Abstract

Deprived neighborhoods in Sweden in which criminal networks have a negative impact on local residents are labeled as “vulnerable neighborhoods” by the police. The method used by the police to classify such neighborhoods is largely based on perceptions, which raises issues of subjectivity and potential biases. The present study explores the characteristics of such neighborhoods based on registry data on socio-demographics and crime. The study employs data in the form of a grid of 250 x 250 meter vector grids (N=116,660) with data on population, foreign background, employment, age characteristics, household type, and eight types of crime. Generalized mixed-effects models of vector grids nested in municipalities were fitted to analyze the characteristics of vector grids classified as vulnerable (N=1678). Several variables are significantly associated with a vector grid being classified as vulnerable, with the proportion of the population that is foreign born, and the proportion with foreign-born parents, being the strongest predictors. In addition, we consider whether there are systematic differences between municipalities and develop a model based on regression coefficients to predict whether a vector grid is vulnerable. The model reclassifies 39.8 percent of the vector grids, identifying locations that statistically resemble vulnerable neighborhoods but are not classified as such, and vice versa.

Keywords

  1. vulnerable neighborhood; Swedish conditions; deprived neighborhood; crime; policing

Introduction

Over the past few years, there has been political debate in Sweden and the neighboring Nordic countries about the problems that Sweden is experiencing with gang violence and “vulnerable neighborhoods” (Johansson, 2018). In Norway and Denmark, these problems, which are often associated with migration and/or integration, are labeled svenske tilstander (Swedish conditions), and it is stated in the political debate that these two countries need to avoid ending up in a situation similar to that in Sweden (Lokland & Nilsson, 2018; Johansson, 2018; Lie, 2019). This debate has primarily been of a qualitative nature with a focus on vulnerable conditions in these neighborhoods, resembling how a ghetto is a racialized problem area, while skid row is its less racialized and less criminogenic counterpart (Huey & Kemple, 2007; Lynch et al., 2013). Meanwhile, very few quantitative studies have directed a central focus towards the characteristics of these vulnerable neighborhoods (Puur, 2020). A significant question in this context is whether it is possible to statistically determine if these areas differ from their surrounding environments and, if so, how?
Vulnerable is the label applied by the Swedish Police Authority to deprived neighborhoods in which criminals have a major impact on the local community (Polisen, 2017; Gerell et al., 2020). There are currently 60 areas that the Swedish Police have identified and labeled in this way, the majority being located in and around major cities (Polisen, 2019). This type of neighborhood tends to be characterized by higher levels of fear among residents than other residential areas and higher levels of crime (Brå, 2018), but the differences are generally not so pronounced (Gerell et al., 2020). For some types of crime, however, there are larger differences, and this coincides with particular types of crime that have received a great deal of media attention since 2010. There has been an increase in gun violence in Sweden (Sturup et al., 2019), which is, to a large extent, linked to vulnerable neighborhoods (Brå, 2015; Gerell et al., 2021). Vulnerable neighborhoods in Stockholm and Malmö, for instance, experience around 4–5 times as many shootings per capita as other parts of these cities (Gerell et al., 2020). Detonations of hand grenades follow a similar geographical pattern, although these also take place in city centers (Sturup et al., 2020). Similarly, torched cars have been linked to deprived neighborhoods (Gerell, 2017a; Malmberg et al., 2013), as have open drug markets (Magnusson, 2020; Gerell et al., 2021). It seems clear that there are problems with crime and fear among the residents of vulnerable neighborhoods and that these problems feed into political discourses in the Scandinavian countries on vulnerable neighborhoods, immigration and “Swedish conditions”.
To some extent, prejudice, and even racism, is likely to be a contributing factor to the problems in these neighborhoods. In the US, the policing of drugs is much heavier in black communities than in white ones, and it has been argued that this is driven in part by race and in part by economic interests (Lynch et al., 2013). The policing of minorities has given rise to the Black Lives Matter movement, which highlights how police violence disproportionally affects minority communities (Dillon & Sze, 2016). Similar processes exist in Europe, and in the Nordic countries it has been argued that the policing of minorities, and in particular police stops and controls, can increase feelings of exclusion. Young men with a minority background who are stopped a lot by the police perceive this as a marker of their unbelonging, reinforcing feelings of being excluded from the majority society (Solhjell et al., 2019). Such processes can serve to increase patterns of segregation and exclusion and in part contribute to the problems some neighborhoods suffer.
At the same time, however, the method employed by the police to determine which neighborhoods are labeled as vulnerable is largely based on subjective police perceptions (Puur et al., 2019). Whether or not an area is labeled as vulnerable depends on whether the local police perceive there to be criminal networks in the neighborhood who have a large impact on the local community (Polisen, 2017). There may be many biases that affect whether a neighborhood is classified as vulnerable. The amount and quality of police intelligence work may differ across police districts, and problems the police know little about will not be included. Similarly, there are large differences across police districts throughout Sweden in the level – and type – of problems they experience. This may affect how the police in different areas rate the problems found in a neighborhood that may be assessed as vulnerable.
In this paper, we will attempt to quantify the underlying and distinctive characteristics of neighborhoods that are labeled as vulnerable without using any police intelligence reports or subjective measurements. Our work is partly based on previous work (Puur, 2020) that used demographic and socio-economic data to analyze the characteristics of such neighborhoods. We use registry data on unemployment, age groups, household composition, foreign background, and police-reported crime to outline the characteristics of vulnerable neighborhoods. Thus, the main aim is to see which types of variables have the strongest impact on whether a location is classified as vulnerable. Secondary aims are to analyze whether there are differences between municipalities, and whether, or how, our statistical analysis identifies places as vulnerable that have not been identified as such by the police.
Our findings indicate that neighborhood grids classified as vulnerable are characterized by measurable and systematic differences compared to other grids across Sweden. These quantifiable differences emerge in different ways. For example, vulnerable neighborhood grids have a higher proportion of residents of foreign background, of young residents, larger households, larger populations, and also more gun violence and more (police-reported) narcotics crime. However, the findings also show that vulnerable neighborhood grids are characterized by lower levels of other types of crime, such as bicycle thefts and cases of assaults on public officials. A somewhat surprising result is that the statistical relationship between vulnerability and high levels of unemployment is rather weak. A strong statistical relationship between vulnerable neighborhood grids and the variables focused on foreign background raises several questions. It could be that the type of criminal organizing that lends itself to violence involving weapons, such as shootings, is more common among immigrant groups. A contributing factor could be that some young people, many of whom are immigrants with insecure home conditions, are more exposed to active recruitment into criminal networks than other young people with more stable living conditions (Polisen, 2017). It is also possible that some form of prejudice or similar plays a role, with criminal networks in neighborhoods characterized by having large proportions of residents of foreign background receiving more attention from the police, e.g., in the form of more regular routine checks (Kardell, 2006; Schclarek Mulinari, 2017). The present paper cannot answer such questions, but by analyzing systematic differences between vulnerable and non-vulnerable locations, we provide a starting point for further analysis.

Vulnerable neighborhoods, crime and scientific approaches?

The phenomenon of vulnerable neighborhoods, including their characteristics and crime, has been indirectly addressed in several scientific approaches. The relationship between deprived neighborhoods and crime may, for example, be explained by social disorganization, which is the theoretical idea that socially and economically disadvantaged neighborhoods are less capable of coordinating themselves to attain common goals and to monitor deviant behaviors (Bursik & Grasmick, 1993; Sampson et al., 1997; Sutherland et al., 2013). Factors such as low socio-economic status, residential instability, population heterogeneity, structural density, family disruption, and urbanization play a crucial part in generating environments that are more favorable to crime, since the ability of residents to practice social control in the neighborhood either fosters criminal behavior or diminishes it (Bruinsma et al., 2013; Bursik & Grasmick, 1993; Sampson, 2006; Sampson & Groves, 1989). Further, Sampson and colleagues (1997) have argued that these neighborhoods often lack collective efficacy, which means that residents show little trust in one another and are unwilling to intervene for the common good of the neighborhood (Sampson, 2006). Both ethnic heterogeneity and residential mobility make social organization among residents difficult, the former because there might be communication barriers that inhibit efficient social coordination and the latter because social networks are repeatedly disrupted (Sampson et al., 1997).
Social structures can also exert a certain pressure on some individuals in society to engage in various types of behavior, including criminal behavior. Merton’s theory of strain describes life goals as being culturally defined in a given society and as being shared via common societal values, whereas the opportunities for achieving these goals may vary substantially and may depend on where you come from or where you live (Agnew, 2011). Therefore, living in socially and economically disadvantaged neighborhoods creates a discrepancy between the goals set by society and individuals’ means or opportunities to achieve them, which can lead to some individuals pursuing alternative and sometimes deviant methods. Strain thus increases the risk of deviant behavior among individuals with limited opportunities.
The perception of some members of society as being deviant is also important for explaining criminal behavior (Becker, 1963), with juveniles from socially and economically deprived areas, for example, being at higher risk of being publicly labeled (Farrington, 1977). These are instances where juveniles are labeled as potential criminals, not because of their own deviant behavior but rather because of reputations based on the socio-economic conditions of their families and because they live in socially deprived areas (McAra & McVie, 2012). Equally, juveniles from certain socio-economic groups are more likely to be identified as suspects due to a higher police presence in certain neighborhoods (Sampson, 1986). In Sweden, Wästerfors and Burcar Alm (2020) found that young people from ethnic minorities who live in socially disadvantaged neighborhoods experience police encounters both more often and more negatively. This might be due to more active crime-control strategies in certain neighborhoods. There have been reports of individuals from minority groups being stopped by police several times a month during so-called routine checks (Schclarek Mulinari, 2017). The disproportionate monitoring and surveillance of specific groups and places, actions based on casual suspicion, and unexplained stops tend to weaken the legitimacy of the police. It can also weaken feelings of belonging in society among men of minority backgrounds who bear the brunt of being stopped (Solhjell et al., 2019).

Geographical analysis of deprived neighborhoods and crime

There have been prior attempts to examine vulnerable neighborhoods using registry data, but with the exception of Puur (2019), this has not been done directly in relation to the police definition of vulnerable neighborhoods. The National Council for Crime Prevention in Sweden used median income, social assistance, and the proportion of the population aged 16–30 to classify neighborhoods as socially vulnerable. They found that the most vulnerable 10 percent of neighborhoods had more crime and fear of crime but had, in general, not experienced worse trends over time than other urban areas (Brå, 2018).
In relation to deprived neighborhoods, structural characteristics are usually measured using units such as census tracts. A key concept that is often employed is that of concentrated disadvantage, which represents a compound measure of demographic or economic factors that together generate a concentration of disadvantage (Sampson et al., 1997; Gerell & Kronkvist, 2017; Jones & Pridemore, 2018). Measures of concentrated disadvantage typically include variables such as unemployment, education and/or social assistance recipiency as indicators of socio-economic status, in addition to sometimes including variables such as race, single-parent households, and similar (Sampson et al., 1997; Gerell & Kronkvist, 2017). Jones and Pridemore (2018) have found that concentrated disadvantage at the neighborhood level (measured using census tracts) contributes to crime on street segments and ought, therefore, to be considered alongside micro-level crime measures. A similar finding was noted by Gerell (2018), who found that crime at bus stops was influenced both by place-based characteristics, such as whether there was a bar nearby, and by the characteristics of the surrounding neighborhood, such as concentrated disadvantage or collective efficacy. Also, Kim (2018) used street segments to analyze the relationship between structural characteristics and crime using a similar concentrated disadvantage index, which included poverty, single-parent households, average household income and school degree, and which was calculated using census data at the block level and was then subsequently linked to street segments.

The Swedish Police specification of vulnerable neighborhoods

The first report on vulnerable neighborhoods was presented by the Swedish Police in 2014 and identified 63 such areas in Sweden (Polisen, 2014).1 The means employed to identify these areas were fairly crude and based on an unstructured method involving interviews with local police officers. This was followed by a second report in 2015, which listed 53 neighborhoods (Polisen, 2015). A third report was released in 2017 in which the methods used to analyze neighborhoods had been given a broader theoretical basis and had become more standardized (Polisen, 2017). The most recent report to date was presented in 2019 and listed 60 vulnerable neighborhoods (Polisen, 2019).
The current process employed by the police to determine whether a neighborhood is to be classified as vulnerable begins by posing a few basic questions to the local police districts regarding criminal organization and how criminality affects the local community in their district. For each question, the police districts can choose between four grades of severity in their responses. If all questions are graded with one of the two lowest grades of severity, the process ends. These districts are not considered to contain a vulnerable neighborhood (Gerell et al., 2020).
If at least one item is listed as having a more grave situation, the local police district is asked to provide a substantial amount of follow-up information. They are asked to note the location of the problem on a map of the district, with some districts having multiple neighborhoods of this kind. For each neighborhood that they identify, they are asked to complete a survey containing 80 questions on crime, extremism, and criminal organization in the neighborhood (Litbo et al., 2019).
The police district is then asked to list the individuals who are causing the problems, e.g., the criminal network of interest, dividing them into four broad categories labeled A–D. The A-category comprises those who generate criminal opportunities via connections, companies, or other assets. The B-category are the doers: those who are selling drugs, shooting people and similar. The C-category comprises the youths associated with the network, who are of importance since they generate disorder in the neighborhood and are at risk of being drawn into more severe criminality. Finally, the D-category are the children who are at risk of being exploited. The police rarely have good knowledge of the D-category and obtain a proxy for the size of this group by requesting statistics from Statistics Sweden on the number of children in the neighborhood in each cohort that have failed in school (Gerell et al., 2020).
Finally, the district is asked to mark on a map whether there are any locations where drugs are sold openly or where the police have trouble working, for instance, locations where they cannot park a police car without a risk of it being vandalized. The district then completes an area document in which the problems in the neighborhood are summarized in text format. The material is sent to the National Operations Department, where analysts go through the material.
In order to be classified as vulnerable, a listed area should be a neighborhood,2 and this neighborhood should be mostly residential and deprived. At this stage, locations that are not deemed to be neighborhoods, that are not deprived or that are in the city center (and thus not mostly residential) will be discarded from the analysis. For the remaining neighborhoods, analysts try to establish whether the degree of criminal impact on the local community is high enough to warrant the neighborhood being classified as vulnerable.
Evidently, this process involves multiple steps in which subjective perceptions can, and will, have an impact on the outcome. Police districts differ in many ways, and the individual police officers tasked with conducting the analysis differ even more. There may be differences in the level of knowledge/intelligence about neighborhoods, in how this is rated, in how it is described, and in how the descriptions are understood and evaluated by the National Operations Department of the Swedish Police (Nationella operativa avdelningen (Noa)).

Research design and data

Research questions

In this section we present our research questions, outline the demographic data and models employed, and describe our deliberations regarding geographical units of analysis and crime data.
The research questions we aim to answer in this paper are:
1.
What are the characteristics of vector grids in vulnerable neighborhoods?
2.
What differences exist between municipalities in how vector grids are classified as vulnerable?
3.
What differences can be identified between vector grids classified as vulnerable by the police and a classification based on statistical analysis of demographic and crime-related variables?
From a statistical and methodological perspective, the first question will be addressed by applying generalized linear mixed effects models with a logit link. The second question will be investigated by fitting a single-level logit link regression using dummy variables for the municipalities that have a vulnerable neighborhood and by providing examples of neighborhoods that our models suggest could be classified differently from the way they have been classified by the police. In this context, and with regard to the third research question, it should be noted that the police do have a substantial amount of well-developed intelligence and knowledge regarding how they assess and classify vulnerable areas, and the methods, analysis, and findings from this paper should be regarded as complementary and not as the only way of understanding systematic similarities and differences between vulnerable neighborhoods.

Identifying appropriate geographical units of analysis

There are several challenges when trying to analyze the characteristics of vulnerable neighborhoods identified from a police perspective. One problem is that the vulnerable areas are hand-drawn on maps by the police. This means that they do not match any administrative or statistical designation of neighborhoods. To solve this problem, and as mentioned above, we have used national grid data for populated areas of Sweden. All densely populated parts of Sweden have been divided into 250 x 250 meter vector grids, and our main analysis focuses on the characteristics of those grids that intersect with a vulnerable neighborhood. In total, the dataset consists of 116,660 vector grids, of which 1,678, or approximately 1.4 percent, present a geographical overlap with vulnerable neighborhoods. On average, these grids have a much higher population density (mean = 380) than the others (mean = 73) and together account for 638,000 of the total sample of 8,863,000 Swedish residents, or 7.2 percent of the population living in densely populated urban areas. Since the data are focused on densely populated urban areas, the dataset does not cover the entire Swedish population, which was 10.12 million in 2018.

Data

The data used in this analysis are drawn from five sources. As indicated above, these sources comprise: (1) a vectorized grid-net of Sweden with registry data on the population from Statistics Sweden; (2) crime data for eight crime categories provided by the police; (3) municipal level data on population and municipal boundaries provided by the Swedish Election Authority; (4) the areas that the police have labeled as vulnerable (also from the police); and (5) Electoral participation data on the election district level from the Swedish election authority which is used in a supplementary analysis as a test of social capital/collective efficacy.

General data management

Statistics Sweden deliver data based on a grid net covering Sweden, in which the whole country is divided into three subcategories. Places with no population, or very low population density, such as certain rural areas and areas of forest, are excluded. Grids outside urban areas and with low population density have a size of 1,000 x 1,000 meters. Vector grids with a high population density are registered with a size of 250 x 250 meters. These grids cover the populated parts of towns and cities. Because the focus of this study is on densely populated areas, only the 250 x 250 meter grids have been included in the analysis.
Several steps were taken to generate our final dataset. In the first of these, the geography of the four datasets containing the original source data were brought together, which resulted in a combined dataset containing 116,660 geographical grids with data from at least one of the four datasets. This means that we have generated as much geography as possible by allowing every vector grid that is present in at least one of the four datasets to be included.
In the second step, we combined these grids with a second layer of geographical data on the location of Swedish municipalities produced by the Swedish Election Authority. Each grid thus contains information about the municipality in which its centroid is located. In the third step, we added municipality population data from Statistics Sweden for the year 2018, so each grid contains information on the number of residents in the municipality in which it is located. In the fourth step, we linked data from police records on crime locations to the grids so that each grid contains information about the number of crimes recorded within it. All data management was performed in the statistical software package R (R Core Team, 2020), using RStudio and the dplyr (Wickham et al., 2020) and sf (Pebesma, 2018) packages. Regressions were fitted using the lme4 (Bates et al., 2015) package.

Registry grid data and municipal population

Four different sets of registry data from Statistics Sweden have been employed to capture foreign background (2018), employment (2017), household structure (2018), and age (2017).
The foreign background dataset includes four variables, which capture whether the residents and their parents were born in Sweden or abroad. In the present study, we use the proportion of the population who are foreign born and the proportion with two foreign-born parents.
The employment data include the number and proportion of people aged 20–64 who are in employment. Being employed is defined as having worked at least one hour per week in the measurement month, i.e., November 2017 (Statistics Sweden, 2021). In this paper, we use the proportion of the employed population aged 20–64.
The age-group dataset consists of the number of people in different age groups. In the present study, we use the proportion of the population comprised of children aged 0–15 years and the proportion comprised of youth/young aged 16–24.
The household structure data are divided into single-person households, partners without children, partners with child(ren), single parents with child(ren), and other households. Inclusion in the “other households” group means that at least one person in the household is neither a partner nor a child. This group includes for instance multi-generation households, friends living together, and similar. In the present study, we have used other households, since Puur (2020) showed that the presence of such households is associated with vulnerable neighborhoods.

Crime data

The dataset containing crime data from the police consists of reported outdoor offenses from the years 2016–2019 and data from the police registry on illegal firearm discharges from the month of November 2016 through 2019.
The statistics can be influenced by the willingness to report crimes, and in locations where people distrust the police and/or are afraid to contact the police due to a perceived risk of retaliation from criminals, the proportion of offenses that are reported to the police is likely to be lower. This is in turn likely to be associated with the socio-economic status of a location, with disadvantaged areas being characterized by both lower trust in the police and more fear of criminals (Goudrian et al., 2004). In Sweden, the difference between deprived neighborhoods and other places is, however, not particularly pronounced. The levels of trust are somewhat lower in deprived neighborhoods, but trust levels remain at a fairly high level overall (Brå, 2018). A recent report has also shown that, net of individual level controls, there were no reporting differences for robberies or sexual offenses associated with living in a deprived neighborhood, but there was a decline in the rate of reporting for assaults (Brå, 2021).
The illegal firearm discharge registry is a separate record, maintained by Swedish Police since November 2016, that keeps track of gun violence. Since there is no offense code in the official statistics that captures all gun violence, the police deemed it necessary to track this issue separately. The incidents included in this registry may cover a wide range of offense categories, with one incident often comprising more than one offense. Common offenses in the illegal firearm discharge registry include homicide, attempted homicide, aggravated assault, vandalism, and crimes listed under the Weapons Act.
The data on offenses reported to the police comprise four types of violent crime: assault, robbery, assaulting a public official, and public endangerment through the use of explosives (this offense code was introduced in 2018). It should be noted that some of the illegal firearm discharges are likely to be included here as well, but since there are very few illegal firearm discharges in comparison to the number of assaults, this is unlikely to have any major impact on our results.
To shed more light on the relationships, we also consider four non-violent crimes: bicycle theft, theft from a motor vehicle, vandalism, and narcotics offenses. As can be seen from Table 1, many of the variables are over-dispersed, in particular the crime variables, which are very unevenly distributed across the grids.
Table 1 Descriptive statistics for the grid-level variables included in the study. Reported as counts for population and crime, and as proportions for all other variables. Valid observations are based on queen contiguity values as discussed below.
VariableValid observationsVector-grid meanStandard deviation
Population112 94978.1150
Proportion employed108 201.864.289
Proportion foreign born112 949.118.171
Proportion with foreign-born parents112 949.031.067
Proportion “other households”112 943.048.125
Proportion aged 0-15112 352.170.163
Proportion aged 16-24112 352.083.118
Shootings116 660.008.112
Explosions116 660.003.066
Bicycle theft116 6601.667.5
Narcotics offenses116 6602.4317.0
Robberies116 660.1821.30
Vandalism116 6602.8833.3
Assault in public environment116 660.7895.22
Assaulting a public official116 660.1011.18
Theft from motor vehicle116 6601.175.44

For analysis, all variables were standardized into z-scores.

Election participation data

In a supplementary analysis, we also consider the share of the population that participates in national elections as a measurement of social capital or collective efficacy, which has been linked to crime (Sampson et al., 1997) and to vulnerable neighborhoods (Polisen, 2017). While election participation is a more general social capital variable than collective efficacy, prior studies have used similar data to operationalize collective efficacy (Weisburd et al., 2014). The election data is on electoral district level, and we have assigned each vector grid the value of the electoral district that it overlaps the most with using the st_join function in R.

Analysis and research design

Since vulnerable neighborhoods by definition have to be large and, on average, include 31 vector grids, we need to model spatial dependencies or clustering of the data in our analysis. We do this by calculating the mean value for each variable in adjacent grids (Queen contiguity), which we call cluster vector grids. We assume that this aggregate value for multiple grids will be more predictive of a location being a vulnerable neighborhood than the individual value in the vector grid itself. This is because clusters of grids are more geographically similar to a vulnerable neighborhood than the value of a single grid. Our assumption was tested by using the different versions of our variables in regression models, and the assumption turned out to be correct (results not shown). We therefore only use the mean value for adjacent vector grids in the paper.
Our analysis focuses on whether or not a vector grid is part of a vulnerable neighborhood. Since we expect there may be subjective differences between different police districts or municipalities in how they assess vulnerable neighborhoods, we fit the models as generalized mixed effects models with vector grids nested in municipalities. Before fitting our models, we standardized our continuous variables into z-scores. To consider differences across municipalities, we ran single-level models with dummies added for each municipality that has at least one vulnerable neighborhood using the fast dummies module in R. This generates a dichotomous variable for each municipality that has at least one vulnerable neighborhood. The regressions were fitted using the glmer command in the lme4 library with a logit link, since the outcome variable is dichotomous.
We test the associations with our independent variables in three main steps, each of which is divided into smaller parts. In the first step, we perform our main analysis, which corresponds to the first research question: What are the characteristics of vector grids in vulnerable neighborhoods? The models are mixed effects models, which take both the grids and the municipality into account.3 We first fit a model that only includes population at the municipal geographical level as an independent variable. In the second model we add crime, and in the third model population-based variables. This results in global level associations for whether a grid is considered vulnerable. These models include a random intercept at the municipal level.
To further assess our findings, we also provide a fourth model where we add electoral participation as that is an important variable theoretically. This variable is available on a different geographical unit of analysis, electoral districts, and these models are, therefore, fit as three-level models of vector grids nested in districts nested in municipalities.
In the second step, we compare systematic differences between municipalities – in line with the second research question. We fit a single-level model corresponding to model 3 in table 2 but with dummy variables for each municipality with at least one vulnerable neighborhood. This is to see whether there are systematic differences between municipalities in how many vector grids are classified as vulnerable, with the size of the dummy coefficient being indicative of a difference.
In the third step, which corresponds to the third research question, we use the same regression model, but without municipality dummies, to obtain predictive values for whether a vector grid meets the model criteria for being vulnerable, and we use weights from this regression to generate a vulnerability index. The index is calculated by multiplying the value of a variable in each vector grid by the regression coefficient for the same variable. We do this by using both only significant coefficients and using all coefficients for the index. Using these indexes, we then produce maps showing the similarities – and differences – between how our model predicts vulnerability and the way in which the police have classified vector grids as being vulnerable. We do this for three municipalities: the municipality with the most vulnerable neighborhoods and the two that have the highest and lowest municipal dummy variable coefficients from step two outlined above. The latter two can be interpreted as capturing municipalities for which the model identifies a potential bias in the classification, i.e., municipalities that have more or fewer grids than the model would suggest. A high value for the dummy variable means the model is suggesting that the municipality should have a higher number of grids that are classified as vulnerable, and vice versa.

Results

We began by analyzing bivariate associations between the independent variables and the dependent variable (results not shown). All variables present associations in the expected direction, as discussed below. The crime variables are positively associated with a vector grid being labeled as vulnerable. The strongest association is for shootings, followed by robbery, narcotics and theft from motor vehicles. Among the non-crime variables, higher employment presents a negative association and is thereby linked to a lower probability of an area being classified as vulnerable, while having higher proportions of children and youth present a positive association. However, these associations are weaker than the association with population, which is in turn weaker than the associations with the variables related to foreign background.

Main findings

Based on these findings, we proceed with our main analysis (Table 2). The generalized linear logistic regressions are fitted using four main models, with the first model only including the municipal population, and with the second adding cluster levels of crime. The third model adds the cluster mean population, in addition to the cluster proportions of foreign born, of foreign-born parents, of employed, of children, of youth and of other households, and the fourth adds electoral participation.
Model 1 shows that there is a strong association between municipal size measured by population and the likelihood of a grid being classified as vulnerable. This is unsurprising and, to some extent, simply reflects the need for a large population to generate the type of local criminal networks that are in focus in vulnerable areas. However, this result constitutes a useful point of departure for further analysis. As shown in the high ICC-values, most of the variance is situated between municipalities rather than within municipalities.
Model 2 shows a strong relationship between levels of assault, shootings, and narcotics crime in clusters of vector grids and being classified as vulnerable. Explosions, robberies, and theft from motor vehicles also exhibit positive, yet weaker, associations. Vandalism is unrelated to the outcome, while assaults against public officials and bicycle theft present strong negative associations with being labeled vulnerable. The strong negative coefficient for assaults on public officials is somewhat unexpected given that attacks against the police in vulnerable neighborhoods have received widespread attention.
Model 3 shows that population presents a significant, and fairly strong, association with grids being classified as vulnerable by the police. A one standard deviation increase in population, corresponding to 150 residents per vector grid in the area, has an impact that is approximately twice as large as that of a one standard deviation increase in shootings (corresponding to 0.11 shootings on average or about one shooting for clusters that include the full nine vector grids). Employment is the only population-based variable that is not significant, which is somewhat surprising given that unemployment is one of the variables explicitly discussed by the police in their definition of vulnerable neighborhoods. There are higher values registered for the proportions of foreign born, foreign-born parents, children, youth, and other households in vulnerable grids. The strongest associations are noted for the foreign-background variables. A one standard deviation increase in the proportion of foreign born (17 percentage points) has an impact that is approximately twice as large as that of a similar increase in population, or more than four times as large as a similar increase in shootings measured in standard deviations.
Adding the population-based variables reduces the coefficients for all crime variables except vandalism. Shootings, narcotics, assaulting public officials and bicycle theft are the only crime variables that retain significance. The reduction is quite dramatic for assaults, with a shift from a large and highly significant coefficient to becoming non-significant. Bicycle theft has the largest coefficient among the crime variables, and the coefficient is negative, which means that one of the primary crime-related factors for understanding grids labeled as vulnerable is that they have much lower levels of police-reported bicycle thefts.
To sum up, then, the third model shows that vulnerable vector grids tend to be in clusters of vector grids characterized by more shootings and narcotics crime but fewer bicycle thefts and assaults on public officials. The surrounding vector grids also tend to have a larger population, a higher proportion of children and youth, a higher proportion of the “other household” type and a much higher proportion of residents of foreign background, all expressed in terms of standard deviations.
Table 2 Generalized linear mixed effects model with logit link on whether a vector grid is in a vulnerable neighborhood. All vector-grid variables use the mean for cluster vector grids. All variables are standardized, and random intercepts at the municipal level are included in all models. * p < .05 ** p < .01 *** p < .001
VariableModel 1Model 2Model 3Model 4
 Coeff (p)Coeff (p)Coeff (p)Coeff (p)
Shootings .476 ***.162 ***.194 ***
Explosions .085 ***.024.036
Assault .904 ***-.010-.316
Assaulting a public official -1.13 ***-.181 ***-.033
Robbery .083 *.016.446 ***
Theft from motor vehicle .093 ***-.007-.445 ***
Bicycle theft -.904 ***-.434 ***-1.366 ***
Vandalism -.001-.020-.106
Narcotics .420 ***.223 ***.452 **
Population  .438 ***1.395 ***
Proportion Employed  -.042-.073
Proportion Foreign born  .748 ***1.135 ***
Proportion foreign-born parents  .572 ***.661 ***
Proportion aged 0-15  .205 **.292 *
Proportion aged 16-24  .085 *.252 *
Proportion “other households”  .277 ***.558 ***
Electoral participation   -355 ***
Municipal-level variable    
Municipal population3.83 ***3.51 ***2.68 **-.035
     
ICC.769.761.657.822
Intercept-12.3-12.4-12.69.63
AIC11651.18687.65481.33317.6
N116660116660116391111768
In a final model we assess whether a measurement of social capital – the share of residents who voted in the last election (2018) – impacts on our understanding of the characteristics of vector grids in vulnerable neighborhoods. In Table 2, model 4, we present the results from this three-level model of vector grids nested in electoral districts nested in municipalities. There are several substantial differences in this model, with perhaps the most prominent change being that the municipality population is rendered non-significant, while the electoral participation variable is highly significant instead. In addition, assaulting a public official becomes non-significant, while robbery becomes significant and positive, and theft from motor vehicle becomes significant and negative. The coefficients for several other variables change, too, but not significance or the sign. This highlights the importance of considering social capital, collective efficacy, and similar variables as important measures in understanding vulnerable neighborhoods.

Systematic differences between municipalities

The above analysis has shown how we can understand (clusters of) vector grids that are labeled as vulnerable, but we have yet not touched on possible differences between municipalities. However, our random intercept models registered large differences in the local intercepts, and we now proceed to test systematic differences between municipalities, in line with our second research question.
We do this by fitting our models without a random intercept, i.e., as single-level models, in 27 separate regressions that include a dummy for each municipality that has a vulnerable neighborhood (Table 3).4 As expected, we find that the coefficients tend to be positive, which is natural since we only include dummies for neighborhoods that have vulnerable neighborhoods, while there will also be similar locations elsewhere. Huddinge is the municipality with the lowest coefficient, followed by Landskrona, whereas Örebro and Jönköping have the highest coefficients. We interpret this as showing that our model is suggestive of there being more/larger areas that would be classified as vulnerable in Huddinge and Landskrona, and fewer/smaller areas in Örebro and Jönköping. We also fitted this as one single regression with all dummies included, yielding similar differences but different coefficients (results not shown).
Table 3 27 municipality dummy regressions, fitted as single-level models with the same variables as Model 3 in Table 2 (with the exception of the random intercept), with a separate regression for each municipality with a vulnerable neighborhood adding its dummy variable. The municipality of Ale is in italics to highlight the fact that this municipality does not have a vulnerable neighborhood. * p < .05 *** p < .001
Municipality (ID)Municipal dummy coefficientCountyPopulation in municipality 2018
Järfälla (123).384Stockholm78 480
Huddinge (126)-.241Stockholm111 722
Botkyrka (127).398Stockholm93 106
Haninge (136).805 **Stockholm89 989
Upplands-Bro (139).335Stockholm28 756
Sollentuna (163).730 *Stockholm72 528
Stockholm (180).483Stockholm962 154
Södertälje (181).879 **Stockholm97 381
Sundbyberg (183).504Stockholm50 564
Uppsala (380).651 *Uppsala225 164
Eskilstuna (484).394Södermanland105 924
Linköping (580).513Östergötland161 034
Norrköping (581).712 *Östergötland141 676
Jönköping (680).944 ***Jönköping139 222
Växjö (780).862 **Kronoberg92 567
Malmö (1280).693 *Skåne339 313
Landskrona (1282)-.002Skåne45 775
Helsingborg (1283).118Skåne145 415
Kristianstad (1290).560Skåne84 908
Halmstad (1380).440Halland101 268
Ale (1440).414Västra Götaland30 926
Göteborg (1480).633 *Västra Götaland571 868
Trollhättan (1488).527Västra Götaland58 728
Borås (1490).438Västra Götaland112 178
Örebro (1880)1.02 ***Örebro153 367
Västerås (1980).205Västmanland152 078
Borlänge (2081).631 *Dalarna52 224

Predicting vulnerable neighborhoods

The final step of the analysis involves using the results from our regressions to consider places that the model believes should be classified as vulnerable but which are not, or vice versa. This corresponds to our third research question. We will not be exploring this in detail in this paper, but, for exploratory purposes, we will calculate predictions and show predictive maps of the two municipalities whose dummies produced the highest and lowest coefficients: Huddinge (lowest) and Örebro (highest), in addition to the municipality with the largest number of vulnerable neighborhoods (Gothenburg).
The random intercept model provides a good estimate of how to understand which vector grids are likely to be designated as vulnerable, but, for predictive purposes, the random intercepts make it more difficult to estimate global predictive values. We therefore fit single-level models and use the coefficients from the regression as weights to calculate a predictive index for whether or not a vector grid should be considered vulnerable. The single-level regression yields similar results, although some coefficients change. The most substantial changes are that the coefficient for the municipal population variable is much lower and that a significant association is registered between robbery and being vulnerable.
We use the coefficients to calculate a prediction index, in which each grid is assigned the value of each significant coefficient times the value of the variable. We then test which threshold value on this index is needed to produce the same number of vector grids as the number classified as vulnerable by the police (1,678), and we use this value (5.835) as a threshold for predicting which of the vector grids should be classified as vulnerable. This results in 60.9 percent of the vector grids classified by the police as vulnerable being classified as vulnerable by our model, while the remaining 39.1 percent of vector grids are substituted by different vector grids. We have also repeated this process using only the significant variables in the model. This yields very similar results but with slightly more vector grids being replaced (39.8 percent).
Figure 1 shows the western section of Huddinge, which roughly corresponds to half of the municipality.5 There is a neighborhood classified as vulnerable by the police in the westernmost part (Vårby), which mostly consist of vector grids identified as vulnerable by the model, but which has a section in the middle that is classified as non-vulnerable – presented as translucent cells in the figure. The model identifies a few vector grids to the east of this neighborhood as vulnerable in addition to a few grids scattered in other locations. Importantly, though, the model also identifies a large cluster of vulnerable vector grids in the south-eastern section of the map (red cells). This corresponds to the Flemingsberg area. If we were to construct vulnerable neighborhoods based on the model, this part of Huddinge would emerge as a new vulnerable neighborhood.
For Örebro Municipality, the model very accurately predicts the neighborhood classified as vulnerable by the police in the northern part of the municipality (Vivalla). As regards the other vulnerable neighborhood (Oxhagen/Varberga) in the municipality, however, the model shows this neighborhood mostly to consist of vector grids that are not vulnerable, with only six of 21 vector grids being classified as such. The model thus suggests either that the size of the southern vulnerable neighborhood should be reduced or that it should be declassified as vulnerable based on the model predictions using the variables included in this analysis.
The two examples described above involve medium-sized cities. In our final example, we present Gothenburg, the second largest city in Sweden. Gothenburg has nine vulnerable neighborhoods, which is the highest number for any city in the country. As can be seen from Figure 3, our model suggests that several areas in north-eastern Gothenburg that the police have not classified as vulnerable have characteristics that are similar to those of vulnerable neighborhoods. There are already six vulnerable neighborhoods in the north-eastern section of the city, but the model classifies large parts of neighboring areas as also being vulnerable.
In addition, the model suggests that the Backa neighborhood (in the center of the map) should be somewhat smaller. Interestingly, the model also identifies an area in the city center as being vulnerable. This appears to be driven by a very high number of narcotics offenses, robberies and a high proportion of youth. Since this is a very different type of location, it highlights a need to adjust this type of model to account for city center effects in future studies.

Discussion

This study has shown that the strongest predictors of a vector grid being labeled as vulnerable are the foreign-background variables and the size of the municipality in which the grid is located. The prominence of these variables is not surprising, but the fact that their coefficients are so much larger than many of the other variables is surprising and will be discussed further below. Of the crime variables, being vulnerable is positively related to shootings, narcotics crime, and, in some models, robberies It is strongly negatively associated with police-reported bicycle thefts and, in most models, with assaulting a public official.
It is already well established that gang violence, including shootings and explosions, is strongly linked to these neighborhoods (Gerell et al., 2021; Sturup et al., 2020). Similarly, registered narcotics crime is largely a function of policing, and the police have prioritized vulnerable neighborhoods a great deal over recent years. The criminal networks that are active in vulnerable neighborhoods do tend to be involved in the narcotics trade, and this fact, combined with the heavy policing of such neighborhoods, makes this finding unsurprising, too. While unsurprising, it should, however, be noted that heavy policing of a neighborhood can lead to negative consequences for the neighborhood, and, in particular, for young men with a minority background (Solhjell et al., 2019). Policing can thus both be part of the problem and part of the solution, and it is of some importance to find the right balance of repression and building relations to achieve real progress (Gerell et al., 2020).
Few would have guessed, however, that the coefficient for bicycle thefts would be greater than that presented by these other offense types. This may, to some degree, be due to fewer residents in vulnerable neighborhoods using bicycles, but perhaps more to the point, it may be that fewer people in these neighborhoods have insurance that covers bicycle theft or a bicycle that is sufficiently expensive for an insurance claim to be likely to generate any money. This would mean that fewer bicycle thefts would be reported in vulnerable neighborhoods as compared to other places. In addition, all of the crime variables are, to some extent, likely to be subject to more underreporting in vulnerable neighborhoods than elsewhere, since trust in police is lower in these areas (Brå, 2018), and since reporting is associated with socio-economic status (Goudrian et al., 2014). This means that we may be underestimating the crime rates in vulnerable neighborhoods. It should, however, be noted that the difference in trust between deprived neighborhoods and other neighborhoods is relatively small in Sweden, and that neighborhood deprivation is only linked to lower reporting propensities for some types of crime (Brå, 2021).
Employment, which is, to some extent, a proxy for socio-economic status, is, unexpectedly, not associated with grids being labeled as vulnerable net of controls. This variable also presented only a weak bivariate correlation with being designated as vulnerable to begin with. This was not expected as unemployment is highlighted as one of the key variables in police discussions of vulnerable neighborhoods. In supplementary analysis, we fitted our main result models but without the foreign-background variables (results not shown). The models indicate that vulnerable vector grids do tend to have lower employment. This points to the strong association between (un)employment and foreign background. Vulnerable neighborhoods are poorer than other neighborhoods, but they are more characterized by foreign background than by unemployment. Whether low socio-economic status is a key driver of neighborhoods being classified as vulnerable will need to be considered with better SES-data in the future. Studies on segregation and residential mobility have highlighted the importance of income and of house prices (Wessel & Lunke, 2021), while also noting that where a person lives can impact on their earnings (Wessel & Magnusson Turner, 2021). The findings in this paper should be explored further with the addition of income, education and other variables to better capture the impact of socio-economic status. This can also be done by indexing several of these variables to represent from affluent to vulnerable area grids (Guldåker et al., 2021). Our analysis also reveals that considering electoral participation as a proxy for social capital can shed further light on the mechanisms behind classifying neighborhoods as vulnerable. Adding election districts and electoral participation to the analysis significantly impacted on how municipal population and crime was related to designation as vulnerable. Electoral participation in itself has a strong association with whether a location is characterized as vulnerable or not.
The factors that drive the vulnerable classification thus appear to mainly involve strong concentrations of residents with a foreign background, living in highly populated grids and municipalities, combined with shootings and narcotics crime (and fewer than expected bicycle thefts). The fact that vulnerable neighborhoods tend to have many residents of foreign background is well known, but the magnitude of the importance of the foreign-background variables on the classification of vector grids as vulnerable is still surprising. This study cannot say why this is, but there are some possible explanations that deserve mentioning. Prior studies have noted that there are potential biases in the police and the justice system (Kardell, 2006; Kardell, 2011), and it is possible that something similar may be leading to a stronger focus on crime or criminal networks in neighborhoods with a large proportion of residents of foreign background. If this is the case, then neighborhoods with large shares of residents from foreign backgrounds would stand out. It is also possible that the types of crime or of criminal organizing that, according to the police, characterize vulnerable neighborhoods are more likely among people of foreign background. We are not aware of any studies that have considered this issue, but it has previously been shown that the types of criminal networks that are common in these neighborhoods – street gangs or loosely organized criminal networks – tend to have a high proportion of foreign-born members (Rostami et al., 2018).
When comparing the model results with the police’s designation of vulnerable neighborhoods, we can see that the predictive model appears to capture somewhat similar areas. There are also differences. Some neighborhoods that are classified as vulnerable by the police are not predicted to be vulnerable by the model, and vice versa. While this is just one among several possible models, this suggests that there may be areas that may be reclassified in the future, going either from vulnerable to not vulnerable or from not being vulnerable to being classified as such. While the police classification is based on intelligence and experience from the neighborhoods that our study does not have access to, the statistical method outlined here may serve as a complementary tool. Neighborhoods that are predicted to be vulnerable by the model but that are not classified as such by the police may, for instance, be worth keeping an eye on in case criminal networks grow stronger in the future at such locations.
When it comes to comparisons across municipalities, there appear to be some differences. Some municipalities have a lower likelihood than others of having vector grids classified as vulnerable. Huddinge, in particular, stands out as having a lower likelihood of vector grids being classified as vulnerable than other municipalities with vulnerable neighborhoods. This may be due to local differences in knowledge, judgment, or intelligence, or it may just be that our statistical model identifies anomalies.
While this study provides no definitive answer to how vulnerable neighborhoods might be identified without the use of police intelligence and similar, it shows that it is feasible to statistically identify locations similar to those identified by the police, which can provide additional insight into the structural features associated with the vulnerability of neighborhoods. It also provides some indication as to differences between municipalities and highlights how some vulnerable neighborhoods differ statistically from other vulnerable neighborhoods. We believe this can serve as a starting point for both a better understanding of vulnerable neighborhoods and for the method used by the police to classify such neighborhoods. For future studies, a longitudinal perspective would be of interest, both to understand the development of neighborhoods before they are classified as vulnerable and to test whether a statistical model can accurately predict changes in the police classification of vulnerable neighborhoods over time.
In addition, future studies would do well to incorporate more and better variables. The present study lacked a good measurement of socio-economic status, and adding income, education or another SES-related variables would be desirable. In addition, both Swedish Police (Polisen, 2017) and researchers (Sampson et al., 1997) highlight the importance of social capital and, in particular, collective efficacy in understanding vulnerable neighborhoods. Such data is hard to come by on a national level, but we added electoral participation data in an exploratory model, which have been used as a proxy for social capital/collective efficacy before (Weisburd et al., 2012). Improving on the present paper with more data – and, in particular, more datapoints over time – can help us gain an understanding of why some neighborhoods experience increased problems with gangs and crime, and perhaps also of what Swedish conditions actually are and how they can be understood.

Footnotes

1
Due to a typo in the report, which stated that there were 55 neighborhoods, references to the number 55 persist in some discussions on the topic, e.g., “55 no-go areas in Sweden” in international media.
2
Some police districts report entire cities as struggling with criminal networks, and others report small rural locations with very few residents. Both these cases would be excluded in the current context due to their not being of the neighborhood type.
3
The models were fit using the bobyqa optimizer to reduce problems associated with non-converging models.
4
There are 26 municipalities with a vulnerable neighborhood, but in one case this neighborhood is right on the border with another municipality, with some of the vector grids being assigned to the other municipality. We include this other municipality (Ale) for reasons of transparency.
5
The rest of the municipality contains no grids that are predicted to be vulnerable.

References

Agnew, R. (2011). Revitalizing Merton: General strain theory. The origins of American criminology, 16, 137-158. https://doi.org/10.4324/9781315133683-7.
Bates, D., Maechler, M., Bolker, B., & Walker, S. (2015). Fitting Linear Mixed-Effects Models Using lme4. Journal of Statistical Software, 67(1), 1-48. https://doi.org/10.18637/jss.v067.i01
Becker, H.S. (1963). Outsiders: Studies in the Sociology of Deviance. New York: Free Press.
Bruinsma, G., Pauwels, L., Weerman, F., & Bernasco, W. (2013). Social Disorganization, Social Capital, Collective Efficacy and the Spatial Distribution of Crime and Offenders: An Empirical Test of Six Neighbourhood Models for a Dutch City. British Journal of Criminology, 53(5), 942. https://doi.org/10.1093/bjc/azt030
Bursik, R. J., & Grasmick, H. G. (1993). Neighborhoods and crime: The dimensions of effective community control. New York: Lexington Books.
Brå [Brottsförebyggande rådet (National Council for Crime Prevention)] (2018). Utvecklingen i socialt utsatta områden i Urban Miljö 2006-2017. Brå-rapport 2018:9. Brottsförebyggande rådet, Stockholm.
Brå (2021). Anmälningsanalys vid fyra typer av brott. En statistisk analys av skillnader i anmälningsbenägenhet. Kortanalys 1/2021. Brottsförebyggande rådet, Stockholm.
Dillon, L., & Sze, J. (2016). Police power and particulate matters: Environmental justice and the spatialities of in/securities in US cities. English Language Notes, 54(2), 13-23. https://doi.org/10.1215/00138282-54.2.13.
Farrington, D. (1977). The effects of public labeling. British Journal of Criminology, 17:112-125.
Gerell, M. (2017a). Collective efficacy and arson: The case of Malmö. Journal of Scandinavian Studies in Criminology and Crime Prevention, 18(1), 35-51. https://doi.org/10.1080/14043858.2017.1298172
Gerell, M. (2018). Bus stops and violence, are risky places really risky? European Journal on Criminal Policy and Research, 25:4, 351-371. https://doi.org/10.1007/s10610-018-9382-5.
Gerell, M., Hallin, P. O., Nilvall, K., & Westerdahl, S. (2020). ATT VÄNDA UTVECKLINGEN – från utsatta områden till trygghet och delaktighet. Malmö University Publications In Urban Studies. MAPIUS 26.
Gerell, M., & Kronkvist, K. (2017). Violent crime, collective efficacy and city-centre effects in Malmö. British Journal of Criminology, 57(5), 1185-1207. https://doi.org/10.1093/bjc/azw074
Gerell, M., Sturup, J., Rostami, A., Magnusson, M., Nilvall, K., & Khoshnood, A. (2021). Open Drug Markets, Vulnerable Neighbourhoods and Gun Violence in Two Swedish Cities. Journal of Policing, Intelligence and Counter Terrorism, 16:3, 223-244. https://doi.org/10.1080/18335330.2021.1889019
Goudriaan, H., Lynch, J. P., & Nieuwbeerta, P. (2004). Reporting to the police in western nations: A theoretical analysis of the effects of social context. Justice quarterly, 21(4), 933-969. https://doi.org/10.1080/07418820400096041
Guldåker, N.; Hallin, P.-O.; Nilvall, K.; Gerell, M. Crime Prevention Based on the Strategic Mapping of Living Conditions. ISPRS Int. J. Geo-Inf. 2021, 10, 719. https://doi.org/10.3390/ijgi10110719
Huey, L, Kemple, T (2007) ‘Let the streets take care of themselves’: Making sociological and common sense of ‘skid row’. Urban Studies 44: 2305–2319. https://doi.org/10.1080/00420980701540911
Johansson, L. (2018). Frykten for svenske tilstander. Norsk medietidsskrift, 25(03), 1-14. https://doi.org/10.18261/issn.0805-9535-2018-03-06.
Jones, R. W., & Pridemore, W. A. (2018). Toward an Integrated Multilevel Theory of Crime at Place: Routine Activities, Social Disorganization, and The Law of Crime Concentration. Journal of Quantitative Criminology, 35(3), 543–572. https://doi.org/10.1007/s10940-018-9397-6.
Kardell, J. (2006). Diskriminering av personer med utländsk bakgrund i rättsväsendet – en kvantitativ analys. In Sarnecky, J. (ed.), Är rättvisan rättvis? Tio perspektiv på diskriminering av etniska och religiösa minoriteter inom rättssystemet. Stockholm, Fritzes, 67-109.
Kardell, J. (2011). Utländsk bakgrund och registrerad brottslighet: Överrepresentationen i den svenska kriminalstatistiken. Stockholms Universitet, Kriminologiska institutionen.
Kim, Y. A. (2018). Examining the Relationship Between the Structural Characteristics of Place and Crime by Imputing Census Block Data in Street Segments: Is the Pain Worth the Gain? Journal of Quantitative Criminology, 34(1), 67–110. https://doi.org/10.1007/s10940-016-9323-8.
Lie, E. (2018) Peter Lindström og Ulf Sempert (red.): Kriminologi och poliskunskap. Mötet mellan forskning och praktik. Nordisk politiforskning, 6(01), 80-84. https://doi.org/10.18261/issn.1894-8693-2019-01-07.
Litbo, S., Puur, M., & Gerell, M. (2019). Delrapport 2. Ett nytt förslag till lokalt lägesbildsverktyg. Malmö Universitet.
Lokland, I. A., & Nilsson, L. (2018). “Det svenska tillståndet”: en kvalitativ publikstudie om hur norska och danska nyhetskonsumenter uppfattar Sverige. Lund University Student thesis in Journalism.
Lynch, M., Omori, M., Roussell, A., & Valasik, M. (2013). Policing the ‘progressive’city: The racialized geography of drug law enforcement. Theoretical criminology, 17(3), 335-357. https://doi.org/10.1177/1362480613476986
Magnusson, M. M. (2020). Mapping Open Drug Scenes (ODS). Crime and Fear in Public Places, 305-325. https://doi.org/10.4324/9780429352775-21.
Malmberg, B., Andersson, E., & Östh, J. (2013). Segregation and urban unrest in Sweden. Urban geography, 34(7), 1031-1046. https://doi.org/10.1080/02723638.2013.799370
McAra L., & McVie, S. (2012). Negotiated order: The groundwork for a theory of offending pathways. Criminology and Criminal Justice 12(4): 347-375. https://doi.org/10.1177/1748895812455810
Brå (2015). Skjutningar 2006 och 2014 – Omfattning, spridning och skador. Stockholm: National Council for Crime Prevention.
Pebesma, E., 2018. Simple Features for R: Standardized Support for Spatial Vector Data. The R Journal 10 (1), 439-446, https://doi.org/10.32614/RJ-2018-009
Polisen (2014). En nationell översikt av kriminella nätverk med stor påverkan i lokalsamhället. Stockholm: Rikskriminalpolisen.
Polisen (2015). Utsatta områden – sociala risker, kollektiv förmåga och oönskade händelser. Stockholm: Nationella Operativa Avdelningen.
Polisen (2017). Utsatta områden – Social ordning, kriminell struktur och utmaningar för polisen. Stockholm: Nationella Operativa Avdelningen.
Polisen (2019). Kriminell påverkan i lokalsamhället – En lägesbild för utvecklingen i socialt utsatta områden. Stockholm: Nationella Operativa Avdelningen.
Puur, M. (2020). The grid of Sweden. A micro-unit analysis of vulnerable neighborhoods. Degree project in criminology, Malmö University, Faculty of Health and Society.
Puur, M., Litbo, S., & Gerell, M. (2019). Slutrapport. Utsatta områden i Sverige och Polisens lägesbildsverktyg.
R Core Team (2020). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.
Rostami, A., Mondani, H., Carlsson, C., Edling, C., & Sarnecki, J. (2018). Våldsbejakande extremism och organiserad brottslighet i Sverige. Forskningsrapport 2018: 4. Stockholm: Institutet för framtidsstudier.
Sampson, R. (1986) Effects of socio-economic context on official reaction to juvenile delinquency. American Sociological Review, 51: 876-885. https://doi.org/10.2307/2095373
Sampson, R. J. (2006). How does community context matter? Social mechanisms and the explanation of crime rates. In: Wikström, P-O., & Sampson, R. J. (eds.), The explanation of crime: Context, mechanisms and development. Cambridge: Cambridge University Press, 31- 60.
Sampson, R. J. & Groves, W. B. (1989). Community Structure and Crime. Testing Social-Disorganization Theory. American Journal of Sociology, 94, 774-802. https://doi.org/10.1086/229068
Sampson, R. J., Raudenbush, S. W., & Earls, F. (1997). Neighbourhoods and Violent Crime: A Multilevel Study of Collective Efficacy. Science, 277: 918–24. https://doi.org/10.1126/science.277.5328.918
Schclarek Mulinari, L. (2017). Slumpvis utvald: Ras-/etnisk profilering i Sverige. Civil Rights Defenders; Kriminologiska institutionen, Stockholms universitet.
Solhjell, R., Saarikkomäki, E., Haller, M. B., Wästerfors, D., & Kolind, T. (2019). “We are seen as a threat”: Police stops of young ethnic minorities in the Nordic countries. Critical Criminology, 27(2), 347-361. https://doi.org/10.1007/s10612-018-9408-9.
Sturup, J., Gerell, M., & Rostami, A. (2020). Explosive violence: A near-repeat study of hand grenade detonations and shootings in urban Sweden. European journal of criminology, 17(5), 661-677. https://doi.org/10.1177/1477370818820656
Sturup, J., Rostami, A., Mondani, H., Gerell, M., Sarnecki, J., & Edling, C. (2019). Increased gun violence among young males in Sweden: A Descriptive National Survey and International Comparison. European Journal on Criminal Policy and Research, 25(4), 365-378. https://doi.org/10.1007/s10610-018-9387-0.
Sutherland, A., Brunton-Smith, I., & Jackson, J. (2013). Collective Efficacy, Deprivation and Violence in London. British Journal of Criminology, 53(6), 1050–1074. https://doi.org/10.1093/bjc/azt050
Wessel, T., & Lunke, E. B. (2021). Raising children in the inner city: still a mismatch between housing and households?. Housing Studies, 36(1), 131-151. https://doi.org/10.1080/02673037.2019.1686128
Wessel, T., & Magnusson Turner, L. (2021). The migration pathway to economic mobility: Does gender matter?. Population, Space and Place, 27(4), e2419. https://doi.org/10.1002/psp.2419
Weisburd, David L., Groff, Elizabeth, Yang, Sue-Ming. 2012. The Criminology of Place: Street Segments and Our Understanding of the Crime Problem. New York: Oxford University Press.
Weisburd, D., Groff, E. R., & Yang, S. M. (2014). Understanding and controlling hot spots of crime: The importance of formal and informal social controls. Prevention science, 15(1), 31-43. https://doi.org/10.1007/s11121-012-0351-9.
Wheeler, A. P., Gerell, M., & Yoo, Y. (2020). Testing the Spatial Accuracy of Address-Based Geocoding for Gunshot Locations. The Professional Geographer, 72(3), 398-410. https://doi.org/10.1080/00330124.2020.1730195
Wickham, H., François, R., Henry, L., and Müller, K. (2020). dplyr: A Grammar of Data Manipulation. R package version 1.0.2. https://CRAN.R-project.org/package=dplyr
Wästerfors, D., & Burcar Alm, V. (2020). “They are harsher to me than to my friend who is blonde”. Police critique among ethnic minority youth in Sweden. Journal of youth studies, 23(2), 170–188. https://doi.org/10.1080/13676261.2019.1592129

Information & Authors

Information

Published In

Go to Nordic Journal of Urban Studies
Volume 2Number 110 June 2022
Pages: 4062

History

Published online: 10 June 2022
Issue date: 10 June 2022

Authors

Affiliations

Manne Gerell [email protected]
Associate Professor, Department of Criminology, Malmö University
(Corresponding author)
MSc (Master of Science in Criminology), Department of Criminology, Malmö University
Nicklas Guldåker [email protected]
Associate Professor, Department of Human Geography, Lund University

Metrics & Citations

Metrics

Citations

Export citation

Select the format you want to export the citations of this publication.

View Options

View options

PDF

Download PDF

Get Access

Restore guest purchases

Enter your email address to restore your content access:

Note: This functionality works only for purchases done as a guest. If you already have an account, log in to access the content to which you are entitled.

Figures and Media

Figures

Media

Tables

Share

Share

Share the article link

Share on social media

Share on Messenger