Introduction

Streets, as critical spaces for public activities and transportation in urban areas, play a vital role in ensuring urban safety and stability. Enhancing the quality of streets is a significant focus in urban studies. Strategies for improving street quality include not only considerations of landscape aesthetics and pedestrian friendliness but also prioritizing safety and security, as these are fundamental requirements for pedestrian use of streets (Park & Garcia, 2020). However, the high density of daily activities and intensive pedestrian flows on streets provide potential targets and opportunities for criminal activities, making streets a preferred venue for offenders to commit crimes during their routine activities. This dynamic increases the likelihood of criminal incidents (Brantingham & Brantingham, 1993; Feng et al., 2016; Mao et al., 2021). Additionally, streets often serve as escape routes and places of refuge for offenders, underscoring their critical role in both the occurrence and prevention of urban crime.

Routine activity theory (Cohen & Felson, 1979) and crime pattern theory (Brantingham & Brantingham, 1993) suggest that there is a close relationship between people’s daily behavior patterns and the urban environment. As the primary spatial carriers of routine activities, the design and layout of streets significantly influence behavioral patterns, thereby closely linking spatial pathways and activity nodes (key destinations that attract human activities, such as transit hubs, commercial centers, and public facilities) to the occurrence of street crimes. In recent years, increasing attention has been given to the relationship between various street environmental elements—such as road networks, facilities, vegetation, housing, demographic, and socioeconomic characteristics—and criminal activities (Birks & Davies, 2017; Davies & Johnson, 2015; Johnson & Bowers, 2010). Among studies focusing on street network morphology, existing literature has primarily explored geometric features, such as street length, density, and hierarchy, as well as topological forms like permeability and accessibility, and their impacts on certain crime types (e.g., burglary, robbery, assault) (Mahfoud et al., 2020; Kelsay & Haberman, 2021; McCutcheon et al., 2016; Davies & Bowers, 2018). Nevertheless, studies that systematically examine the role of street configurational types, including street network patterns and intersection forms (see section “Elements of Street Spatial Environment”), in relation to a broader range of crimes remain limited.

Building on previous research, this study extends the current understanding by incorporating street configurational types into the framework of topological street morphology. Using assault, robbery, larceny, and motor vehicle theft as focal crime types, this study comprehensively examines the influence of both topological and geometric street characteristics on criminal activities. Furthermore, it investigates the potential impact of nodal features that may act as crime attractors or generators, as well as socioeconomic characteristics, aiming to enrich the research framework on the relationship between street environments and crime.

Literature review

Theoretical background

Routine activity theory posits that crime is more likely to occur when a suitable target lacking in capable guardianship converges in space and time with a motivated offender (Cohen & Felson, 1979). Therefore, which areas within the street network are more likely to become routes or nodes of pedestrian flow directly affects the likelihood of encounters between potential offenders and targets. Crime pattern theory suggests that offenders form cognitive maps of urban space through their routine activities and tend to choose familiar environments as crime targets (Hill et al., 1987). The morphology of the street network directly determines the accessibility, continuity, and selectivity of these cognitive maps, thereby shaping the locations where they are more likely to encounter or become aware of criminal opportunities (Brantingham & Brantingham, 1993).

Street networks play a crucial role in shaping how offenders perceive and choose crime opportunities. The activity nodes and paths within a street network significantly determine the spatial structure and specific settings of individuals’ daily activities. Activity nodes in the street network refer to locations where people linger for extended periods or engage in frequent activities, such as retail stores, restaurants, entertainment venues, places of worship, schools, and office buildings (Kim & Wo, 2023). These activity nodes can serve as crime attractors—places that create well-known criminal opportunities to which strongly motivated (e.g., popular shopping areas)—and crime generators—locations where large numbers of people converge for non-criminal purposes, unintentionally increasing the likelihood of crime due to high activity volumes (Brantingham & Brantingham, 1995). Paths within the street network connect these nodes and represent the built environment features that facilitate movement from one node to another. These may include railroads or train systems, public transit networks, vehicular roads, and pedestrian pathways (Kim & Wo, 2023). The routes that offenders frequently traverse influence their awareness of potential victims and, to some extent, constrain their choice of criminal activities, as their movements are limited by the configuration of the street network. In modern cities, streets function as the “skeleton” of urban areas, defining the layout and organization of built environment elements and determining patterns of daily activities. The street network, as the core structure connecting activity nodes and paths, dictates how people move within urban spaces, thereby influencing pedestrian and vehicular flow patterns (Davies & Johnson, 2015). Consequently, the street network serves as both the foundational framework of urban spatial structures and a key factor in crime occurrence and prevention, shaping the location, frequency, and distribution of criminal activities.

Additionally, street crimes are closely associated with demographic, social, and economic characteristics, such as population density, gender and age composition, education levels, travel patterns, and economic conditions (Wu et al., 2015; Yue et al., 2018). These socioeconomic factors interact with the structural attributes of the street network to jointly influence the spatial distribution and patterns of crime.

In summary, routine activity theory and crime pattern theory provide critical theoretical foundations for understanding criminal behavior from the perspective of street networks. Street crimes result from the interplay between spatial environments and behavioral agents, with both elements collectively shaping crime occurrences and spatial distributions. The interwoven effects of street networks, socioeconomic characteristics, and individuals’ daily activity patterns jointly determine the spatial characteristics and distributional dynamics of urban crime.

Street morphology and crime

The morphological characteristics of streets not only influence criminals’ perceptions of the spatial environment and opportunities for crime but also guide the movement of both criminals and potential victims within the space (Kim, 2018). Street morphology generally includes geometric and topological forms.

Geometric forms are intuitive and easily perceived aspects of street morphology, mainly encompassing density, length, width, and hierarchical attributes. These features are often interrelated—for example, wider streets are typically part of higher-order street hierarchies (e.g., arterials or main roads), while narrower streets are more common in lower-order networks. High-density, short-length street networks tend to form compact neighborhoods with more intersections, leading to higher frequencies of pedestrian (Ewing & Cervero, 2010; Frank et al., 2006). According to Jane Jacobs’ “eyes on the street” theory, frequent pedestrian activity and engagement enhance street surveillance, thereby reducing crime (Jacobs, 1961).

Conversely, low-density areas or streets with greater length are often associated with larger blocks and fewer intersections. This discourages pedestrian activity, weakens natural surveillance, and creates more opportunities for criminal behavior. Studies have shown that street length is significantly positively correlated with burglary incidents (Yue et al., 2018; Wu et al., 2015). However, the relationship between street network density and crime varies by crime type. For instance, street density has been found to be positively correlated with violent crimes, such as assault (Du & Law, 2016), but negatively correlated with property crimes (Ye et al., 2018). Street width also plays a critical role. As Jan Gehl emphasized in Cities for People, street width directly influences traffic speed and the frequency of pedestrian activities, which in turn affect urban safety and vitality (Gehl, 2010). Similarly, street hierarchy impacts pedestrian volume and natural surveillance density, which are key determinants of crime likelihood. Studies on the impact of street hierarchy on crime rates have yielded varying results. For example, Agnew explored the effect of motorway expansion on burglary rates, finding that areas closer to motorway networks might experience more burglaries due to easier escape routes for criminals and a lower likelihood of police capture (Agnew, 2020). Other studies have found that major roads are significant crime hotspots (Wuschke et al., 2021). In China, the risk of residential burglary is higher on side roads compared to other types of road sections (Wu et al., 2015).

In summary, the geometric forms of streets influence crime occurrence both directly and indirectly by affecting pedestrian and vehicular flows, natural surveillance, and social interactions. Street hierarchy also plays an important role in shaping the likelihood of different crime types. Therefore, when investigating the relationship between street density and crime, it is essential to consider variations in road hierarchy and crime type to ensure a more comprehensive understanding.

Topological forms mainly include more abstract configurational features such as accessibility, connectivity, and permeability. Existing research has found that crime is correlated with the topological forms of urban environments. Accessibility describes the ease with which a street can be reached within a network. Research suggests that property crimes are most likely to occur on streets with high traffic volumes and high accessibility, as these streets attract more potential targets (Beavon et al., 1994). Connectivity refers to the number of street segments directly connected to a given segment within the street network, and in space syntax, it measures how many segments can be reached without turning, with higher connectivity indicating better local integration and greater potential for pedestrian and vehicle movement. Mahfoud et al. (2020) found that streets located at the network center with higher connectivity are expected to have higher burglary rates due to greater pedestrian and vehicular traffic flows. Permeability refers to the ease with which pedestrians and vehicles can access a street network, thereby modulating crime risks by influencing how people enter an area (Johnson & Bowers, 2010; Weiner & Jeffery, 1971). High-permeability streets often encourage greater pedestrian and vehicular flows, enhancing natural surveillance and reducing crime rates (Rudlin & Falk, 1999). Conversely, low-permeability streets are considered more dangerous because they often contain crime opportunities that lack adequate natural surveillance, increasing the risk of criminal activity (Beavon et al., 1994). However, Newman (1972) argued that low-permeability streets are generally safer because they restrict access to non-residents, allowing neighbors to better monitor their community. Additionally, Armitage (2007) and Johnson & Bowers (2010) found that high-permeability street networks could increase the risk of residential burglary. These differing perspectives highlight the duality of permeability’s impact on crime.

Betweenness centrality, a key measure of street permeability, has been widely used in research on the spatial distribution of crime (Yue et al., 2018). Numerous studies have explored the relationship between betweenness centrality and crime. For instance, Davies & Johnson (2015) and Musah (2020) found that streets with high betweenness centrality often correlate with higher burglary rates. Davies & Johnson (2015) demonstrated the persistent importance of “betweenness” at various radii in influencing human behavior patterns. This underscores the need to further investigate the relationship between mobility patterns at different scales and crime. Frith (2017) suggested that the scale of the radius could modulate the impact of betweenness on crime—smaller radii better reflect neighborhood-level activities, while larger radii align with city-level transportation network functions. Kim & Hipp (2020) analyzed betweenness centrality within 1/4-mile and 1-mile radii to capture local and non-local pedestrian flow patterns and incorporated poverty levels as an interaction term. Their results showed that in high-poverty areas, increased non-local betweenness initially led to higher violent crime rates, which then decreased. In contrast, in streets with better living conditions, non-local betweenness only had a crime-enhancing effect.

The type of street intersection is a direct structural expression of betweenness centrality and also reflects street permeability to a considerable extent. For example, open intersections (e.g., cross junctions) typically connect streets in four directions and serve as convergence points for multiple routes, significantly increasing route choice and spatial accessibility, thereby enhancing the overall permeability of the street network. In contrast, T-junctions restrict directional flow, and cul-de-sacs (or dead-end streets), which connect in only one direction and lack through-connectivity, reduce the continuity and accessibility of the network. This suggests that intersection types not only reflect physical morphological characteristics but also determine the locational attributes and passability of a given street segment within the overall network structure. Johnson & Bowers (2010) found that cul-de-sacs are generally safer as they limit non-resident access and enhance natural neighborhood surveillance, outcomes that are inherently shaped by their distinctive network configuration. Hillier (2004) noted that simple, straight cul-de-sacs connected to main roads can serve as safe spaces, but when cul-de-sacs form complex systems, they may become high-risk areas for crime. This underscores the need to balance permeability and safety in street network design by considering the combination and functionality of different intersection types. Despite these findings, systematic research on the relationship between street intersection types and crime remains scarce. In particular, the quantitative measurement of intersection types has not been fully explored. Drawing on Marshall’s Street and Patterns, this study adopts four indicators—X-junction ratio, T-junction ratio, cell ratio, and cul-de-sac ratio—to quantify and describe the combination of intersection types systematically. The study attempts to explore the influence of these configurational morphology indicators on crime.

The relationship between street networks and crime is complex and multidimensional. Although previous studies have explored the connection between street network morphology and crime, most have been limited to single-indicator analyses or simplistic combinations of street network morphology and demographic characteristics. Failure to account for other potential influencing factors (e.g., surrounding environments and socioeconomic conditions) may result in spurious correlations between street characteristics and crime (Summers & Johnson, 2017). Moreover, existing research on street networks and crime has rarely focused on the relationship between street configurational types and crime. This limitation hinders a deeper understanding of the relationship between street spatial environments and criminal behavior. Therefore, a comprehensive and systematic analysis of the relationship between street environments and crime requires the integration of multiple factors and the adoption of a more holistic research framework.

Nodes and crime

Activity nodes are facilities located along streets that attract a large volume of daily activities, consequently bringing in both potential criminals and crime targets. Conversely, these nodes can also reduce crime by deterring potential offenders. According to their functions, activity nodes can be classified into three types: crime generators, crime attractors, and crime inhibitors (Brantingham & Brantingham, 1993). Crime generators are public spaces that are easily accessible and attract large crowds. These locations often become crime hotspots due to frequent human movement and congregation, providing ample opportunities for criminal activities. Examples of such facilities include banks, restaurants, hotels, parking lots, and bus stations (Haberman & Kelsay, 2020; Liu & Du et al., 2017; Zeng et al., 2021).

Crime attractors, in contrast, may not always draw large crowds but are functionally suited for motivated offenders to identify targets. These locations are often hotspots for illicit activities, including bars, nightclubs, liquor stores, and recreational venues (Lee et al., 2021; Sytsma et al., 2021). Such places are ideal for crime, partly due to the prevalence of cash transactions and the reluctance of many victims to report incidents to the police. In contrast, police stations and community policing centers act as crime inhibitors, effectively reducing the incidence of crime by deterring potential offenders (Bernasco & Block, 2011). The presence of these facilities enhances surveillance and law enforcement response, creating an environment less conducive to criminal activities.

Social and economic attributes and crime

Street spaces accommodate people’s daily activities, thereby influencing the occurrence of crime. The street network facilitates and restricts population movement, shaping the daily activity patterns of urban residents (Mahfoud et al., 2020). Additionally, differences in individuals’ socioeconomic attributes further impact the characteristics of daily activities on the streets, such as family size, gender, racial disparities, educational attainment, and the proportion of the unemployed population (Liu et al., 2018; Wu et al., 2015). Gender and age composition, as the most common indicators, are often used to characterize the risk of crime (Heidensohn, 1989; Yue et al., 2018). For instance, research suggests that the age range of 15-30 is the peak age range for offenders (Wu et al., 2015), and males are considered the primary perpetrators of crime (Heidensohn, 1989). Beyond gender and age, racial composition and education levels also correlate closely with crime. Higher education levels and specific racial characteristics may influence community vitality, thereby affecting crime frequency (Liu et al., 2018). For example, neighborhoods with a higher percentage of well-educated residents tend to report more residential burglary cases (Wu et al., 2015). Furthermore, racial disparities are closely tied to social inequality, and studies have found that higher levels of inequality often result in higher crime rates (Hipp, 2007).

Beyond demographic characteristics, socioeconomic conditions also have significant associations with crime. Low income and high unemployment rates are often viewed as indicators of poverty and social inequality, strongly correlated with crime risk and crime rates in impoverished areas (Sampson et al., 1997). Regarding the relationship between unemployment and crime, two contrasting perspectives exist in academic research. One perspective suggests that unemployment promotes crime because unemployed individuals, reliant on economic conditions for their livelihoods, are likely to become potential offenders (Cantor & Land, 1985; Yue et al., 2018). However, another viewpoint, from the perspective of routine activity theory, suggests that unemployed individuals staying at home can act as natural surveillance, potentially reducing crime to some extent (Paternoster & Bushway, 2001).

Housing vacancy rates and rental rates also influence street mobility and accessibility, indirectly affecting crime occurrence. Studies indicate that areas with higher housing vacancy rates tend to experience higher crime rates (Lin et al., 2022), while the relationship between rental rates and crime remains inconclusive (Hegerty, 2017). Moreover, the occurrence of violent crimes is associated with the number of vacant houses and the diversity of land use within a region (Sparks, 2011). The relationship between housing vacancy rates and property crimes is more pronounced than that with violent crimes (Gray & Parker, 2023).

In addition to socioeconomic factors, residents’ travel modes can also influence criminal activities. Specifically, travel patterns are both a result of crime rates and a factor influencing crime occurrence. Communities with high-crime rates often see residents reducing their frequency of walking or cycling, instead relying more on driving or public transportation (Ferrell & Mathur, 2012). This shift decreases street-level crowd density and public visibility, potentially providing more opportunities for criminal behavior. Walking activity is positively correlated with violent crime (Reid et al., 2024), suggesting that increased pedestrian activity may be closely linked to the occurrence of violent crimes. Furthermore, in Heinen’s (2023) study, many respondents reported experiencing crime-related incidents during travel. Specifically, the proportions of victims by travel mode are 35% for walking, 11% for cycling, and 24% each for driving and public transportation. These findings further validate the relationship between travel patterns and criminal incidents, highlighting that different travel modes may influence crime types and frequencies.

In summary, street crime is closely related to the socioeconomic characteristics of populations, with varying degrees of association across different attributes. However, given the diversity of social conditions in different cities, existing studies lack comprehensive consideration of how urban socioeconomic attributes collectively impact crime risks.

Current study

The literature review reveals that street morphology, activity nodes, and socioeconomic attributes all influence crime. However, existing research presents inconsistent findings, particularly regarding the impact of permeability on crime. These discrepancies are closely tied to the complexity of street environments. One major reason is that most studies focus on specific spatial elements within streets, such as individual indicators of street morphology, nodes, or certain facility points, while rarely considering the combined effects of street network morphology, activity nodes, and socioeconomic characteristics on crime.

To address this gap, this study aims to investigate how street environmental factors influence criminal behavior when incorporating street network morphology elements. In this research, street network morphology refers to compositional, constitutional, and configurational characteristics of street networks, including geometric features (e.g., road network density and hierarchy) and topological features (e.g., the number of intersections, permeability, and the less-explored street configurational types). These characteristics reflect the organization of street networks in spatial configurations and traffic flows (Marshall, 2005; Hillier & Iida, 2005). The concept of the street environment integrates street network morphology, node features, and socioeconomic attributes, providing a comprehensive representation of how streets and their surrounding environments influence crime mechanisms.

This study builds its framework (Fig. 1) on routine activity theory and crime pattern theory, incorporating three dimensions: street network morphology, activity nodes, and socioeconomic attributes.

Fig. 1
figure 1

Research framework.

Full size image

Data and methods

Study region

Detroit is the largest city in the state of Michigan, USA, with an area of approximately 370.2 km². Scarce economic opportunities in Detroit have led to an annual unemployment rate of 14.2% from 2000 to 2022 (Johncox, 2023), with 31.8% of the population living below the poverty line (equivalent to 202,000 out of 635,000 individuals) (Detroit, 2023b).

It is estimated that Detroit, being one of the most dangerous cities in the United States, experiences a crime incident every three minutes, ranking 17th among the cities with the highest crime rates globally. On average, there are 20.59 violent crimes and 45.50 property crimes per 1000 residents in the Detroit area (Detroit, 2023a). Residents of Detroit generally consider the northern part of the city to be the safest. In the southeastern region of Detroit, the annual likelihood of becoming a crime victim may reach as high as 1 in 13, compared to 1 in 71 in the northern areas (CrimeGrade, 2024). Considering that most of the population and socioeconomic data provided by the Census Bureau are at the block group level, in order to accurately integrate population and socioeconomic indicators into spatial locations, block groups are chosen as the analytical units. This study includes a total of 879 block groups (Fig. 2).

Fig. 2
figure 2

Block group and city boundary in Detroit, Michigan.

Full size image

Address-level crime data

The crime incident data used in this study were geocoded using the City of Detroit’s internal GIS infrastructure, which converts reported addresses (typically street-level addresses or intersections) into geographic coordinates (latitude and longitude) through an address-based geocoding process. According to the dataset documentation from the Detroit Open Data Portal, this system geocodes incidents to their actual reported locations rather than relying on spatial approximations such as block centroids.

While minor spatial inaccuracies may occur, particularly for crimes reported near intersections, alleys, or other non-standard locations, the average positional error is typically within 10 to 20 meters, which is acceptable for neighborhood-scale spatial analysis. To further ensure spatial validity, we excluded crime records with missing, implausible, or out-of-bound coordinates, such as those falling outside city boundaries or within non-residential zones like industrial areas or bodies of water.

This study focuses on street-level crimes and includes two major categories based on offense type: violent crimes (assaults and robberies) and property crimes (larcenies and vehicle thefts). These categories account for the majority of reported crime incidents in Detroit and are particularly relevant for evaluating environmental or spatial risk patterns (Fig. 3).

Fig. 3
figure 3

Number of crimes by type in 2019 (ranked from high to low).

Full size image

Elements of street spatial environment

The street spatial environment in this study comprises two aspects: street network morphology and activity nodes. The street network data for Detroit is from the Open Street Map (OSM) website for the year 2019.

Geometric features primarily examine overall road network density (the total length of streets per unit area) and the proportional lengths of roads by classification, including major roads, minor roads, and local roads (calculation methods are detailed in Table 1). According to the encoding of the United States Census Bureau’s vector road network data, major roads include limited-access highways, overpasses, and highway ramps; minor roads include community streets and urban streets; local roads encompass service roads, alleys, lanes, and private roads.

Table 1 Environmental elements of the street space.
Full size table

Topological features consist of multiple types, including street intersection types, street network patterns, and road network permeability. The specific definitions are as follows:

  1. (1)

    Street intersection types are classified into “T-junctions” and “X-junctions.”

  2. (2)

    Street network pattern is composed of four indicators proposed by Marshall: X ratio, T ratio, Cell ratio, and Cul ratio. It includes “X-cell”, “T-cell”, “T-tree”, and “X-tree” configurations (Marshall, 2004), as depicted in Fig. 4.

    Fig. 4
    figure 4

    Street configurational types. (Note: The characteristics of the street network patterns are 397 organized based on Marshall’s work (Marshall, 2005).

    Full size image

  • X Ratio and T Ratio: The X ratio and T ratio represent the proportions of different types of intersection nodes in the street network. The X ratio represents the proportion of four-way junctions to the total number of intersections, while the T ratio represents the proportion of three-way junctions to the total number of intersections. In a street network consisting solely of X-junctions and T-junctions, the sum of the X ratio and T ratio equals 1.

  • Cell Ratio and cul Ratio: The cell ratio and cul ratio, respectively, indicate the proportions of closed-loop units and dead-end roads in the street network. The cell ratio represents the proportion of closed-loop cells to the total number of cells and cul-de-sacs combined, while the cul ratio represents the proportion of cul-de-sacs to the same total. The sum of the cell ratio and cul ratio equals 1.

Street network patterns provide an initial understanding of the spatial layout and characteristics of street networks. For example, networks with cell-based layouts generally offer better accessibility compared to tree-like networks, making them more convenient for traffic and usage. Additionally, networks dominated by X-junctions tend to be more regular compared to those dominated by T-junctions.

Street permeability is characterized by betweenness centrality, a network measure introduced by Freeman (1977) that quantifies the relative importance of nodes or segments by calculating the frequency with which they lie on shortest paths between other nodes (Freeman, 1977). The calculation formula is:

$${B{etweenness}}_{i}^{r}=\mathop{\sum}\limits _{j\in N-\left\{i\right\}\,:\,d\left[i,j\right]\le r}\frac{{n}_{jk}\left[i\right]}{{n}_{{jk}}}$$
(1)

In Equation (1), \({B{etweenness}}_{i}^{r}\) represents the betweenness centrality of street segment i within a search radius r, \({n}_{{jk}}\) denotes the total number of shortest paths between street segments j to k, while \({n}_{jk}\left[i\right]\) indicates the number of shortest paths from street segment j to k that pass through street segment i. Considering different modes of travel or transportation, conventional maximum travel distances are taken into account. The average betweenness centrality is calculated with radii of 800 m, 1500 m, and n (equivalent to infinity) to represent pedestrian, bicycle, and car travel modes, respectively. Higher betweenness centrality values indicate that a street segment functions as a key connector or “bridge” within the network, frequently used in shortest routes between other locations. This reflects the segment’s strategic importance in facilitating pedestrian and vehicular movement. In this study, we use betweenness centrality as a proxy for street permeability, as it captures how accessible and traversable a street segment is within the network. This indicator is computed using the Spatial Design Network Analysis (sDNA) tool. The calculation provides a betweenness value for each street segment. These values are then weighted by the lengths of the respective street segments to compute the average betweenness for all street segments within each block group. The result is the overall betweenness centrality for the block group.

The node data is downloaded from the OSM website in 2023 and includes Points of Interest (POI) data relevant to crime, comprising 10 types of nodes: ATMs and banks, bus stops, bars, parking lots, restaurants, leisure facilities, hospitals, schools, places of worship, and police stations. The time frame for this data is 2023. Since leisure facilities, schools, places of worship, and police stations are absent in most units or typically only present in one, they are treated as dummy variables. A value of 0 indicates the absence of such a facility within the block group, while 1 indicates its presence. The measurement unit for the other six types of facilities is either count or frequency. It is important to note that the POI data reflect the post-COVID-19 period, during which some minor changes in the distribution and operation of facilities may have occurred. For example, certain leisure facilities, bars, and restaurants experienced temporary closures or operational restrictions during the pandemic, which could have influenced their presence or activity levels recorded in 2023. However, given the relatively stable urban fabric of mature cities like Detroit, these changes are expected to be limited in scope and unlikely to significantly affect the overall spatial patterns observed.

Street socioeconomic attributes

Socioeconomic attributes in this study refer to a set of composite indicators describing the socioeconomic conditions of surrounding areas. Population data are from the U.S. Census Bureau for the year 2019 (U.S. Census Bureau, 2019).

Specifically, population characteristics include racial composition (White, African American, and Asian populations), the number of residents with higher education (holding a bachelor’s degree or above), low-income individuals (below the poverty line), and populations using different travel modes (car, public transit, or walking). Socioeconomic conditions include the unemployment rate, housing rental rate, and housing vacancy rate. While the unemployment rate, rental rate, and vacancy rate are expressed as percentages, all other indicators are measured in absolute numbers (persons).

Research method

The study first employs kernel density estimation to describe the spatial distribution patterns of assault, robbery, larceny, and vehicle theft crimes in Detroit for the year 2019. Recognizing that the spatial distribution of crime is closely related to walking distances and pedestrian flows (Gilderbloom & Riggs, 2015; Ackerman & Rossmo, 2015), an 800-meter radius is selected as the KDE bandwidth. This distance approximates pedestrian accessibility, covering essential services and public activity spaces in daily life, thereby simulating the spatial diffusion of criminal activities. Subsequently, Moran’s I is used to evaluate the clustering degree of crime points. Table 1 presents the set of street network spatial environmental factors, encompassing street network morphology, activity nodes, and socioeconomic attributes.

To analyze the relationship between street spatial environment and various types of crimes, regression analysis is employed to investigate the influence of street environmental characteristics on crime. Since the dependent variables represent the counts of different types of crime incidents, which are count variables, a count model (Poisson or Negative Binomial) is employed. Furthermore, due to the presence of overdispersion, the Negative Binomial Regression model is ultimately selected for analysis. Since this study focuses on street crimes, with all incidents occurring on streets, and the total street length within block groups varies significantly, the total street length in each block group is included as an exposure variable to model crime rates. Referring to Stata’s operating manual (StataCorp, 2023), it is more appropriate to model the incidence than the count directly when the exposures differ by a relatively large amount. Prior to conducting Negative Binomial Regression analysis, multicollinearity among explanatory variables is assessed, and eventually, 30 variables are included in the model, with variance inflation factor (VIF) values all below 10.

Results

Descriptive statistics

Table 2 presents the descriptive statistics of the variables. The mean number of assault cases across Detroit neighborhoods is the highest, followed by larceny, vehicle theft, and robbery. This indicates that assault crimes had the highest total number of incidents in Detroit in 2019. The standard deviations, from largest to smallest, are for larceny, assault, vehicle theft, and robbery, indicating that the variation in crime numbers across neighborhoods is greatest for larceny crimes, while robbery and vehicle theft crimes have relatively smaller variations between neighborhoods.

Table 2 Descriptive statistics of variables (N = 879).
Full size table

The road types within Detroit neighborhoods predominantly consist of minor roads, with the average proportion of minor road length reaching 92.8%. This is followed by major roads and local roads. The intersections are mainly X-shaped and T-shaped. Among the activity nodes, bus stops have the highest average number per neighborhood, followed by restaurants, parking lots, hospitals, bars, ATMs, and banks. The number of bus stops within each block group shows greater variability compared to other facilities.

Regarding elements of routine behavioral factors, the African American population in Detroit is the largest, with an average of 601.35 individuals per block group, and the variation in African American population numbers across neighborhoods is quite large. The Asian population is the smallest, with an average of only 12.84 individuals per block group. On average, there are 1130 individuals with a bachelor’s degree and 2640 individuals with low income per block group; the unemployment rate is 16.64%, with some block groups having rates as high as 78.13%. In terms of transportation, most people travel by car, with fewer individuals using other modes of transportation. Additionally, the average housing vacancy rate is 29.44%, and the average housing rental rate is 48.93%.

Spatial analysis of crime hotspots

The distribution of the four types of crime, as calculated through kernel density estimation, reveals positional differences in these criminal activities across various block groups and street segments (Fig. 5). The results indicate that the most prominent hotspots for assault, robbery, larceny, and vehicle theft are located in the central southern region (Central Business District South, University South, Downtown, and Greektown). Satellite maps show that this area serves as a hub for commercial and social interactions, attracting higher foot traffic. Factors such as high population density, frequent commercial activities, and significant mobility increase the opportunities for criminal activities, making this region the most concentrated area for crime. Larceny crimes display a marked clustering trend, with the main hotspot in the southern region, while other residential or industrial areas exhibit relatively lower crime densities. In comparison, assault crime hotspots cover a larger area, forming broader zones predominantly in the western and northeastern regions. Secondary hotspots for motor vehicle theft are more scattered and cover smaller areas, primarily located in the northwestern region, as well as in the central (Boston Edison, Rosa Parks East) and eastern (Denby and East English Village) neighborhoods. Additionally, due to the smaller number of robbery incidents compared to the other three crime types, secondary hotspots for robbery are not visually distinguishable on the map.

Fig. 5: Spatial distribution of four types of crime hotspots in 2019, based on kernel density estimation.
figure 5

a shows assault hotspots mainly in northern and central areas; b shows robbery hotspots scattered across the city; c highlights larceny clusters concentrated in the downtown core; d displays stolen vehicle hotspots in central and northeastern neighborhoods.

Full size image

To further investigate the spatial clustering of various types of crime, Moran’s I is employed to conduct a spatial autocorrelation analysis for the four types of crime within the study area (Table 3). The results show that the Moran’s I indices for all four types of crime are greater than 0, indicating a positive spatial correlation. This suggests that the crime data for all four categories exhibit spatial clustering. Ranking the z-values from largest to smallest reveals that the degree of spatial clustering, from strongest to weakest, is larceny, vehicle theft, assault, and robbery.

Table 3 Spatial autocorrelation analysis of four types of crimes in 2019.
Full size table

Local autocorrelation analysis is employed to investigate the clustering of crimes in specific spatial contexts. Figure 6a–d visually displays the significant and non-significant spatial autocorrelation distributions for assault, robbery, larceny, and vehicle theft crimes, respectively, highlighting regions of relative importance and insignificance for these crimes. For assault crimes, there are five regions exhibiting high-high clusters, where high-crime rates are surrounded by high-crime rates (i.e., spatial autocorrelation), though these are somewhat scattered. Robbery and larceny crimes each show only one region with high-high clustering, primarily located in the southeastern area. Vehicle theft crimes have two regions with high-high clustering, situated in the northern and southeastern areas. Overall, the four types of crimes are primarily concentrated in the central southern part of the city, which is Detroit’s central commercial and administrative district. This area includes many public transportation stations, parks, businesses, and public service facilities, resulting in high traffic and pedestrian volumes. The complex social environment of this region, characterized by features that attract and facilitate crime, makes it more prone to higher crime rates.

Fig. 6: Cluster and outlier analysis of four crime types using Local Moran's I statistic.
figure 6

a shows five high–high clusters (HH) of assault; b shows one significant HH cluster for robbery; c highlights a dense HH larceny area in the central region; d presents HH clusters of stolen vehicle thefts along key corridors. All HH areas are outlined in black boxes.

Full size image

Spatial distribution of street network configurational types

Figure 7 illustrates the spatial distribution of street network morphology values across block groups in Detroit. The results reveal that most block groups exhibit high cell ratios, generally above 0.8. Areas with low cell ratios (and correspondingly high cul ratios) are primarily concentrated in the southeastern and southern regions, with sporadic occurrences in other areas (Fig. 7a). Block groups with a high X ratio (and low T ratio) for intersection composition are mainly distributed in the western, northwestern, and northeastern areas. Conversely, areas with a low X ratio (and high T ratio) are concentrated in the central and southern regions (Fig. 7b). Further analysis indicates that road networks in most block groups are dominated by “T-cells” and “X-cells”, with fewer instances of “T-trees” and “X-trees”.

Fig. 7: Spatial distribution of street network morphology indicators.
figure 7

a Cell ratio; b X ratio; c intersection count; d betweenness centrality (R = 800); e betweenness centrality (R = 1500); f betweenness centrality (R = n) (Note: for (a, b), a higher cell ratio indicates a lower cul ratio, and a higher X ratio indicates a lower T ratio. The cul ratio and T ratio are not displayed here).

Full size image

Block groups with a high number of intersections are primarily located in the southern central region (Fig. 7c). A comparison with Fig. 5 reveals that these areas overlap significantly with crime hotspots. Regarding permeability, the road network under walking travel mode exhibits a spatial pattern of decreasing permeability outward from a central peak in the southern region (Fig. 7d). In contrast, permeability under public transit and automobile travel modes does not show a clear trend and is generally at lower levels, with high-permeability areas scattered sporadically (Fig. 7e, f).

(Note: For Fig. 7a, b, a higher cell ratio indicates a lower cul ratio, and a higher X ratio indicates a lower T ratio. The cul ratio and T ratio are not displayed here.)

The relationship between street spatial elements and crime

Table 4 presents the outcomes of the negative binomial regression analyses examining the associations between street spatial features and four crime types: assault, robbery, larceny, and vehicle theft. To ensure comparability, population-related variables were rescaled to units of 1000 individuals.

Table 4 The results of the negative binomial regression model for street network spatial environment indicators and the number of four types of crimes.
Full size table

Road density is significantly and negatively associated with both assault and larceny. A 1 km/km² increase in road density corresponds to a 2.8% reduction in the assault rate and a 3.2% reduction in the larceny rate, holding all other variables constant. The proportion of minor roads shows a consistently positive association with crime: a 0.1-unit increase in this proportion is associated with increases of 10.4%, 11.1%, 4.3%, and 8.4% in assault, robbery, larceny, and vehicle theft, respectively. In contrast, the proportion of local roads shows a significant inverse relationship with larceny. A 0.1 increase in local road share is linked to a 22.5% reduction in the larceny rate, suggesting a potential protective effect.

In terms of topological attributes, the cell ratio exhibits significant positive associations with assault, robbery, and larceny, with stronger effects observed on assault and robbery. No significant link is observed between the cell ratio and vehicle theft. The X ratio is negatively correlated with assault, robbery, and larceny, and positively correlated with vehicle theft; however, none of these results are statistically significant. The number of intersections shows a negative association across all four crime types. Each additional intersection is linked to decreases of 1.6%, 1.8%, 1.2%, and 1.2% in assault, robbery, larceny, and vehicle theft, respectively. Moreover, block groups characterized by high betweenness centrality (R = 800) tend to exhibit elevated rates of all four crime types, indicating that highly traversed street networks may facilitate increased crime opportunities.

For activity node indicators, the number of dining establishments is positively associated with assault. Each additional dining venue corresponds to a 5.5% increase in the assault rate. Robbery appears more sensitive to multiple node types. Bus stops and dining venues are both positively associated with robbery, with increases of 3.4% and 11.8%, respectively. Interestingly, the number of bars is negatively associated with robbery, with each additional bar reducing the robbery rate by 17.8%. Larceny is also influenced by the presence of bus stops and dining establishments, which are associated with increases of 1.4% and 6.1%, respectively. Additionally, neighborhoods featuring recreational facilities experience larceny rates that are 124.6% higher than those without. For vehicle theft, dining establishments contribute to a 6.7% increase, while the number of parking lots has a mitigating effect, decreasing the vehicle theft rate by 6.1% per additional lot.

Socioeconomic variables also play an important role. Assault is positively associated with the number of African American residents, car commuters, pedestrians, and low-income individuals. A 1000-person increase in each group is linked to multiplicative increases of 1.408, 1.923, 11.079, and 1.406 in the assault rate, respectively, with the walking population exhibiting the greatest influence. The number of residents with a bachelor’s degree demonstrates a dual effect: negatively associated with assault and robbery, but positively associated with larceny and vehicle theft. For vehicle theft in particular, a 1000-person increase in this group elevates the crime rate by a factor of 2.840. Ethnic composition shows further variation, as African American populations are associated with increases in vehicle theft, whereas higher numbers of Asian residents are linked to significant reductions, lowering the expected rate by more than half (rate ratio: 0.479).

Housing characteristics further clarify area-level crime patterns. The vacancy rate is inversely related to assault, larceny, and vehicle theft, with each 1% increase corresponding to reductions of 0.3% to 0.4% in these crime rates. By contrast, the rental housing rate is positively associated with all four crime types. A 1% increase in rental rates corresponds to rises of 0.5%, 0.5%, 0.6%, and 0.2% in assault, robbery, larceny, and vehicle theft, respectively.

Discussion

Our study constructs an environmental factor set from three aspects: street morphological characteristics, activity nodes, and routine behavior subjects to explore the impact of the street environment on violent and property crime incidents in Detroit. The results indicate that the influence of street environmental factors varied among assault, robbery, larceny, and vehicle theft. Specifically, higher road density is associated with lower incidence rates of assault and larceny. This finding aligns with previous studies (Liu & Du et al., 2017; Mao et al., 2021; Mao & Ding, 2014). Increasing road density brings more pedestrian and vehicular traffic and more roadside buildings, enhancing natural surveillance and deterring criminal behavior. Thus, while increased foot traffic may slightly elevate opportunities for crime, the suppressive effect of improved natural surveillance outweighs the additional opportunities. The study also found that the proportions of different road types significantly influence crime. The proportion of minor roads had a significant positive impact on all four types of crime, whereas the proportion of local roads was statistically associated with a negative impact on larceny only. Although the impacts on the other three crime types are not statistically significant, they also showed a negative trend. This disparity may stem from the prevalence of minor roads in Detroit’s block groups. As the primary conduits for daily travel, minor roads facilitate greater social interaction. However, due to staggered travel patterns, natural surveillance on these roads is not concentrated at a specific location or time. Consequently, when motivated offenders, suitable targets, and the absence of capable guardians converge in space and time, crimes are more likely to occur. In contrast, local roads, which primarily comprise alleys and private roads, are less common and provide greater privacy. This aligns with the environmental characteristics conducive to larceny, which often requires concealment.

Among the topological features of street networks, particularly the quantifiable configurational metrics proposed by Marshall, we found that a high cell ratio positively influences violent crimes such as assault and robbery but has no significant impact on motor vehicle theft. This discrepancy may stem from the differing opportunity structures and behavioral patterns of offenders. Violent crimes typically rely on proximity and rapid escape routes, making them more likely to occur in areas with a high cell density. In contrast, motor vehicle theft depends more on environmental concealment and the offender’s selective actions. Thus, while areas with high cell ratios provide more opportunities for crime, these opportunities do not affect all crime types equally. The study did not find significant effects of the X ratio on the rates of assault, robbery, larceny, and motor vehicle theft. This contrasts with prior research conducted in Shanghai (Mao et al., 2022), which showed significant correlations between street network morphology and larceny. The difference may be attributed to the fact that the Shanghai study focused solely on the relationship between street network features and theft crimes, whereas this study also incorporated activity nodes and socioeconomic factors. When accounting for additional environmental factors, some influences may become statistically insignificant. Regarding the relationship between permeability and crime, most existing studies have focused on residential burglary (Armitage, 2007; Davies & Johnson, 2015; Johnson & Bowers, 2010; Yue et al., 2018). This study extended the analysis to examine the relationship between permeability and four types of crime: assault, robbery, larceny, and motor vehicle theft. The results show that only road network permeability within an 800-meter radius (representing walking travel mode) has a positive effect on all four crime types. This finding suggests that neighborhoods with greater overall walkability tend to experience higher rates of assault, robbery, larceny, and motor vehicle theft. This result differs from previous studies, which have highlighted varying effects of small- and large-radius permeability on different crime types.

Regarding activity nodes, studies have shown that street robbery crimes increase in areas near grocery stores, shops, dining facilities, ATMs, and banks because cash payments are more common at these locations, making customers carrying cash attractive targets (Bernasco et al., 2016; Bernasco & Block, 2011; Haberman & Ratcliffe, 2015). Similarly, this study found that areas with a higher number of dining establishments also experience increased assault, robbery, larceny, and vehicle theft crimes, though no significant relationship between ATMs/banks and robbery/larceny crimes was found. This positive association may be partly explained by the higher pedestrian flow generated by restaurants, which increases the number of potential targets and reduces the ability of natural surveillance, especially during peak hours. In line with routine activity theory, the concentration of people in these locations enhances the likelihood of encounters between motivated offenders and suitable targets in the absence of capable guardianship. Additionally, this study identified that areas with more bars experience more robbery crimes, as intoxicated customers are more vulnerable targets, and other intoxicated individuals cannot act as guardians (Bernasco & Block, 2011; Conrow et al., 2015). Furthermore, bus stops are found to have a significant positive impact on robbery and property crimes, confirming previous research conclusions that bus stops are considered crime attractors for property loss (Zahnow & Corcoran, 2019). Other studies consider bus stops as crime generators due to the high daily pedestrian flow, which increases the presence of attractive targets (Lee et al., 2023; Liu & Jiang et al., 2017; Zhang et al., 2024). Areas with more parking lots are less likely to experience vehicle theft crimes, possibly due to centralized vehicle management and the presence of surveillance cameras and security personnel, making it difficult for criminals to commit crimes.

Regarding the relationship between the socioeconomic attributes of daily activity agents and crime, the study results indicate that an increase in the number of African Americans is significantly positively associated with assault and motor vehicle theft crimes. This finding contrasts with a study on Detroit that suggests an increase in both White and Black populations helps significantly reduce crime in neighborhood groups (Lin et al., 2022). This discrepancy can be attributed to several factors. On one hand, the differences may stem from the different types of crimes examined in the two studies. On the other hand, the study by Lin et al. (2022) primarily used the SAR model to analyze the effects of population and socioeconomic factors on crime, without considering the role of spatial environmental factors in the urban context. If spatial factors have a significant influence on crime, their omission in a model that focuses only on demographic and socioeconomic characteristics could obscure the actual effects of these environmental variables. This could lead to spatial autocorrelation in the relationship between socioeconomic characteristics and crime, as spatial factors would not be accounted for in the analysis. Additionally, the number of people with a bachelor’s degree or higher has a significant positive impact on larceny and vehicle theft crimes. Further data analysis reveals a significant correlation between the proportion of bachelor’s degree holders and high-income population (r = 0.667, sig = 0), as wealthier individuals often become targets for criminals. Moreover, housing vacancy rates negatively impact assault, larceny, and vehicle theft crimes, contrary to previous research conclusions. For instance, Chen explored the longitudinal association between housing vacancy patterns and crime in New Orleans, finding a significant correlation between housing vacancy and the increase in neighborhood property and violent crimes (Chen & Rafail, 2020). Higher rental rates increase the likelihood of assault, robbery, larceny, and vehicle theft crimes. High rental rates indicate a frequent presence of non-local populations, which might weaken community cohesion and reduce informal social control, increasing crime opportunities according to the collective efficacy theory (Sampson, 2013).

This study presents two primary innovations. The first is the expansion of the indicators used to study the relationship between street environments and crime. By incorporating Marshall’s concept of street configurational types into the topological morphology analysis framework, this research includes indicators such as “X ratio,” “T ratio,” “cell ratio,” and “cul ratio.” These indicators not only characterize the basic geometric forms of street networks but also convey the configurational features of the street network, such as intersection types and their proportions, and the connectivity of the road network. These metrics broaden the analytical scope beyond the simple geometric features typically focused on in existing studies, such as street length and density. This allows for a more nuanced exploration of the relationship between street morphology and criminal activity. The second innovation lies in the extension of the existing crime pattern theory (Fig. 8). Crime pattern theory suggests that street paths and nodes influence people’s daily activities, which in turn indirectly affect the occurrence of crime. However, the theory does not delve deeply into how morphological patterns, such as path intersections and closures, specifically influence criminal behavior. By introducing street configurational and morphological features, this study further explains how street paths and activity nodes, through their structural characteristics (such as intersection types and accessibility), influence the selection of daily activity routes, thereby affecting the risk of crime. This extension enriches the theoretical framework and provides new perspectives and guidance for street design.

Fig. 8
figure 8

Expansion of crime pattern theory.

Full size image

However, this study has some limitations. Firstly, the types of crimes considered need to be expanded. Previous research indicates that the relationship between different crime types and street environment varies; this study focuses on violent crimes like assault and robbery and property crimes like larceny and vehicle theft. Future research should explore other crime types. Secondly, the time span of the research data can be extended. This study analyzes data from only one year, which cannot verify the impact of changes in environment and population factors on crime. Future studies could consider using panel data analysis. Thirdly, the study primarily examines the configurational form of street paths (intersection types) in Detroit, where intersection configurations are relatively simple. Future studies could explore areas with more diverse intersection configurations to assess how these configurational indicators impact crime. Moreover, since the study employs block groups as the unit of analysis, there exists inherent heterogeneity in street-level metrics within each block group. While the aggregated street network indicators can reflect the presence of certain key arterial streets within the area, they cannot fully represent all individual street segments. Therefore, the interpretation of area-level metrics requires distinct consideration from traditional street-level analyses. Fourthly, the time gap between crime data from 2019 and POI data from 2023 may cause measurement errors affecting the accuracy of associations between environmental features and crime. Although Detroit’s built environment is generally stable over short periods, this temporal difference is a limitation. Future studies should use longitudinal or contemporaneous data to better capture dynamic interactions between urban environments and crime patterns.

Conclusion

This study takes Detroit as an example, using crime data from 2019 on assault, robbery, theft, and vehicle theft to analyze their spatial distribution characteristics. The study incorporates street configurational types into a topological analysis framework, considering street network morphology (including geometric and topological forms), activity nodes, socioeconomic attributes, and constructs street environment indicators to explore the relationship between street environments and crime. The results show that the street network morphology can influence crime occurrences to some extent. For example, increasing road density helps reduce theft crimes, while areas with higher proportions of community and urban streets may facilitate the occurrence of violent crimes such as assault and robbery. Additionally, the impact of cell density on crime is moderated by the type of crime, and the influence of street permeability on crime varies depending on travel patterns and ranges. Different types of crimes are also associated with different node facilities. Regarding socioeconomic factors, after considering racial composition, travel patterns, educational levels, and housing occupancy characteristics, the study finds that neighborhoods with more individuals holding a bachelor’s degree or higher tend to have a lower likelihood of violent crimes but a higher likelihood of property crimes. Housing rental rates show a positive impact on all four types of crimes. Based on these findings, we offer practical design recommendations for urban planners and policymakers to improve crime prevention. Enhancing road network connectivity and density can increase pedestrian and vehicle activity, boosting natural surveillance and reducing theft opportunities. Special attention should be given to managing and designing community and urban streets by improving lighting, visibility, and encouraging community involvement to lower the risk of violent crimes. Additionally, optimizing the placement and management of activity nodes, such as dining facilities, bars, and bus stops, can help reduce their potential as crime attractors through increased security and regulation of alcohol-related venues. Different crime types have distinct environmental characteristics in their high-crime areas, which can provide local governments with reference points for crime prevention and control measures.