1)] was then applied to each time series to facilitate the identification of scaling behavior and volatility.

Figure 3 shows the log-transformed monthly frequency of news articles per category. Police confrontation stands out with the highest and most volatile coverage, while homicides, drug trafficking, and generic violence exhibit more irregular and lower-intensity patterns.The descriptive statistics below reinforce this interpretation:

Figure 3

The high coefficient of variation in homicide and trafficking categories indicates erratic coverage, with several months registering no mentions. In contrast, police confrontation not only has the highest average visibility but also displays significant burstiness, with monthly counts ranging from 1 to over 100 articles.

Power-law temporal analysis

To assess whether media attention follows temporal power-law dynamics, we applied linear regressions to the log-transformed time series. The independent variable was time (measured in months), and the dependent variable was the log₁₀ of article count per category.

Only police confrontation shows a statistically significant positive slope (p < 0.05), suggesting that its visibility has increased over time in a log-linear pattern–consistent with temporal power-law dynamics. The remaining categories do not show clear or statistically meaningful trends in log space, which suggests that their visibility patterns are more random, irregular, or diffuse.

Modeling media visibility of police confrontation

Given the results from the log–log regression analysis, we observed that only the category police confrontation exhibited a statistically significant growth trend over time in logarithmic scale. This suggests that journalistic attention to this type of violence follows a pattern compatible with temporal power-law dynamics–with increasing and concentrated visibility.

The other categories (homicide, drug trafficking, and generic violence) did not demonstrate significant or consistent trends. Their irregularity, low frequency, and statistical insignificance indicate that more advanced temporal modeling would not yield meaningful or interpretable patterns at this stage.

Therefore, in the following analysis, we focus exclusively on the category police confrontation, applying two complementary modeling techniques.

Negative Binomial regression (NB)

To statistically model the increase in journalistic attention to police confrontation over time, we applied a Negative Binomial Regression (Hilbe, 2011) to the monthly article counts for this category. This model is appropriate for count data exhibiting overdispersion, where the variance exceeds the mean—an expected characteristic of media coverage dynamics.

We specified the model as:

Where:

• y_t is the number of articles in month t,

• μ_t is the expected article count,

• β₀ is the intercept, and

• β₁ represents the temporal effect (i.e., trend).

The positive and statistically significant coefficient for time (β₁ = 0.035, p < 0.001) indicates a consistent growth trend in the number of news articles on police confrontation throughout the study period.

In exponential terms, the monthly growth rate is exp(0.035)≈1.036, suggesting that the expected number of articles increases by approximately 3.6% each month.

The strength and significance of the trend support the findings from the exploratory log–log analysis, reinforcing that police confrontation visibility is not random, but follows a compounding trajectory.

This result suggests the presence of nonlinear escalation dynamics, potentially tied to cycles of state intervention and heightened media attention in response to disruptive events. In the next section, we complement this long-term modeling with Interquartile Smoothing (IQR) to detect and interpret anomalous bursts in coverage.

Interquartile smoothing (IQR)

To complement the long-term modeling performed via Negative Binomial regression, we applied an Interquartile Range (IQR) outlier detection procedure to identify short-term anomalies in the journalistic coverage of police confrontation. This method is useful for detecting spikes in visibility that may correspond to specific high-impact events or media cycles not captured by linear trends.

We calculated the first quartile (Q1), third quartile (Q3), and the interquartile range (IQR = Q3 − Q1) of the monthly distribution of article counts. Outliers were defined as months in which the count exceeded Q3 + 1.5·IQR or fell below Q1 − 1.5·IQR. The results are visualized in Figure 4.

Figure 4

The IQR smoothing analysis reveals five prominent outlier peaks, where the number of articles covering police confrontations in Manguinhos sharply exceeded the typical monthly distribution. These peaks may be associated with:

• Large-scale police operations,

• Mass casualty events, or

• Highly visible incidents with political or media ramifications.

Importantly, these anomalies signal moments of media overexposure, where visibility diverges from routine coverage patterns. While NB regression captured the structural growth over time, IQR highlights the episodic nature of visibility spikes, offering a more nuanced view of media attention dynamics.

This dual approach–NB for trend modeling and IQR for shock detection–provides a robust framework to understand both the systemic escalation and exceptional surges in media coverage of urban violence.

Results

The combined analytical strategy employed in this study allows for a detailed understanding of how different types of violent events gain visibility in online media over time in Manguinhos. The log–log transformation and exploratory modeling demonstrated that only one category—police confrontation—exhibited a statistically significant temporal trend compatible with a power-law distribution, with a monthly growth rate of approximately 3.6

In contrast, other violence categories—such as drug trafficking, homicide, and generic violence—showed no significant temporal trends in log-scale, indicating that their occurrence in the media is more sporadic or evenly distributed, lacking the compounding growth structure that characterizes police confrontations.

The Negative Binomial (NB) regression further reinforced these findings, revealing a strong, statistically significant upward trajectory for media coverage of police confrontations, with time as a key explanatory variable. This supports the hypothesis that this specific type of violence has increasingly dominated the public narrative, possibly driven by cycles of state intervention and community-level escalation.

To capture more transient fluctuations, Interquartile Smoothing (IQR) was applied to identify anomalous peaks in monthly article counts. Five months stood out as statistical outliers, suggesting moments of media saturation or exceptional incidents. These spikes likely reflect large-scale police operations, violent confrontations with high death tolls, or controversial episodes that temporarily boosted visibility well beyond average levels.

The dual approach—combining long-term trend modeling with short-term anomaly detection—offers valuable insights into how media visibility does not simply mirror the occurrence of violence, but rather follows its own logic of amplification, repetition, and disruption. In particular, the alignment (or lack thereof) with the Fogo Cruzado data set will be explored in subsequent stages of analysis to assess how journalistic coverage compares with real-world event reporting.

Discussion

Our findings are of great concern. The structural violence experienced daily by the Manguinhos community, as captured by the Fogo Cruzado dataset, remains largely ignored by both the media and the broader public. To interpret this pattern, we draw on Hannah Arendt's concept of the “banality of evil,” not in its original historical sense, but as a lens to understand the normalization of violence in marginalized urban contexts.

In Manguinhos, everyday violence—threats, killings, domestic abuse, and other forms of harm—rarely becomes visible unless it escalates into catastrophic events, such as major confrontations between police forces and heavily armed criminal factions. Outside of these exceptional moments, violence is rendered ordinary, part of the “normal” life of a vulnerable and disenfranchised population. In this sense, there is a banality not only in the perpetration of violence, but also in its systematic invisibility.

This dynamic is not new. Decades of sociological research have shown that urban violence in Brazil is deeply embedded in broader processes of marginalization and uneven state presence. The work of Anthony and Elizabeth Leeds, for instance, emphasized the importance of fostering community development rather than treating marginalized territories as spaces to be controlled or eliminated. Their findings demonstrated that sustainable social development was both possible and already occurring, even in the absence of consistent state support.

Subsequent policies, such as community policing initiatives and later the UPPs (Police Pacifying Units), sought to address these structural challenges. However, as widely documented, these efforts were discontinuous and ultimately replaced by intermittent cycles of violent confrontation between state forces and criminal groups. Rather than resolving structural violence, such interventions often reinforced its episodic and highly visible manifestations.

These cycles of confrontation help explain the patterns identified in our data. Violent events do not occur—or, more importantly, are not reported—in a continuous and proportional manner. Instead, they follow what (Barabási 2010) describes as bursty dynamics: long periods of relative invisibility punctuated by sudden, high-intensity events that attract disproportionate attention.

A recent large-scale police operation illustrates this pattern. In a single day, an intervention involving thousands of officers resulted in over one hundred deaths and triggered widespread media coverage, social media reactions, and international attention. Events of this magnitude temporarily dominate the public narrative, while the everyday violence experienced in places like Manguinhos remains largely unreported.

In analytical terms, the temporal dynamics observed in our dataset reflect this structure. Media visibility does not evolve as a smooth trajectory that could be adequately approximated by polynomial functions of time. Instead, it combines a gradual long-term escalation with sudden spikes of attention triggered by high-impact events. This dynamic can be formally described as:

where the exponential component captures the long-term increase in journalistic attention, while the burst component represents episodic shocks associated with major confrontations or large-scale police operations. Because these bursts are inherently discontinuous, they cannot be adequately represented by polynomial functions, which assume smooth and continuous variation over time.

This helps explain why the suffering of marginalized communities remains largely invisible in public discourse. After each major event subsides, everyday life resumes under conditions that remain unchanged, yet largely unreported. As such, media attention does not mirror the lived reality of violence, but rather follows its own logic of amplification and disruption.

Few lines capture this condition as poignantly as Emily Dickinson's reflection on pain:

• Pain – has an Element of Blank –

• It cannot recollect

• When it began – or if there were

• A time when it was not –

Conclusion

This study contributes to the understanding of media attention dynamics surrounding urban violence in Brazil by examining how different forms of violence gain visibility over time. Focusing on Manguinhos (Rio de Janeiro) between 2020 and 2024, and combining news data from Meltwater^® with community-reported data from Fogo Cruzado, we identify systematic asymmetries between the occurrence of violence and its representation in the media.

Our findings show that most forms of violence remain either stable or largely invisible in journalistic coverage, reinforcing processes of normalization and neglect. In contrast, police confrontations emerge as a distinct category characterized by both exponential growth in coverage and episodic spikes of attention. This pattern suggests that media visibility follows a nonlinear dynamic driven by a combination of long-term escalation and burst-like events, rather than reflecting the continuous reality of violence experienced in marginalized communities.

Methodologically, this study models media visibility as overdispersed count data, using Negative Binomial regression to estimate long-term temporal trends and capture systematic growth patterns. This approach allows us to identify statistically significant differences across categories, demonstrating that only police confrontation exhibits a consistent upward trajectory over time. To complement this analysis, we apply Interquartile Range (IQR)–based outlier detection to identify short-term anomalies in media coverage, isolating high-intensity events that generate disproportionate attention.

Taken together, these approaches allow us to characterize media attention as the combination of a gradual exponential trend and irregular bursts of visibility. Importantly, this structure cannot be adequately captured by polynomial specifications, which assume smooth and continuous variation over time, and therefore fail to represent the discontinuous and shock-driven nature of media dynamics.

These results indicate that media attention does not proportionally mirror lived experiences of violence, but instead operates through selective amplification mechanisms that privilege high-impact, state-centered events. As a consequence, public perception and policy debates may become disproportionately shaped by episodic interventions, while structural and everyday forms of violence remain underrepresented.

The proposed framework offers a replicable approach for analyzing visibility dynamics, journalistic bias, and attention cycles in contexts of chronic violence. Future research should further explore the relationship between reported and experienced violence by systematically comparing media data with alternative sources such as community-based monitoring systems. Such efforts are essential for advancing debates on representation, accountability, and the role of media in shaping responses to urban violence.

References

• 1

  ÅgerfalkP. J.ConboyK.MyersM. D. (2020). Information systems in the age of pandemics: COVID-19 and beyond. Eur. J. Inf. Syst. 29, 203–207 doi: 10.1080/0960085X.2020.1771968

  • CrossRef
  • Google Scholar

• 2

  AdornoS. (2002). Exclusão socioeconômica e violência urbana. Sociologias4, 84–135 [in Portuguese]. doi: 10.1590/S1517-45222002000200005

  • CrossRef
  • Google Scholar

• 3

  BarabásiA.-L. (2010). Bursts. London: Penguin Books.

  • Google Scholar

• 4

  Cunha A. P. Mota J. C. Bastos F. I. (in press). A city under siege: time-series analysis of shootouts the association with health living conditions in Rio de Janeiro, Brazil, 2017 to 2022. J. Interpers Violence. In press. 10.1177/08862605251356234

• 5

  DiasC. C. N. (2013). PCC: Hegemonia nas Prisões e Monopólio da Violência. Sao Paulo: Saraiva Jur.

  • Google Scholar

• 6

  Fogo Cruzado (2024). Fogo Cruzado Database: Armed Violence Incidents in Rio de Janeiro Metropolitan Area, 2016–2024. Available online at: https://fogocruzado.org.br/ (Accessed March 31, 2024).

  • Google Scholar

• 7

  HilbeJ. M. (2011). Negative Binomial Regression, 2nd Edn. New York, NY: Cambridge University Press. doi: 10.1017/CBO9780511973420

  • CrossRef
  • Google Scholar

• 8

  HirataD. V. (2021). Sobreviver na Adversidade: Mercado e formas de vida. São Carlos, SP: EdUFSCar.

  • Google Scholar

• 9

  MansoB. P. (2020). A República das Milícias: Dos esquadrões da morte à era Bolsonaro. São Paulo. Todavia [in Portuguese].

  • Google Scholar

• 10

  MisseM. (2019). The puzzle of social accumulation of violence in Brazil: some remarks. J. Illicit Econ. Dev.1, 177–182. doi: 10.31389/jied.32

  • CrossRef
  • Google Scholar

• 11

  MotaJ. C.CunhaA. P.BastosF. I. (2025). Armed violence and shootouts in Rio de Janeiro: relevance to impoverished communities and primary care health services. Ciên. Saúde Colet. 30:e1298202. doi: 10.1590/1413-812320253010.12982025

  • CrossRef
  • Google Scholar