Shootings across the rural–urban continuum

Abstract

In this article, we investigate situations involving firearm violence in Sweden. The spatiotemporal distribution of records is assessed in different contexts across the country and linked to land use, demographic, and socio-economic characteristics by area and by street segment. The findings emphasize the prevalence of evening shootings, particularly in economically disadvantaged areas where young people congregate in public places often in residential areas, in parks, in playgrounds, and in transit areas. Although two-thirds of shootings took place in larger urban municipalities, our study sheds light on the non-uniform distribution of gun violence along the rural–urban continuum. We conclude by offering suggestions for future research and practical interventions to address this pressing issue that negatively affects people and communities.

Introduction

Firearm violence is a rare phenomenon. Yet, firearm incidents have the potential to cause death and inflict long-lasting physical and psychological harm upon victims and their families and friends, with implications for the whole community (Wilhelmsson et al., 2022). Historically, shootings have been associated more with metropolitan areas than rural communities (Evans & Taylor, 1995; Hemenway et al., 2020; Sturup et al., 2018). Drawing from environmental criminology, this article reports on daily, weekly, and seasonal patterns of firearm violence across the rural–urban continuum in Sweden. Using Geographical Information Systems (GIS) and regression models, the study assesses land uses that have a significant predictive role for shootings after controlling for area characteristics, and socioeconomic and locational factors across different types of municipalities. The ultimate goal is to identify potential commonalities and differences in the situational conditions of firearm incidents, which can be valuable for research and crime prevention practices.

Sweden constitutes an interesting case study. First, although the rate of lethal violence has been largely stable in the last ten years, the country has experienced an increase of more than 250% in gun violence (BRÅ, 2022). Half of this has occurred in smaller and medium-sized municipalities compared with one-third of these incidents occurring a couple of years ago, with a significant increase in violent encounters involving young males (Sturup et al., 2019). Small and remote municipalities vary significantly in terms of their criminogenic conditions, economic activities, population density, and access to services, in the same way that urban areas can differ in terms of size, and population (Ceccato & Dolmen, 2011). Second, most of the literature about firearm violence is both urban-centric and dominated by North American literature. Sweden is a sparsely populated country with a relatively low overall population density (26 persons per km²) compared to the European average (109 persons per km² in 2018) (Eurostat, 2022). A rural–urban continuum is an appropriate framework for analysing firearm violence. There are dynamic movements of individuals, activities, and commodities where criminal activities occur. In an interconnected, globalised world, these locations can exhibit rural characteristics in certain aspects and urban characteristics in others. According to Dewey (1960), differences along the rural–urban continuum derive from the fact that variations in population size and density induce variations in anonymity, division of labour, informally and formally formed relationships, and symbols of status which are independent of personal acquaintance. All these factors can significantly affect the criminogenic and safety conditions of such areas (Ceccato & Abraham, 2022).

In this study, we count “every bullet fired” as an incident since every bullet fired increases the risk for the victims of the shootings. Firearm discharges may result in deaths, other injuries, and, in other cases, property damage, in some cases, all three types of harm. We use “land use” as a general term to represent land that is allocated and utilised with a wide range of activities and functions within the built and natural environments, including residential, commercial, recreational, educational, and industrial uses, as well as parks, forests, and points of interests (POIs), such as schools, stations, and playgrounds. We recognise that different land uses affect social dynamics and people’s routine activity, thus influencing the occurrence of shootings across both urban and rural–urban settings.

Theoretical background: the context of lethal violence in public places

Environmental criminology theory can offer valuable insights into understanding violence by examining the spatial and situational factors that contribute to criminal behaviour. There are locations that attract criminals and those that generate criminal opportunities, as they may attract individuals with violent tendencies, while others may provide opportunities for disputes to escalate into shootings (Kinney et al., 2008). At neighbourhood level, limited economic resources, cultural diversity, frequent residential changes, and family breakdowns contribute to a breakdown in community cohesion, ultimately resulting in higher rates of crime and juvenile delinquency (Shaw & McKay, 1942). A lack of social cohesion and/or collective efficacy (Sampson, 2017) leads to the presence of disorderly conditions where shootings are more likely to occur. At the micro level, understanding individuals’ daily routines and activities (Cohen & Felson, 2010) can help identify when and where confrontations leading to shootings are more likely to occur, as well as the neighbourhood contexts. Drawing from environmental criminology, we discuss the current evidence about temporal and spatial patterns of firearm violence in the international literature focusing on the role of land use in both urban and rural areas.

Daily, weekly and seasonal patterns of lethal violence

There is a temporal variation in lethal violence, which is not limited to gun violence only [see for instance, Hakko et al. (1998)]. McDowall and Curtis (2014) found a peak of homicides during the summer in large US cities while Oliphant (2021) shows that such temporal variation depends on where the incident happens with milder winters being associated with more winter shootings (for a review see Ceccato and Westman, 2024). Several studies found that shootings were significantly associated with on-premises alcohol outlets during the night only (Erickson et al., 2015; Morrison et al., 2017). In Sweden, it has been shown that shootings between 2011 and 2015 were fairly evenly distributed across weekdays, with Sundays being the most common (16%) and Wednesdays the least (12%). There were also more shootings in the summer (29%) than in the winter (22%) (Sturup et al., 2018).

Gun violence exhibits near-repeat patterning, meaning that after a shooting there is an elevated risk for a follow-up shooting within a short period (typically a few weeks) and a short spatial distance (typically a few hundred meters), a pattern that is suspected of being driven by conflicts and retaliations (Ratcliffe & Rengert, 2008). Papachristos et al. (2015) investigated the concentration of non-fatal gunshot injuries within risky social networks in Chicago and found that 70% of all non-fatal gunshot victims were located in co-offending networks comprising less than 6% of the city’s population. The near-repeat patterning of gun violence appears to be particularly strong in Sweden (Gerell et al., 2021).

Spatial patterns: land use and neighbourhood context

The geography of firearm violence reflects the activity spaces and daily routines of the victim. Brolan et al. (2016) found that the most common locations for a hit to take place were related to the homes of the target(s), whether the location was on the doorstep or actually within the home of the victim. The next most common location was in an enclosed public space, such as a restaurant, pub, or in club. In New Zealand, assaults (including lethal violence) and almost half the homicides took place around private homes (Langley et al., 1996). Alejandro Giménez-Santana et al., (2018) also found that in Bogotá, Colombia, the highest risk for homicide was detected close to environments known by the victim, such as drugstores, public schools, medical clinics, community kitchens, and public hospitals. In Boston, in the US, Barboza (2018, p. 356) found that “gun violence clusters in neighbourhoods on blocks within short distances from schools. The analysis revealed that up to six times as many shootings occur within 400 m of schools than would be expected if the shootings were located around the city in a way unrelated to schools”. In St. Louis, USA, Fox et al. (2021) also found that homicides were associated with ATMs and convenience stores. Connealy and Piza (2019) found that in Denver, USA, the presence of convenience stores was significant for gun violence as well as robbery and aggravated assault. Things that attracted firearm crime were identified in the study by Muggy et al. (2022) as hotels/motels, alcohol outlets, and restaurants in New Orleans, and rail transit stations in Los Angeles. Oliphant (2021) found that the connection between on-premises alcohol outlets and shootings was observed only during the night and no connection was detected between schools and shootings.

International evidence shows an overall increasing trend in the association between lethal violence and the presence of bars or licensed premises, though evidence seems to depend on the types of premises and types of violence, and times of the day. Alcohol outlets were statistically associated with the incidence of homicides (Jennings et al., 2014), but proximity to alcohol outlets was mostly unrelated to the risk of gun assault (Morrison et al., 2017). In cases in New Zealand, Langley et al. (1996) suggested that a large share of homicide cases occurred in or around licensed premises, but the authors warned that such associations have decreased over time. In his study of alcohol outlets and firearm violence in Philadelphia, USA, Jay (2020) found that beer stores within one block and bars/restaurants on the same block were associated with the increased risk of shootings. The authors found that a late-night alcohol licence was a significant spatial correlate of gun-related deaths in Texas, USA. Many bars and other licensed premises tend to result in an increase in lethal violence in the area. Britt et al. (2005) found that the value of alcohol outlet density would result in an increase in the number of violent acts in the neighbourhood of five crimes per 1000 individuals per year.

Even for drugs that are not associated with violence such as cannabis, the evidence also indicates links between the location of marijuana dispensaries and lethal violence. Hughes et al. (2020) also revealed a statistically significant positive interaction between medical marijuana dispensaries and murder and aggravated assault in Denver, USA. Contreras (2017) also indicated that marijuana dispensaries were associated with changes in violent crime rates, in particular homicides and robberies. In Sweden, cannabis trade locations were associated with places of violence and open drug markets in the municipality and region of Stockholm (Ceccato et al., 2023). It has also been shown that places where drugs are sold openly have elevated levels of gun violence in Sweden, especially in combination with “vulnerable neighbourhoods”, deprived neighbourhoods where gangs are active (Magnusson, 2020). At drug market locations in such neighbourhoods, there are 42 times more shootings per square kilometre as in the rest of the city of Malmö for instance (Gerell et al., 2021). In the US, Semenza et al. (2022) investigate the role of firearm dealers in shootings in relation to disadvantaged neighbourhoods. They found that disadvantage influences the relationship between firearm dealers and shootings in that dealers significantly attract shootings in disadvantaged neighbourhoods, while in non-disadvantaged communities, shootings are located further from gun stores.

However, the urban landscape provides crime detractors, in this case against lethal violence. Fondevila et al. (2021) found that the rate of homicides increased as the distance from a police station increased, up to 600 m. However, the frequency of homicides then decreased the more the distance increased beyond 600 m. The average distance between stations was 1035 km, thus the highest homicide rate seems to occur half way between neighbouring police stations in Buenos Aires, Argentina. Clery et al. (2020) reported that in Detroit, over 50% of violent crimes occurred within 5 km of a hospital, and 99% within 10 km. In contrast, Levine et al. (2021) found no significant relationship between hospital proximity and firearm homicides in Harris County. However, in Bogotá, proximity to medical clinics and public hospitals correlated with increased homicide rates (Giménez-Santana et al., 2018). Kondo et al. (2017) found that being under tree cover was inversely associated with gunshot assault, especially in low-income areas, after controlling for other neighbourhood characteristics. Muggy et al. (2022) identified several firearm crime repellents, including universities/colleges and public/private schools.

Since the Chicago school and social disorganisation theory (Hardyns & Pauwels, 2017; Shaw & McKay, 1942), income has been used in many studies as an indication of poor socio-economic conditions, although its overall effect on violence has often been ambiguous. Recently, Inlow (2019) investigated how land use affects the frequency of homicides in Portland, USA, and found that for every 1% increase in mixed-use residential zoning per kilometre, the logged frequency of homicide increased by 0.016, compared with single-family residential and controlling for all other neighbourhood demographic and socio-economic variables. Barton et al. (2021) demonstrated that in Baton Rouge, USA, the physical environment was a significant predictor of neighbourhood homicide, in particular vacant housing. People who live in public housing are more vulnerable. The probability of a homicide occurring is significantly greater in an internal type of environment when the victim is a resident of public housing (Jennings et al., 2014). In St. Louis, USA, Fox et al. (2021) found that homicides were associated not only with vacant buildings and public housing but also with a few other land use characteristics. In Stockholm, Sweden, gun violence tends to be more concentrated in locations with high population density, high proportions of foreign-born residents, and lower rates of employment than in locations that have not previously had any gun violence (Magnusson, 2023). It has also been shown that gun violence locations tend to have higher levels of crime (Magnusson, 2023), and, more specifically, that there is a spatial association between hand grenade detonations and shootings (Sturup et al., 2020). For a complete review of the correlates of firearm violence, see Ceccato and Westman (2024).

Firearm violence across the rural–urban continuum

Knowledge of where shootings occur is a critical component of firearm violence prevention efforts as it enables law enforcement, community organisations and policymakers to target resources and interventions effectively, address the causes of violence, and ultimately reduce impact on communities (Nance et al., 2002; Reeping et al., 2023; Schleimer et al., 2023). As illustrated in the previous section, current knowledge on lethal violence caused by firearms is dominated by evidence from North American metropolitan areas; some of it is outdated (Kowalski & Duffield, 1990; Nance et al., 2002), and tend to be by county type (Reeping et al., 2023), with little focus on rural areas. Others more specifically study the specific aspect of firearm violence. For example, Schleimer et al. (2023) found that adolescent involvement in violence correlated with the carrying of handguns across the rural–urban continuum. However, they suggest that the strategies for mitigating the risks associated with the carrying of guns may vary between rural and urban communities. Despite the pervasive nature of gun violence, the proportion of the rural population in a county has an inhibitory effect on homicide. Yet, previous research rarely explores the temporal dimension of these incidents, and no studies have assessed in detail the influence of different types of land use on firearm violence at the street segment level across the rural–urban continuum.

This study asks four research questions:

• 1. What is the spatio-temporal distribution of shootings in Sweden?

• 2. Which land uses and socio-economic characteristics of neighbourhoods are associated with firearm violence?

• 3. Does firearm violence vary across the rural–urban continuum?

• 4. What do these patterns mean for the prevention of firearm violence?

Sweden as a case study

According to the Swedish National Council for Crime Prevention (2017), close to 40% of all homicides in Sweden are committed with a firearm, and a considerable number of individuals are injured and wounded in firearm-related violence (Khoshnood, 2018; Sturup et al., 2019). The shooters tend to be male, to have a low education, have been previously convicted of crimes, and have a drug disorder (Khoshnood et al., 2023).

Shootings are commonly connected to conflicts within or between criminal networks. The conflicts often arise out of narcotic market disputes, but personal motivations can also be a cause. A key insight from interviews with people involved in firearm violence in Sweden is that the conflicts are more important in themselves than the reasons for the conflicts. Once a conflict is in progress, everyone involved needs to respond to the slightest sign or risk of getting a worse reputation and a lower status in the criminal environment (Jönsson & Nilsson, 2019). This may, in part, explain why Swedish gun violence exhibits much stronger near-repeat patterning than gun violence in the US (Sturup et al., 2018; Gerel et al., 2021). Shooting someone can be seen as an investment in a criminal career. It is perceived as strengthening the shooter's reputation and position in the criminal environment and can, therefore, be of importance in itself without taking into account the immediate reason why it was committed (Jönsson & Nilsson, 2019).

Out of Sweden’s 10.4 million inhabitants, approximately 1.2 million reside in rural areas (Macrotrends, 2022). Sweden has 290 municipalities (Fig. 1). To characterise the rural–urban continuum, we adopted the classification of the Swedish Association of Local Authorities and Regions (SALAR), as illustrated in Table 1. Most of the largest urban areas are located in the central-southern parts of the country, with the most remote rural areas in Northern Sweden (Fig. 1).

Fig. 1

Swedish municipalities in the study area. The municipalities with shootings (in grey) are split by type: urban, accessible rural, and rural

Table 1 Groups of municipalities according to the Swedish Association of Local Authorities and Regions—SALAR

Most of the population is concentrated in the southern half of the country, which also encompasses the most accessible rural areas. While rural areas have become more prone to crime in the past few decades, urban areas still exhibit higher crime rates across all categories of offences (Ceccato & Abraham, 2022). Figure 1 shows the location of the shootings in the 290 municipalities and the concentrated pattern in the three major cities in Sweden, Stockholm the capital, Gothenburg and Malmö.

Data and methods

The database used in the study is described below and in detail in Appendix, following the analytical strategy shown in Figs. 2 and 3.

• Recorded firearm incidents: The data were obtained from November 2016 to March 2020 from the National Operations Department (NOA) of the Swedish Police Authority following the data protection and privacy laws. The database contains information on the date of each shooting, the time of the day, the type of injury, and the geographical coordinates.

• Geographical data from Sweden’s DeSOs (Demographic statistic areas): is the smallest unit of analysis in Sweden and is a nation-wide breakdown that follows regional and municipal boundaries. The DeSO system divides Sweden into areas, each of which initially have between 700 and 2700 residents. The division takes account of geographic conditions to ensure boundaries align with streets, waterways, and railways as much as possible. We obtained demographic and socioeconomic data from Statistics Sweden (SCB) at the DeSO level for 2018 and 2019. The geographical data are from 290 municipalities, 4789 DeSO areas and 542,679 street segments (Fig. 2).

• Land-use data: The land-use data come from Open Street maps and the Open Data Portal. These include data from residential, industrial, commercial and retail sources, as well as parks, bus stops, playgrounds, restaurants, fast food establishments, convenience stores, supermarkets, cafés, car parks, kiosks, ATMs, bars/pubs and premises selling alcohol, schools, kindergartens, playground areas, and railway stations. All Points of Interest (PoIs) within each DeSO area are counted and land use areas are calculated as a proportion of their own DeSO area. The central point of the municipality was, as a rule, the central station (railway or bus) located in the municipality. Measures of distance from shootings to PoIs are calculated. For the model of street segments, data from The Swedish Traffic Authority were used for street segments. Note that we use the variable “speed limits” for two reasons that matter in the Swedish context. The speed limit is an indication of the type of built environment in Swedish cities, especially larger built environments, where different types of architectural styles impose a separation between main streets/highways and pedestrian paths close to buildings (in Swedish, this is called trafikseparering). These enclosed residential spaces could hypothetically affect the location of the shooting because of the possibility of there being easy escape routes and hiding places for offenders. If the speed limit is low, the area is most likely to be inside a residential area. Several cases of homicides have been related to the use of scooters in these areas with low-speed limits¹). The second reason is that the variable speed in small towns seems to indicate centrality and major services and access routes that could be relevant, according to the literature (Wright & Ribbens, 2016).

• Demographic and socioeconomic data: Data from SCB was obtained to create variables in the resident population, the young male population, unemployment, the composition of households, the foreign population, and the average income per DeSO. Another level of data was gathered in a grid composed of 250 m² and 1000 m² units.

Preparation of the dataset and data quality

Swedish national police collect data on shootings which are officially called “illegal firearm discharges” from local police districts to be published as national statistics. The national police have only used totals per region and month for their statistics and sent us the raw input data from police regions, including coordinates and dates (N = 1687). Our raw data included 1687 incidents, a total that is higher than the national police officially reported counts for shootings in that period (N = 1055). Inspecting the data, we noted that a large number (N = 519) were exact duplicates or had the same date which was removed (648) reducing the total number of shootings (N = 1039). In this study, the total incidents varied in the temporal and spatial/modelling analyses. For the totals in the temporal analysis, we counted the number of victims within each row and also accounted for an event where there were no harmed victims recorded (no dead and no injured). These events were coded as 1, and only when there was no recorded harm done to humans or perhaps a hole in the wall. For example, a row with 1 dead and 2 injured got the N = 3; a row with 0 dead and 1 injured got the N = 1; a row with 0 dead and 0 injured but a sign of a bullet in a wall, got the N = 1. This procedure increased the total number of incidents because, in several locations, there were multiple victims (N = 1067), excluding a few points that had no date or time stamp. The total incidents may have slightly varied depending on the temporal resolution of the analysis and availability of the data (hour, day, and season). For the seasonal analysis, we used data from 2017 to 2020 because we had data for the full 4 years. For the spatial analysis, we used the original dataset N = 1039. Records were later linked to the exact geolocation in space using GIS and the crime coordinates (SWEREF 99) using a map of Sweden. The dataset was used to run the temporal and spatial analysis as reported in “Temporal and Spatial Analysis”.

In the 290 municipalities, the counts of incidents were linked to each geographical unit (municipality, DeSO area and then street/road segment) as illustrated in Fig. 2 and worked as a base for assessing the geographical distribution and modelling discussed in “Modelling firearm violence”. First, to calculate the land uses closest to shootings, we used the “near” function in ArcGis, with the option to select a pair of x and y coordinates (for each incident) for each land use and choose the facility (e.g., bar, school) closest to each incident and provide a measurement in meters. The data preparation for the modelling also involved cleaning and linking databases. Given that we had many DeSO areas and street segments with no firearm violence, we restricted the modelling to municipalities that contained at least one incident.

Fig. 2

The two data resolutions: modelling firearm violence by street segment (a) and DeSO area (b)

Temporal and spatial analysis

First, the dataset was organized by time of day, days of the week, and months of the year to investigate temporal patterns by types of injuries. The hourly variation was described using the proportion of incidents by time of the day for all years. Differences in levels of shootings by season and week were tested using one-way ANOVA between the full years of data. For seasonal analysis, we had a total of 947 incidents: Winter N = 207, Spring N = 211, Summer = 283, Fall N = 246. Second, the distribution of shootings was assessed across the rural–urban continuum by the proportion of cases occurring by municipality type (Tables 1 and 2). Third, the distribution of shootings in relation to the main types of land use was also examined (Table 3).

Table 2 Population distribution and distribution of shootings by type of municipality in Sweden, 2017–2020

Table 3 Shootings per street segment and land use type and sections of the study area

Modelling firearm violence

Potential land use factors that are associated with shooting incidents in Sweden were examined by types of municipalities and by the rural–urban continuum. Additionally, a range of demographic and socioeconomic variables was examined: income, the proportion of young males in the population, the proportion of privately owned dwellings, and a measure of centrality for each municipality using their contiguous boundaries (distance to the city from the centre of each zone limited by the boundaries of the urban area). A vast majority of shootings occurred within densely populated areas, defined as contiguous built-up areas with a maximum of 200 m between buildings and at least 200 inhabitants by the Swedish National Land Survey. Only street segments and DeSO areas that are in such areas were used for modelling. For modelling, these were aggregated into three classifications; all municipalities deemed as cities by SALAR were designated as Urban, all commuting municipalities Accessible rural and rural municipalities Rural. The SALAR classification is used to define rural and urban contexts (Table 1). Therefore, in this study the modelling section focused on areas within urban perimeters only and on areas on the rural–urban continuum, including for instance, areas of urban fringes, connected villages along roads but excluding roads in between them.

CrimeStat IV software was used to test the models (Levine, 2015). A random-effects Poisson-Gamma regression model with a spatial autocorrelation component was utilized to estimate the association of the land use variables with the shootings (Levine, 2015; Lord et al., 2015). The model estimates the number of shootings as a function of the independent variables shown in Appendix. Formally, the number of shootings is assumed to be Poisson distributed and independent and has the form:

$$y_{i} |\lambda_{i} \sim Poisson\left( {\lambda_{i} } \right)$$

(1)

with the mean, ${\lambda }_{i}$, organized as

$$\lambda_{i} = {\text{exp}}\left( {{\text{X}}_{i}^{T} \beta + \epsilon_{i} } \right)$$

(2)

where exp() is an exponential function, ${\beta }_{i}$ is a vector of unknown coefficients for the k covariates plus an intercept and ${\epsilon }_{i}$ is the model error independent of all covariates, assumed to follow a gamma distribution with a mean equal to 1 and a variance equal to $\tau =\frac{1}{\omega }$ where $\omega$ is a parameter that is greater than 0 (Lord, 2015).

We first model the number of shootings by municipality type and per DeSO areas as a function of the independent variables shown in the Appendix. Second, we model the number of shootings as a function of these independent variables by street segments and municipality type. In both cases, we made two runs, one with all independent variables shown in Appendix and a second with only those independent variables that were statistically significant and relatively independent. To avoid multicollinearity, we used the tolerance criterion and dropped any variable that had a tolerance of less than 0.7 (Goldberger, 1991). This is called the reduced model. The sign of the coefficient indicates whether there is a positive (increasing) or negative (decreasing) effect of the variable. Since the coefficient is the exponent of the natural logarithm (i.e., ${e}^{Coefficient}$), a positive coefficient indicates that the dependent variable effect is greater than 1 for each unit change in the independent variable. In contrast, a negative coefficient indicates the dependent variable effect is less than 1 for each unit change in the independent variable. For example, if the coefficient is 0.9, then the dependent variable changes by 2.46 (i.e., ${e}^{0.9}$); that is, it increases with an increase in the independent variable. On the other hand, if the coefficient is − 0.9, then the dependent variable changes by 0.41 (i.e., ${e}^{-0.9}$); that is, it decreases with an increase in the independent variable.

The Poisson-Gamma was used to model urban and accessible rural municipalities. However, the binomial Logit model by street segment was used for rural areas, given that road segments either had an incident or not (1/0). The formula for the binomial logit model, commonly referred to as logistic regression, is expressed in terms of the log-odds of the probability of an event occurring. In this context, the model predicts the logit of the probability p of a binary Y as a linear combination of the predictor variables X₁,X₂,…,X_k. For more details, see Levine (2015). Significance levels based on credible intervals are shown in the results. In the selection of variables for the final model, only significant variables with tolerances of > 0.7 were kept to avoid excess multicollinearity. Figure 3 summarises the methodological steps of the study.

Fig. 3

The methodological steps of the study

Results

The temporal nature of shootings

There were 1067 records of shootings between 2 November 2016 to 13 March 2020; 47% of these were associated with injuries (35%) or deaths (12%). The remaining 53% constituted records of non-injury firearm discharge or other signs of gun-related violence in a particular place, often associated with witness testimonials such as the sound of gun shots (N_total = 1067, N_dead = 131, N_Injured = 371, N_{No_Injury} = 565). The testimonials are linked with visible, clear evidence of a shoot against a person or property (injuries, bullet holes, or other physical damage). However, the testimonial alone does not make it into the statistics. This group could involve shootings without human injuries. Regarding the time of the shootings, out of the 1,067 incidents, more than half were from 8 p.m. to 1 a.m. (Fig. 4). There are three small peaks in records among those who were killed, at 7 p.m., 10 p.m. and midnight. Among those injured, the peak was around 10–11 p.m., while for shootings with no injuries, the peak of records is around midnight.

Fig. 4

Firearm violence by outcome and time of the day, Sweden. The dotted line is the overall percentage of firearm violence by time of day

No significant variations in the recording of shootings between weekdays and weekends was found. A one-way ANOVA was performed to compare the effect of seasonal variation on shootings and was not statistically significant for this short time series (F(3, 946) = 2522, p = 0.056). ANOVA did not reach statistical significance at the conventional alpha level of 0.05, since the p-value (0.056) is slightly above this threshold. This suggests that, based on our results, there might not be a statistically significant difference between the group means, although slightly more shootings lead to injury or death in the summer than in the winter for example.

Spatial patterns of shootings across the rural–urban continuum

As many as 72% of shootings took place in areas identified as “urban” by the SALAR classification, while 28% in "rural" (note that 38% occurred in the three major Swedish cities (Stockholm, Gothenburg, and Malmö, which include 17% of the Swedish population). Table 2 shows the number of shootings and deaths or injuries by type of municipality. The percentage in brackets shows the proportion of the Swedish population by each type. Shootings are more serious in major cities than in any other municipalities.

Figure 5 helps in understanding the spatial distribution of gun violence and its correlation with urban infrastructure. The concentration of firearm incidents near subway stations and along subway lines could indicate that these areas, being transit hubs, experience higher foot traffic and interaction among diverse populations, potentially leading to more opportunities for violent encounters. The mapping of firearm violence in residential zones highlights the impact on community safety and the possible targeting of these areas due to various socio-economic factors, such as poverty, unemployment, or gang activity. The relationship between roads and firearm violence may point to the ease of access and escape routes for perpetrators. Major roads might facilitate quick movement, making it easier for offenders to carry out violent acts and evade law enforcement. Identifying hotspots can lead to targeted interventions, such as increased police patrols, community engagement programs, or modifications in urban design to enhance surveillance and reduce crime opportunities.

Fig. 5

Source: Authors

Firearm violence in Stockholm in relation to residential areas, subway stations/lines and roads.

Only a small proportion of the street segments had shootings occurring in them (929 out of 2,541,347 in Sweden). Of those that did have a shooting, 51% occurred in residential streets compared to 17% of all street segments that were residential in the country. Table 3 shows shootings by type of street segment, land use types, and PoIs. Thus, shootings are associated with particular types of land use regardless of the type of municipality. Figure 5 shows examples of these places in Stockholm municipality. Showing actual places in Stockholm helps to understand the environmental context—such as the density of buildings, presence of public spaces, and types of businesses or amenities nearby—that might influence the likelihood of violent incidents.

Overall, Fig. 5 serves as a bridge between abstract data and real-world applications, enhancing the understanding of firearm violence patterns in Stockholm and supporting more effective responses.

Modelling situational conditions of firearm violence

Table 4 shows the results of modelling characteristics by DeSO areas. These models are in a reduced (parsimonious) version because multicollinearity was found in the initial runs with all the variables. Incidents by type of outcome separately (death, injury and no injury) were not modelled given the limited number of incidents per type and area.

Table 4 Y = Skewed Poisson Gamma estimation method for counts of shootings in Sweden by municipality types, per DeSO¹ areas

There were three main common findings. First, in the regression models, land use variables such as the availability of playgrounds/kindergartens and the proximity to public transit stops were identified as significant factors to explain shootings. Second, fewer land use variables were significant in the rural municipalities in relation to firearm violence. Third, except for the rural models, firearm violence was highly associated with low-income households.

Table 5 shows the modelled characteristics by street segments. Although neighbourhood characteristics still dominate the models (demography, socio-economic status and street hierarchy), several types of land use were associated with shootings, namely the presence of parks in both models (urban and accessible rural municipalities) and playgrounds in the urban model which indicates that shootings occurred at the core of these neighbourhoods.

Table 5 Y = Skewed Poisson Gamma (urban & accessible rural) and binomial logit (rural) estimation method for counts of shootings in Sweden by municipality type, per street/road segment (N = 542,679 street/road segments)

Note that the effect of the size of many of those variables is substantial. For example, in Table 5, the coefficient for the bus station is 1.9245 which translates into an effect of 6.8517 (i.e., a road segment with a bus station accounted for 6.85 more shootings over the period than a road segment without a bus station). Even some of the weaker associations show substantial size effects. For example, in the same table also for the accessible rural category, the coefficient for young people is 0.0052, which translates into 1.0052 shootings per young person per road segment. Moreover, for the remote rural areas, none of the coefficients were statistically significant in Table 5; the small number of shootings that occurred in those areas diluted the precision of the results. These findings do not mean that the model is useless for rural areas. Again, some of those size effects are large (e.g., the coefficient for young people is similar to that of the urban and accessible rural category, though weaker). Figure 6 shows photographs of places where a firearm incident happened in the middle of a neighbourhood. In smaller municipalities, these incidents were associated with land uses closer to major roads.

Fig. 6

Examples of places where firearm incident led to death: a playground; b bus stop and c park

Discussion of the results

Although the level of violence has become stable in Sweden, firearm incidents are increasing. We compiled more than one thousand incidents and found that about one-sixth resulted in death and another one-third in injury. The other half constituted records of non-injury firearm discharge or other signs of gun-related violence. Signs of the recent increase in firearm incidents have also been observed elsewhere (Gerell et al., 2021) where these incidents could involve serious crimes such as drive-by shootings or robberies without injuries.

In Sweden, firearm incidents showed clear temporal patterns, often in the evenings or night-time. More than half of the cases were recorded from 8 p.m. to 1 a.m. Previous research also showed that shootings, particularly those related to gang violence or disputes, were more likely to happen during the evening and night-time hours (Loeffler & Flaxman, 2018), but such temporal concentrations might vary by geographical location. This evening/nocturnal pattern was also found in the international literature; see for instance, in the USA, close to bars and restaurants after 7 p.m. (Morrison et al., 2017), and in Brazil between 7 p.m. and 1 a.m. (Valente, 2019), although no significant variation was found in Dong et al. (2020) between daytime and night-time. No weekday patterns were found in our study, but there does appear to be some variation between (although not significant at the conventional alpha level of 0.05) the cold and hot months of the year, reflecting changes in people’s routine activity during summer and winter. This pattern has been found elsewhere, particularly in places with mild winters (McDowall & Curtis, 2014); while Szkola et al. (2021) found that crimes involving firearms were concentrated in the hot months of the year, as it was found by studies in the Global South [in Brazil, see Ceccato (2005) and Valente (2019)].

Although shootings clearly follow the urban hierarchy, with major cities having the highest number (e.g. Atun et al., 2019), our findings show that there are also variations across types of municipalities. Smaller cities that are close to the three largest municipalities appear to be affected by a ‘spillover effect.’ The data did not show major variations within ‘accessible rural’ and ‘remote rural municipalities’ for firearm incidents. Firearm violence is usually concentrated in big cities for several reasons; for instance, gangs tend to be more prevalent in urban areas (Ceccato & Westman, 2024). Also, in large municipalities, there may be greater access to firearms. However, even within large cities, certain land uses are more associated with shootings than others.

In Sweden, common variables that were associated in the models with shooting places were areas close not only to playgrounds, parks but also where guardianship was poor and where escape might be facilitated (industrial areas, public transit) (Fig. 5). Playgrounds in the heart of some residential areas and during the day are safe places for children. They are also examples of places that Day and Wager (2010) suggest as being pivotal in nurturing a diverse set of abilities and skills in children and youth. Especially after school hours and in the evening, they offer teenagers and young people on the outskirts of the city a place to gather and provide a sense of independence and freedom, far from the gaze of parents. These places might be where young people who experience social marginalisation might gather (Littman, 2022); note that previous research shows that these individuals may be involved in criminal networks (Jönsson & Nilsson, 2019). Our modelling results show that shootings were more likely to occur in poor areas, in interstitial places of these neighbourhoods where young people live (residential areas), in areas where open drug scenes occur. In large cities, these findings confirm the overlap of the geography of shootings with those for open drug scenes in Stockholm, as previously suggested by Magnusson (2023).

Although shootings in small municipalities are less frequent than in larger ones, they can still have a significant impact on local communities (Möllberg, 2022). In minor cities, firearm incidents were also close to inner city areas, especially to retail establishments, restaurants, and supermarkets connected to larger capacity roads. Like in larger cities, restaurants, bars, and nightclubs can be locations where altercations escalate into shootings along main arterial roads. While in minor cities shootings may occur in the inner city, in large cities, they are not exclusively central. Small convenience stores or shops may serve as informal meeting spots for young people to gather (Oldenburg, 1999), but these facilities may also attract crime (Eck, 2003). In summary, while firearm violence in smaller towns (Bell, 1992; Dewey, 1960; Pahl, 1966) may share similarities with that found in large cities, our findings show they are not the same and therefore pose unique challenges for both law enforcement and their communities, including geographical isolation, different cultural norms, limited resources, and the availability of firearms (but see e.g., Hemenway et al., 2020; Shukla et al., 2019), which is an inherent quality of the gradual transition from urban to rural areas in the Swedish context.

The switching coefficient signs between urban and accessible rural areas suggest that the relationship between land uses and other variables may vary depending on the level of accessibility. This could be due to differences in land management practices, socio-economic factors, or environmental conditions between urban and rural areas. A strong relationship between land use and shootings occur on streets in accessible rural areas, possibly due to less oversight and mixed land use, which may create environments conducive to criminal activity. In contrast, suburban areas often feature planned layouts that incorporate residential zones separate from commercial and recreational spaces, resulting in different patterns of activity and surveillance. This structured land use planning in suburban areas may affect the relationship between land use and shootings, compared to those observed in other areas.

Finally, this study is innovative in its exploration of different scales of analysis (area and street segment) in the modelling section and in providing further evidence of the Modifiable Area Unit Problem—MAUP—in regression analysis (see e.g., Fotheringham & Wong, 1991) in the field of criminology. In the next section, we turn to our conclusions and call for policies that address the unique needs of areas far from Stockholm, Gothenburg, or Malmö.

Conclusions and recommendations

This article sets out to characterise the distribution of firearm incidents both temporally and spatially across urban and rural environments in Sweden and identify land uses associated with these shootings. Shootings occur mostly in the evenings and reflect temporal variations that follow young people’s routine activities. Not all shootings fit into the patterns described previously and therefore interventions should be tailored considering these variations.

Using two different modelling strategies, we show that although the large majority of cases are concentrated to major cities, rural areas, and smaller towns also experience gun-related incidents, and that the prevalence and patterns may differ, which impose different prevention strategies and research challenges. Firstly, demographic diversity and socio-economic variability within the rural–urban continuum can differ vastly, even over short distances, making generalisations about shootings difficult even within municipality types. Secondly, the availability and quality of data on incidents of violence can be inconsistent, as reporting mechanisms and police capabilities may vary between more urbanised and more remote rural areas within the continuum. Additionally, cultural factors, including attitudes toward gun ownership and law enforcement, can significantly influence both the occurrence of shootings and the community’s responses to them. This complexity requires a nuanced approach to research that accounts for the heterogeneity of the rural–urban continuum, demanding new ways to address the multifaceted nature of shootings in these areas.

Although in Sweden, common shooting locations for young people include areas near playgrounds, parks, places with poor supervision, and areas facilitating escape, such as industrial zones and public transit points and restaurants, changing these environments will most certainly not affect the mechanisms that lead to these shootings, or if they do, only marginally. It would be necessary to have an in-depth knowledge of the mechanisms linking offenders and victims in the particular places and situational conditions where shootings occur. This could contribute to a better understanding of the meaning of these settings for those involved in these conflicts. It could be that such violence is viewed as a means for shooters to enhance their reputation within criminal circles, a performative act of young men’s identities (O'Reilly & Crutcher, 2006), thus holding significance on their own, regardless of their direct motives (Jönsson & Nilsson, 2019). If they belong to gangs, they claim to differ from other young people in the neighbourhood by committing these acts of extreme violence close to where they live.

The key to effective prevention is recognising that a one-size-fits-all approach may not work when dealing with firearm violence in diverse settings. In rural areas, in particular, police and community organisations can work together to engage with residents, especially young people. Reducing firearm violence demands a combination of strategies, including knowing the main motivations for these shootings, understanding links with drugs and organised crime and improving community policing practices. Law enforcement agencies can revise the allocation of resources by focusing on patrolling and monitoring these high-risk neighbourhoods and interstitial areas at particular times. Investing in programmes that target youth at risk, offer alternatives to gang involvement, and provide mentorship can help reduce the number of young individuals drawn into gun-related crimes.

In the long run, recognising that firearm crime is a concern, policymakers can consider implementing stricter gun control measures to reduce the availability of firearms, especially focusing on border control and the existent stock of arms in the country, especially in rural areas. Shootings frequently occur in poorer neighbourhoods, often where open drug scenes are present, already known by the police. Stricter regulations and better surveillance of establishments need to be implemented, focusing on place management and the quality of risky facilities that attract open drug scenes and firearm violence.

Despite the many strengths of our research design, there are some limitations. First, we were able to exclude roads in urban areas with no shootings, allowing us to identify covariates for both urban and accessible rural municipalities. However, we were not equally able to model remote rural municipalities per street segment given data limitations. Future research should explore new approaches to defining the rural–urban continuum and seek improved modeling strategies per street segment. This study represents a preliminary step in that direction. Second, for this study, we know little about the offenders and victims. Therefore, future research should focus on understanding the motivations behind these firearm incidents and the role that the situations played in these shootings.

Appendix: Description of the datasets

Type of data	Variables	Small area unit (DeSO) model (N = 4789)			Street segment model (N = 542679)			Year	Source
		Description	Format	Min/Max	Description	Format	Min/Max
Crime statistics (x,y location)	Shooting	Count within DeSO area	Continuous	0/1	Count linked to street segment	Continuous	0/5	2016–2020	Police authority: National Operations Department (NOA)
Land use variables (x,y location)	Bar	Count within DeSO area	Continuous	0/12	Count (within 150 m) per street segment	Continuous	0/3	2020	Open street maps
	Bus station	Count within DeSO area	Continuous	0/2	Count (within 150 m) per street segment	Continuous	0/2	2020	Open street maps
	Convenience store	Count within DeSO area	Continuous	0/12	Count (within 150 m) per street segment	Continuous	0/4	2020	Open street maps
	Gas station	Count within DeSO area	Continuous	0/10	Count (within 150 m) per street segment	Continuous	0/3	2020	Open street maps
	Kindergarten	Count within DeSO area	Continuous	0/8	Count (within 150 m) per street segment	Continuous	0/3	2020	Open street maps
	Playground	Count within DeSO area	Continuous	0/36	Count (within 150 m) per street segment	Continuous	0/9	2020	Open street maps
	Pub	Count within DeSO area	Continuous	0/14	Count (within 150 m) per street segment	Continuous	0/3	2020	Open street maps
	Railway station	Count within DeSO area	Continuous	0/6	Count (within 150 m) per street segment	Continuous	0/2	2020	Open street maps
	School	Count within DeSO area	Continuous	0/10	Count (within 150 m) per street segment	Continuous	0/7	2020	Open street maps
	Sports centre	Count within DeSO area	Continuous	0/5	Count (within 150 m) per street segment	Continuous	0/3	2020	Open street maps
	Supermarket	Count within DeSO area	Continuous	0/7	Count (within 150 m) per street segment	Continuous	0/2	2020	Open street maps
	Park area	Park area/DeSO area	Continuous	0/22	Intersection of park area with street segment	Nominal	0/1	2020	Open street maps
	Retail area	Retail area/DeSO area	Continuous	0/1	Intersection of retail area with street segment	Nominal	0/1	2020	Open street maps
	Industrial area	Industrial area/DeSO area	Continuous	0/2	Intersection of industrial area with street segment	Nominal	0/1	2020	Open street maps
Demographic & socio-economic variables	Young population (age 16–24)	Count of young population (age 16–24) per DeSO area	Continuous	86/2669	Count of young population (age 16–24) per DeSO area intersecting with street segment	Continuous	0/860	2020	SCB
	Low-income household	Count of low-income households per 250m2 or 1000m2. >167,400 kronor disposable income per year, per DeSO area	Continuous	7/924	Count of low-income households per 250m2 or 1000m2. > 167,400 kronor disposable income per year, intersecting with street segment	Continuous	0/560	2017	SCB
Other	Street speed limit	Median of street speed limits within DeSO area	Continuous	20–70	Speed limit attributed to street segment	Nominal	5/120	2020	Open street maps