Exploring the Spatial Relationship Between Crime and Urban Places in Austin: A Geographically Weighted Regression Approach

Abstract

Urban safety is a critical concern for sustainable city development, with crime patterns often linked to localized environmental factors. Understanding the spatial dynamics of safety is critical for informed design and planning of urban environments. This study employs a Geographically Weighted Regression (GWR) approach to investigate how crime in Austin, Texas, correlates with Points of Interest (POIs) such as bars, transit stations, financial businesses, and public spaces, while accounting for localized socio-economic factors. Building on theoretical frameworks like Routine Activity Theory and Crime Pattern Theory, the analysis integrates crime data from the Austin Police Department (APD), POI datasets, and census variables to explore spatially varying relationships often overlooked by traditional global models (e.g., OLS). A novel adaptive geo-grid method refines spatial units by clustering high-density downtown areas into smaller zones and retaining larger grids in suburban regions, ensuring precision without over-fragmentation. Analysis of crime incidents and POI data reveals significant spatial non-stationarity in crime–environment associations. Transportation-related facilities demonstrate strong spatial correlation with crime citywide, particularly forming persistent crime hotspots around transit hubs in areas like Rundberg Lane, South Congress, and East Riverside. Alcohol-related establishments show a strong positive correlation with crime in entertainment districts (coefficient up to 13.5, p < 0.001) but a negligible association in suburban residential areas (coefficient close to 0, p > 0.05). The GWR model significantly outperforms traditional OLS regression, capturing critical local variations obscured by global models. Downtown Austin emerges as a complex hotspot for urban safety where multiple high-risk POI types overlap. This research advances urban design and planning knowledge by providing empirical evidence that environmental factors’ influence on safety is spatially conditional rather than universally consistent, aligning with Crime Pattern Theory and Routine Activity Theory. The findings support place-specific crime prevention strategies, offering policymakers data-driven insights for developing targeted design strategies for urban zones.

1. Introduction

Safety is one of the top issues in American cities, having a significant effect on social stability, economic growth, and public safety. According to information from the Texas Department of Public Safety, statewide homicides declined 10.7% between the years 2022 and 2023 [1]. Looking back between 2019 and 2023, violent crime increased 27.6%, which portends an overarching upward trend. Aside from the direct impact that crime has on human lives, crime also presents economic impacts as it lowers the value of properties and increases public expenditures on law enforcement. Additionally, marginalized populations are disproportionately vulnerable to crime, increasing social injustices. Therefore, understanding the spatial distribution of crime in cities is important to develop effective, place-specific prevention policies for future urban design and planning practices [2]. The rates of crime in Austin, Texas, vary widely across different urban geographies as safety concerns are significantly higher in some areas compared to others, making it a great case study for spatially focused analysis [1].

The rates of crime in Austin, Texas, vary widely across different urban geographies as safety concerns are significantly higher in some areas compared to others, making it a strong case study for spatially focused analysis [1]. These disparities cannot be explained solely by physical environmental features. They also reflect the interplay of broader social, economic, environmental, and technological stressors that cities face simultaneously—a phenomenon described as the polycrisis [3]. In this context, similar neighborhoods with comparable physical layouts or POI compositions can nonetheless experience markedly different crime rates due to variations in socioeconomic inequality, governance capacity, and community resilience. Understanding crime in an era of polycrisis, therefore, requires analytical tools that capture both systemic influences and localized conditions. Geographically Weighted Regression (GWR) meets this need by identifying place-specific crime–environment relationships, enabling precision policymaking where interventions address not only the design of physical spaces but also the management, services, and social supports that shape their use.

There are various urban theories to explain why crime clusters in particular places. Routine Activity Theory (RAT) [4] teaches that crime occurs whenever three circumstances coincide: (1) a motivated offender, (2) an appealing victim or proper targets, which means persons or objects worth the effort that are visible, within reach, and accessible, and (3) the absence of effective protection (e.g., police, surveillance cameras, or other watching eyes) [5]. Not merely based on offender motivation alone, crime exists wherever environmental arrangements support an offending opportunity [6,7]. Crime Pattern Theory is an expansion of RAT that addresses how crime is aggregated geographically within cities [8]. Crime Pattern Theory holds that offenders target places that are familiar to them—such as their home, workplace, or where they travel by habit—creating “crime hotspots” where proper targets are plentiful and good guardianship is poor.

Broken Windows Theory also finishes these models by pointing out the role of disorder in the urban environment in promoting crime [9,10]. Graffiti, property destruction, and trash are signals of social abandonment, which encourage crime and may breed more serious offenses. Some of the crime prevention programs in inner cities incorporate parts of this theory, including public area design and maintenance on a regular basis, and participation by the community [11]. Empirical evidence supports these theories. Research indicates that crime is concentrated in specific areas and specific times, validating the Crime Pattern Theory’s assumptions about the spatial concentration of crime. The Broken Windows Theory has also influenced urban police practice, most notably in New York City, during the 1990s under Police Commissioner William Bratton, where aggressive policing practices were associated with a startling decline in crime rates [7].

A growing body of research has investigated the relationship between urban spatial configurations and crime, highlighting the role of specific land uses and place types in influencing crime patterns. Numerous studies have identified that establishments such as fast-food restaurants, convenience stores, liquor stores, and large-box retailers (e.g., Walmart) are positively associated with higher rates of violent crime [12,13,14,15,16]. In contrast, the presence of green spaces has frequently been linked to reductions in crime, with parks and urban greenery proposed as mechanisms for fostering social cohesion and informal surveillance [17,18,19].

However, findings across the literature are far from consistent. While some studies assert that green space can deter crime by promoting legitimate usage and territorial functioning, others caution that poorly maintained or secluded green areas may provide opportunities for hidden criminal activity [11,20,21,22,23]. These contradictions underscore the need for context-specific, spatially explicit research that can account for local variations in how places influence crime.

Moreover, the current literature reveals important conceptual and methodological gaps. Many studies are constrained by large-level geographic units—focusing on a single city or neighborhood—limiting the generalizability of their findings. The resolution of crime data is rarely high enough to conduct place-based analysis in urban contexts, with most research relying on aggregated data at the census tract or block group level. Traditional approaches aggregate diverse microenvironments within broader geographic units, obscuring nuanced relationships between specific crime types and particular places. This limitation creates a significant analytical blind spot. For example, previous studies may identify that commercial areas experience higher crime rates but cannot distinguish whether robberies cluster around restaurants vs. grocery stores, or how these patterns vary by time of day. Recent studies utilizing high-resolution mobility data (e.g., SafeGraph) demonstrate that transient populations near transit hubs exhibit distinct temporal crime patterns, necessitating dynamic modeling approaches [24]. Additionally, the cumulative effects of multiple urban elements—spanning the physical, social, and institutional environment—are seldom explored in a unified analytical framework. This study utilizes unique high-granular crime records for understanding the heterogeneity of urban spaces. We also employ GWR to examine the interplay between land use, green space, accessibility, and social context and their combined effects on crime distribution. Through this approach, the research contributes to a more nuanced and integrative understanding of urban crime dynamics, grounded in both theory and spatial evidence.

Therefore, this study aims to (1) quantify spatially varying relationships between crime and POIs in Austin using GWR; (2) identify localized high-risk POI clusters; and (3) provide place-specific design strategies for urban safety.

2. Data and Methods

2.1. Data Sources

2.1.1. Study Area: City of Austin, TX

Austin, the capital city of Texas, is located in the south-central region of Texas along the Colorado River. It has experienced rapid population growth, increasing from 790,390 in 2010 to approximately 986,928 in 2024 [25]. The Austin-Round Rock metropolitan area has experienced rapid growth, with a 10.8% population increase from 2020 to 2024 [26]. This expansion is accompanied by strong economic expansion in technology, education, and creative industries. Despite its economic prosperity, the city has experienced socioeconomic disparities and public safety concerns. In 2022, Texas reported a Gini coefficient of 0.476, highlighting significant income inequality [27]. According to Blau (1985), rapid population growth combined with economic disparities can intensify social tensions, contributing to elevated crime rates in areas with pronounced inequality. Research on the spatial relationship between crime and urban environments in Austin offers theoretical insights and empirical data to guide future urban planning efforts [28].

2.1.2. Gathering and Analyzing Mobility Data

This study utilized anonymized mobility data from SafeGraph and Advan, two companies recognized for collecting location-based data from mobile devices. These data providers are adopted in academic research due to their established data reliability and methodological transparency.

SafeGraph, in particular, employs a variety of accuracy-focused strategies to ensure high data quality. These include accuracy metrics, such as the Coverage Rate and the Real Open Rate, which assess the quality and completeness of U.S. Points of Interest (POI) data [29]. To further refine data accuracy, SafeGraph uses multiple truth sets, including government datasets and industry benchmarks like Google, to evaluate and calibrate its data [30]. Additionally, precision assessments such as the Real Open Rate are applied through manual verification of POI samples to determine their legitimacy and current operational status [31].

This dataset provides detailed information on location-specific features such as latitude, longitude, location name, subcategory, and top category. Crime-related POIs were further categorized into six groups based on the Routine Activity Theory and Crime Pattern Theory [8]: alcohol-related venues, financial businesses, transportation locations, public spaces, high-risk services, and retail hubs. Each group is associated with crime types, including violent offenses, property theft, and public disturbances. A total of 57,963 sampled records were analyzed.

The number of visitors from a visitor’s block group (BG) to a POI was adjusted to account for weekly variations in device sampling ratios over the study period, as outlined in Equation (1). To ensure accurate identification of true POI visitors, data points were clustered based on the duration a device remained in a specific area and then compared to known POI boundaries. Points falling within these boundaries were classified as valid visits, while those indicating irregular or erratic movement were excluded.

$$N_{b,p,w}={\displaystyle \frac{R_{b,p,w}\times Po{p}_b}{D_{b,w}}}|$$

(1)

In this study, Nbpw represents the estimated number of visitors from block group b to point of interest p during week w. The term Rbpw refers to the raw number of visits from block group b to POI p in week w, as recorded by SafeGraph. Popb indicates the total population of block group b, which is used to scale the visit data. Dbw denotes the number of sampled devices in block group b during week w, serving as an adjustment factor to correct for sampling bias in the mobility dataset.

$$V{P}_{kt}=\sum _{i=1}^n VB{G}_{ikt}$$

(2)

where VPkt denotes the overall number of visits to POI k during week t, obtained by summing VBGikt, the estimated visits from each contributing block group i. The value n represents the total number of block groups with recorded visits to the POI in that specific week.

2.1.3. Crime Data (Austin Police Department)

For types of crime (violent vs. property), this study utilizes crime incident data provided by the Austin Police Department (APD) [32], which systematically categorizes incidents based on crime type following standards outlined by national crime-reporting systems, including the Uniform Crime Reporting (UCR) program and the National Incident-Based Reporting System (NIBRS). This data provides specific coordinates for crime occurrence location information, which is unique compared to previous studies. APD classifies crimes into two primary categories used in this research. The distinction between violent and property crimes facilitates a nuanced spatial and temporal analysis, crucial for targeted urban planning and crime-prevention strategies.

Violent Crimes: These involve force or the threat of force against persons, including murder and non-negligent manslaughter, aggravated assault, robbery, and sexual assault [33]. Violent crimes significantly disrupt the social fabric of communities by undermining residents’ perceived safety and contributing to broader social distress [34].

Property Crimes: These offenses primarily concern illegal acquisition or destruction of property without direct physical harm to individuals. Some examples include burglary, larceny-theft, motor vehicle theft, and arson [31]. Property crimes indirectly affect community cohesion, economic stability, urban livability, and neighborhood attractiveness [35].

In our crime data, they exhibit distinct daily and seasonal patterns. Violent crimes often peak at night due to reduced visibility and diminished surveillance, whereas property crimes typically increase during daytime when residential occupancy is lower [36,37]. Seasonally, crime rates rise around holidays and summer months, driven by heightened residential vacancy, tourism, and social gatherings, creating conditions conducive to both property and violent offenses [38,39]. While both violent and property crimes exhibit clear daily and seasonal fluctuations, the primary focus of this study is on spatial variation in crime–environment relationships. For this reason, crime incidents were aggregated over the full multi-year study period before spatial modeling, rather than modeled as a function of time. Complementary spatiotemporal approaches—including deep learning with remotely sensed imagery—have shown promise for forecasting crime rhythms at fine scales [40]. This approach controls short-term temporal fluctuations while capturing stable spatial associations.

2.1.4. POI Data

POI has long been widely used to study urban characteristics and human activity patterns, as an important indicator to understand the dynamics of urban space [41]. This study uses POI data to analyze Austin crime patterns.

The selection of POI categories was guided by two criteria: (1) theoretical relevance established in environmental criminology literature, particularly Routine Activity Theory and Crime Pattern Theory, which identify certain land uses as potential crime generators or attractors, and (2) empirical evidence from prior spatial crime studies linking these place types to elevated crime risk [3,4,5]. Categories such as alcohol-related establishments, transportation facilities, and high-risk service locations have repeatedly been shown to concentrate on motivated offenders, suitable targets, and limited guardianship, while public/green spaces and retail areas can function as either protective or criminogenic settings depending on context [42]. These categories typically fall into four main groups, as shown in

Table 1. POI Categories.

Category	Examples of Locations
Alcohol-Related Establishments	Bars, Pubs, and Taverns Nightclubs Beer, Wine, and Liquor Stores
Financial and High-Cash Businesses	Banks Convenience Stores Gas Stations
Transportation	Urban Transit Systems Bus Stations Train Stations Parking Lots and Parking Structures
Public/Green Spaces	Parks and Open Spaces with Low Surveillance
High-Risk Service Locations	Strip Clubs and Adult Entertainment Homeless Shelters and Public Housing Facilities Casinos and Betting Centers Tattoo Parlors and Smoke Shops Used Car Dealers and Auto Repair Shops
Large Crowds and Retail Areas	Shopping Malls Grocery Stores Fast Food Restaurants

. To test the robustness of these categorizations, we also conducted supplementary sensitivity checks by re-running the GWR model with alternative POI groupings (e.g., merging retail and high-risk service locations, excluding rare POI types) and found no substantial changes in the spatial patterns or significance of key predictors. This supports the stability of our results with respect to POI categorization.

2.2. Spatial Regression Method: Geographically Weighted Regression (GWR)

To examine the spatially varying relationship between urban environmental features and crime distribution across Austin, this study employs GWR as the core modeling approach. GWR is designed to capture spatial heterogeneity by estimating localized regression coefficients at each spatial unit [24,43,44,45,46]. This is particularly critical for urban crime studies, where neighborhood-level dynamics, land use patterns, and policing efficacy vary substantially across space and influence the magnitude and direction of crime-generating processes. The selection of GWR is thus both methodological and theoretical: it reflects the recognition that urban phenomena such as crime are embedded in local spatial contexts and that their relationships with environmental attributes are not uniform across a heterogeneous urban landscape.

While alternative local spatial modeling techniques, such as Multiscale Geographically Weighted Regression (MGWR) [26], can account for variable-specific spatial scales, our research focus does not require multi-scalar estimation. The aim of this research is to evaluate the localized strength and direction of crime–environment relationships at a consistent neighborhood scale, enabling direct comparison of predictor effects across space. GWR’s single adaptive bandwidth structure is well-suited for this objective, allowing us to capture fine-grained local variations without overfitting, which is a potential risk when MGWR, or other advanced techniques, are applied to datasets with a limited number of observations per unit. Given the relatively high granularity of our block group–level data and the study’s emphasis on neighborhood-scale heterogeneity, GWR offers a methodologically efficient and theoretically coherent framework. This choice also aligns with Routine Activity Theory and Crime Pattern Theory, which emphasize that environmental influences on crime are inherently place-based and context-dependent.

While GWR effectively captures spatial heterogeneity, recent advances in machine learning offer complementary approaches. Advanced techniques such as Gradient Boosting Decision Trees (GBDTs) have revealed nonlinear relationships between environmental factors and crime, often achieving higher predictive accuracy than traditional linear models [47]. However, GWR remains preferable for interpretable spatial analysis where understanding local coefficient variation is paramount.

The analysis was conducted at the scale of the U.S. Census block group, which offers an appropriate balance between spatial granularity and data reliability for neighborhood-level analysis. Block groups provide sufficient resolution to capture localized variations in crime patterns and urban form while maintaining consistency with demographic and land-use data. Each block group served as an individual spatial unit in the model, with the dependent variable specified as the count of reported crimes within its boundaries.

The set of explanatory variables used in the GWR model includes a comprehensive suite of environmental, socio-spatial, and accessibility-related indicators, reflecting diverse theoretical perspectives on crime generation. These include the number of various identified types of POIs (transportation-related, public and green space, large crowd and retail-related, high-risk service locations, financial and high-cash businesses, and alcohol-related establishments), average distance to the nearest police station for each block group, and total population of each block group (

Table 2. POI Categories.

	Min	Mean	Max	Std
Number of Crime	0	187.48	6120	316.89
Number of Transportation POI (Urban Transit Systems; Bus Stations; Train Stations; Parking Lots and Parking Structures)	0	5.38	129	8.13
Number of Public and Green Spaces POI	0	0.89	14	1.47
Number of Large Crowds and Retail Areas POI (Shopping Malls; Grocery Stores; Fast Food Restaurants)	0	5.21	148	10.5
Number of High-Risk Service Locations POI (Strip Clubs and Adult Entertainment; Homeless Shelters and Public Housing Facilities; Casinos and Betting Centers; Tattoo Parlors and Smoke Shops; Used Car Dealers and Auto Repair Shops)	0	1.48	47	3.52
Number of Financial and High Cash Businesses POI (Banks; Convenience Stores; Gas Stations)	0	3.12	104	7.12
Number of Alcohol Related Establishments POI (Bars, Pubs, and Taverns; Nightclubs; Beer, Wine, and Liquor Stores)	0	8.43	544	26.73
Average distance to police station of the block group	0.16	2.27	9.98	1.57
Total population of the block group	0	2051.62	11,726	1551.25

). The average distance to the nearest police station was calculated to represent spatial accessibility to formal guardianship; specifically, this was operationalized as the average of the shortest distances from each POI within a block group to its nearest police station, thereby capturing the localized accessibility of law enforcement to active places within each neighborhood unit. The total population of each block group was included as a control for exposure and scale.

Prior to conducting GWR, a global OLS model is used to assess overall model fit and test for multicollinearity among explanatory variables.

Crimei = β0 + β1(Transportationi) + β2(PublicGreeni) + β3(Retaili) +
β4(HighRiski) + β5(Financiali) + β6(Alcoholi) + β7(PoliceDisti) +
β8(Populationi) + εi

(3)

In Equation (3), Crimei is the number of crimes in block group i. β0 is the intercept, and β1 to β8 are the global coefficients for each explanatory variable.

The Koenker (BP) test will confirm the presence of spatially non-stationary relationships, justifying the application of GWR. After that, a Poisson-based GWR model was employed in recognition of the count-based nature of the dependent variable (crime incidents).

Crimei = β0(ui,vi) + β1(ui,vi)(Transportationi) + β2(ui,vi)(PublicGreeni) + ⋯
+β8(ui,vi)(Populationi) + εi

(4)

In Equation (4), (ui,vi) are the spatial coordinates (e.g., the centroid) of block group i. k = 1, …, 8 index the covariates, and βk (ui,vi) is the local coefficient for the kth variable at location i. The rest of the terms are the same as in the OLS model, but with location-specific estimates.

The model was estimated using an adaptive neighborhood type—defined by a consistent number of nearest neighbors across space—to account for the uneven spatial distribution of block groups in Austin. To determine the optimal neighborhood size, the Golden Search method was used in conjunction with Akaike’s Information Criterion corrected (AICc) to minimize model complexity while improving local fit. The analysis was conducted using ArcGIS Pro, which produced localized coefficient estimates and local pseudo-R² values, enabling a detailed spatial examination of how the magnitude and direction of relationships between crime and environmental features vary across the city’s diverse neighborhoods.

3. Results and Findings

3.1. Crime Distribution in Austin

Figure 1 and Figure 2 present the spatial distribution of crime and crime-related POIs in Austin. This hotspot analysis identifies crime-concentrated areas and explores correlations with different urban functions.

The research findings reveal a clear spatial clustering pattern of crime in Austin, with six primary types of Points of Interest (POIs) showing varying degrees of association with crime hotspots. These six categories include alcohol-related establishments, financial and high-cash businesses, public and green spaces, high-risk service locations, large crowds and retail areas, and transportation-related facilities. Each type of POI plays a different role in shaping the urban crime landscape, with significant spatial variation in its influence.

Among them, downtown Austin stands out as a zone where multiple POI types overlap. In areas such as the Sixth Street Entertainment District, alcohol-related establishments are highly associated with violent crimes and disorderly conduct. Financial and cash-intensive locations—such as banks, ATMs, and gas stations—also exhibit notable clustering in the downtown area. Likewise, transportation hubs, including bus stops and parking facilities, show a high concentration of crime in this part of the city. Public and green spaces generally report lower crime rates, with moderate clustering observed in certain locations.

From a broader citywide perspective, transportation-related POIs demonstrate the strongest spatial correlation with crime. Consistent and dense crime hotspots are found around transit facilities, not only in downtown Austin but also in areas like Rundberg Lane, South Congress, and East Riverside. In contrast, crime related to retail and alcohol-serving venues tends to be more localized and context-dependent.

High-risk service locations—such as strip clubs, public housing, and homeless shelters—also show a strong positive correlation with crime, particularly in East Austin and the North Lamar area.

Overall, while all six POI types influence the distribution of crime, transportation-related facilities and high-risk service locations display the highest degree of clustering and the broadest spatial extent. Due to the overlapping presence of multiple high-risk POIs, downtown Austin becomes a microcosm of the city’s broader crime dynamics.

3.2. Crime and POI Relationships (Local Effects)

3.2.1. OLS Results and Interpretation

To explore the global relationships between crime incidence and various POI categories, we applied an OLS regression model, the results of which are shown in

Table 3. OLS results—model variables.

Variable	Coefficient	StdError	t-Statistic	p-Value	Robust_SE	Robust_t	Robust_Pr	VIF
Intercept	17.867	12.776	1.398	0.163	16.528	1.081	0.280	--------
Transportation	9.626	1.212	7.944	0.000 *	2.338	4.117	0.000 *	3.169
Public/green Spaces	3.909	4.408	0.887	0.376	5.042	0.775	0.438	1.375
Large Crowds and Retail Areas	−3.804	1.214	−3.134	0.002 *	2.956	−1.287	0.199	5.301
High-Risk Service Locations	15.189	2.371	6.407	0.000 *	2.879	5.276	0.000 *	2.278
Financial and High-Cash Businesses	4.008	1.582	2.533	0.012 *	3.094	1.296	0.196	4.138
Alcohol-Related Establishments	6.758	0.500	13.526	0.000 *	1.277	5.294	0.000 *	5.823
Average distance to police station of the block group	−8.976	3.967	−2.263	0.024 *	3.319	−2.704	0.007 *	1.270
Total population of the block group	0.031	0.004	7.760	0.000 *	0.007	4.342	0.000 *	1.217

* p < 0.001.

. The regression revealed that several POIs, including high-risk service locations (β = 15.19, p < 0.001), alcohol-related establishments (β = 6.76, p < 0.001), and transportation hubs (β = 9.63, p < 0.001), are significantly and positively associated with crime rates.

In contrast, public and green spaces (β = 3.91, p = 0.376) and large crowds/retail areas (β = −3.80, p = 0.199) were not significantly associated with crime. The negative, albeit insignificant, coefficient for retail areas may suggest a potential deterrent effect of higher surveillance or more formal design standards in commercial zones.

Importantly, proximity to law enforcement was inversely related to crime rates, with the “average distance to police station” variable showing a statistically significant negative coefficient (β = −8.98, p = 0.024). This supports the idea that accessible policing contributes to crime deterrence, although the magnitude is not strong enough to suggest it is the sole influencing factor. Additionally, population size positively correlates with crime (β = 0.031, p < 0.001), supporting established findings that denser neighborhoods are more exposed to criminal activity.

The OLS diagnostics (

Table 4. OLS results—diagnostics.

		Dependent Variable	Crime
Number of Observations	530	Akaike’s Information Criterion (AICc) [‘d’]	6652.570
Multiple R-Squared [‘d’]	0.841	Adjusted R-Squared [‘d’]	0.839
Joint F-Statistic [‘e’]	344.781	Prob (>F), (8521) degrees of freedom	0.000 *
Joint Wald Statistic [‘e’]	389.600	Prob (>chi-squared), (8) degrees of freedom	0.000 *
Koenker (BP) Statistic [‘f’]	172.658	Prob (>chi-squared), (8) degrees of freedom	0.000 *
Jarque-Bera Statistic [‘g’]	1064.884	Prob (>chi-squared), (2) degrees of freedom	0.000 *

* p < 0.001.

) result showcases the model’s adjusted R² of 0.839, which demonstrates that roughly 84% of the variance in crime is explained by the included predictors, indicating a high level of explanatory power. However, several diagnostic tests point to limitations in the model’s global assumptions. The Koenker (BP) test is highly significant (χ² = 172.66, p < 0.001), indicating heteroskedasticity—in other words, the strength of predictor relationships varies across space. This violates the constant variance assumption of OLS and suggests that spatial non-stationarity is present.

Further, the Jarque–Bera statistic (χ² = 1064.88, p < 0.001) signals that residuals are not normally distributed, which violates another fundamental OLS assumption. This indicates potential non-linear relationships or omitted spatial variables that the global model does not capture.

Multicollinearity diagnostics also point to a moderate concern. For example, alcohol-related establishments had a Variance Inflation Factor (VIF) of 5.82, slightly above the common threshold of 5, suggesting that correlated predictors may affect the stability and interpretability of coefficient estimates.

Taken together, these diagnostic issues imply that while OLS offers initial insight into the relationships between crime and urban features, it cannot adequately represent local variations. Thus, we adopt GWR in the next section to account for spatial heterogeneity in predictor effects and improve model fidelity.

3.2.2. GWR Results and Spatial Heterogeneity

To address the spatial non-stationarity and nonlinear effects identified in OLS diagnostics, a GWR model was implemented to estimate localized regression coefficients across block groups (

Table 5. OLS vs. GWR coefficient comparison.

Variable	OLS Coeff	Min (GWR)	Q1 (GWR)	Median (GWR)	Q3 (GWR)	Max (GWR)
Transportation	9.626	0	1	4	7	129
Public/green space	3.909	0	0	0	1	14
Large Crowds and Retail Areas	−3.803	0	0	2	6	148
High-Risk Service Locations	15.189	0	0	0	1	47
Financial and High-Cash Businesses	4.008	0	0	1	4	104
Alcohol-Related Establishments	6.758	0	0	3	9	544
Average distance to police station of the block group	−8.976	0.162	1.196	1.961	2.814	9.979
Total population of the block group	0.030532	0	1141.5	1617.5	2418	11,726

), enabling spatial interpretation of how different POI types relate to crime in varying urban contexts (Figure 3).

The GWR results reveal that crime relationships with POIs are highly context-dependent and spatially heterogeneous (Figure 4). In central Austin—particularly the Sixth Street Entertainment District, South Congress, and East Riverside—coefficients for alcohol-related establishments were markedly high, exceeding 500 in some block groups. These locations, characterized by dense nightlife activity, showed the strongest positive correlations between alcohol-related businesses and crime. In contrast, suburban or low-density residential areas exhibited considerably lower coefficients, with some approaching neutrality, reflecting variation in nightlife intensity, population activity, and surveillance.

Transportation-related POIs, including bus and train stations and parking structures, were also strongly associated with crime in high-ridership areas such as downtown and the Rundberg Lane corridor. The spatial clustering of positive coefficients suggests these areas function as mobility hubs where increased transient movement may heighten the risk of opportunistic crime.

High-risk service locations (e.g., adult entertainment venues, public shelters, auto repair shops) and financial or cash-intensive businesses (e.g., gas stations, convenience stores, banks) displayed elevated coefficients in older commercial corridors and transitional neighborhoods. These associations were especially pronounced in areas with weaker infrastructure, mixed land uses, and lower formal guardianship, pointing to compounded spatial and social vulnerabilities.

Retail-oriented POIs exhibited varied coefficients across space. In well-regulated environments such as The Domain—an upscale shopping area—coefficients were low or negative, suggesting a protective or neutral relationship with crime. However, in peripheral commercial strips or low-surveillance areas, retail POIs were positively associated with crimes such as theft and vehicle break-ins.

Green and public spaces also displayed mixed spatial effects. Parks in well-maintained, high-use areas tended to show neutral or negative coefficients, while those in underserved neighborhoods showed positive correlations with crime, often in conjunction with poor visibility, lighting, or informal encampments.

4. Discussion

4.1. Spatial Heterogeneity Is Pervasive

The spatial variation observed in the GWR model underscores a fundamental insight: the relationship between urban features and crime is not uniform but is shaped by specific local conditions. The polycrisis framework suggests that crime disparities between neighborhoods with similar POIs (e.g., downtown vs. suburban bars) may reflect broader systemic stressors, such as income inequality, housing insecurity, or policing disparities [5]. For instance, Rundberg Lane’s high crime rates could stem from overlapping crises (e.g., poverty + transit density), while affluent areas mitigate risks through private security. Future models should incorporate socioeconomic resilience indicators to disentangle these synergies. While prior models like OLS aggregate these relationships into a single city-wide estimate, GWR reveals meaningful variation across neighborhoods, offering a richer understanding of crime-environment dynamics.

In entertainment and nightlife districts, the positive association between alcohol-related establishments and crime reflects established previous findings but also reaffirms the role of localized clustering in amplifying risk. These findings suggest that it is not the mere presence of alcohol establishments that drives crime but rather their spatial concentration in high-activity districts, often accompanied by weak nighttime guardianship and dense social mixing. Transportation nodes and corridors similarly emerge as zones of heightened vulnerability, not necessarily due to the POIs themselves, but because of the social and spatial conditions they generate. High foot traffic, temporal crowding, and anonymity likely converge to create settings conducive to certain types of crime. In such contexts, transit design, surveillance infrastructure, and policing presence become critical mitigating factors. Future urban designers need to be aware of such factors and offer relevant spatial solutions.

The differentiated effect of retail POIs is particularly notable. GWR reveals that the same category of POI—retail spaces—can have opposite effects depending on location. Where management, design, and visibility are strong, retail centers may deter crime through natural surveillance and legitimate presence. In contrast, neglected or isolated retail environments may lack these deterrents, becoming crime attractors instead. This complexity is obscured in global models but becomes clear through GWR’s spatial sensitivity.

Findings on green space further challenge simplistic associations between parks and safety. Recent work shows that design and programming mediate safety outcomes in parks, with well-activated, visible green spaces associated with lower risk [48]. Rather than serving uniformly as crime deterrents, green spaces exhibit criminogenic or protective effects depending on physical design, maintenance, visibility, and neighborhood context. These findings align with studies that caution against viewing green space as inherently beneficial, highlighting the role of planning and programming in shaping its social function.

Crucially, the value of GWR in this study lies not simply in outperforming OLS, but in its capacity to expose the conditionality of crime risk. The model does not suggest that certain POIs are inherently criminogenic but rather that their risk effects are spatially contingent, shaped by land use mix, surveillance, mobility, and socioeconomic context.

4.2. Planning and Design Implications

The findings of this study underscore the importance of adopting spatially differentiated and context-sensitive approaches to urban crime prevention, particularly in cities with diverse land use patterns and social dynamics like Austin. The spatial heterogeneity revealed through GWR challenges the effectiveness of blanket policies and calls for tailored urban design, regulatory, and community-based strategies.

In entertainment districts and transit-rich corridors—such as the Sixth Street Entertainment District, South Congress, and areas surrounding major transportation hubs—where crime risk is amplified by nightlife intensity, transient populations, and concentrated activity nodes, targeted interventions are crucial. These may include increased nighttime policing presence, enhanced public lighting, and application of Crime Prevention Through Environmental Design (CPTED) principles such as sight lines, access control, territorial reinforcement, and active frontages. An updated CPTED toolkit increasingly integrates smart technologies for monitoring and activation, broadening the menu of design-oriented prevention measures [49]. Moreover, public realm management, including street cleaning, crowd control, and formal programming, can enhance perceived safety and signal informal guardianship.

In older commercial corridors and transitional neighborhoods where high-risk service and financial POIs cluster—such as along parts of the Rundberg Lane corridor—urban design must work with social policy. Satellite-derived built-environment indices (e.g., building density) have been shown to differentially influence indoor vs. street crimes, highlighting the need for context-specific design interventions [50]. Investments in streetscape improvements, mixed-use development, and rehabilitation of underutilized properties should be coupled with expanded access to housing, mental health services, and job training [51,52]. These strategies address not only the environmental symptoms of crime but also the socioeconomic vulnerabilities that give rise to it.

In the case of public and green spaces, the findings challenge the assumption that provision alone is sufficient. Parks that are underfunded, poorly designed, or neglected may become sites of disorder rather than safety. Effective strategies should focus on integrating green spaces into the broader urban fabric, ensuring visibility from streets and residences, installing consistent lighting, and promoting regular, inclusive programming—such as community events, sports leagues, and educational activities—to foster natural surveillance and strengthen community ownership. Rather than treating parks as isolated amenities, planners should recognize them as active nodes within the urban network, whose safety depends on surrounding land use, connectivity, and engagement.

Notably, alcohol-related crime hotspots (e.g., Sixth Street) often align with nightlife tourism zones, where business models prioritize revenue over safety [53,54]. Municipalities should balance economic vitality with risk mitigation by: (1) requiring safety certifications (e.g., lighting standards) for liquor licenses; (2) designating ‘entertainment districts’ with coordinated policing and transport; and (3) engaging stakeholders (e.g., bar owners, residents) in co-designing nighttime governance strategies.

Ultimately, this study reinforces the necessity for hyperlocal policy responses, informed by fine-grained spatial analysis. City planning departments, law enforcement agencies, and community organizations must collaborate not only to reduce crime incidence but also to build environments that support inclusive, resilient urban life. This includes integrating crime data into planning workflows, ensuring equitable investment in vulnerable neighborhoods, and prioritizing maintenance and social infrastructure in addition to physical design [54].

Implementation challenges arise when translating place-specific recommendations into practice, which entail non-trivial constraints. Capital upgrades (lighting, cameras, frontage improvements) require multi-year budgeting and sustained O&M funding. Interagency coordination is essential—police, transit operators, public works, and nightlife/economic development offices often pursue misaligned incentives. Many CPTED actions occur on private parcels (e.g., parking lots, building frontages), which necessitate zoning incentives, façade programs, or business improvement districts. Politically, cities must balance nighttime economy goals against noise and safety concerns. Equity safeguards are needed to avoid displacement when safety investments raise amenity values. Finally, data-informed targeting depends on governance, privacy, and community trust in monitoring technologies.

Table 6. POI categories, examples, and practical implications for crime prevention and urban design.

Category	Examples of Locations	Practical Implications
Alcohol-Related Establishments	Bars, Pubs, and Taverns Nightclubs Beer, Wine, and Liquor Stores	In entertainment districts, a strong positive association with violent crime and disorder; targeted nighttime policing, improved lighting, CPTED measures, and crowd management are critical.
Financial and High-Cash Businesses	Banks Convenience Stores Gas Stations	Often located in older commercial corridors with elevated robbery/theft risk, they require surveillance, secure design, and integration with social policy measures such as community economic support.
Transportation	Urban Transit Systems Bus Stations Train Stations Parking Lots and Parking Structures	Consistent crime hotspots in transit-rich corridors; interventions include enhanced lighting, active station staffing, CPTED principles, and design to reduce anonymity.
Public/Green Spaces	Parks and Open Spaces with Low Surveillance	Crime risk varies by visibility, maintenance, and programming; effective strategies involve lighting, sight lines, regular inclusive programming, and integration with surrounding land uses.
High-Risk Service Locations	Strip Clubs and Adult Entertainment Homeless Shelters and Public Housing Facilities Casinos and Betting Centers Tattoo Parlors and Smoke Shops Used Car Dealers and Auto Repair Shops	Positive associations in transitional neighborhoods with weaker infrastructure require environmental design improvements and coordinated social services.
Large Crowds and Retail Areas	Shopping Malls Grocery Stores Fast Food Restaurants	Mixed effects depending on location; well-managed sites deter crime via natural surveillance, while neglected or isolated retail strips may attract crime—design quality and management are key.

summarizes the main POI categories analyzed in this study, along with examples of locations and their practical implications for urban design and crime prevention. These implications are informed by the spatially heterogeneous associations revealed in the GWR analysis and highlight potential strategies for place-specific interventions.

4.3. Limitations and Future Work

While this study contributes to a more nuanced understanding of the spatial relationships between crime and urban features, several limitations still need to be acknowledged to contextualize its scope and inform future research directions.

First, GWR effectively identifies spatial non-stationarity and local associations between POIs and crime, but it does not establish causality. For instance, our model shows that bars correlate with crime in entertainment districts, yet this association may result from contextual factors, such as poor lighting or low surveillance, rather than alcohol sales per se. The observed relationships should therefore be interpreted as spatial correlations, potentially shaped by broader conditions such as land use regulations, socioeconomic inequality, or policing strategies. To better isolate causal mechanisms, future studies could integrate field audits (e.g., CPTED assessments) to evaluate specific environmental attributes [55] that mediate POI–crime dynamics.

In addition, data constraints limit our ability to incorporate certain social dimensions and temporal patterns alongside POI measures. POI datasets, while valuable for identifying formal activity nodes, do not capture all environmental or social variables that may influence crime, such as informal gathering spaces, temporary or seasonal activities, lighting quality, building vacancy rates, or neighborhood-level social cohesion. These unmeasured factors may also shape local crime dynamics and should be considered when interpreting results. Incorporating qualitative methods, historical datasets, or governance-related variables could help unpack these layered influences. Furthermore, future research would benefit from integrating temporally granular data sources—such as GPS-based mobility traces, ride-hailing records, or geotagged social media activity—to align spatial analysis with daily and seasonal crime rhythms [56,57].

Second, the analysis is spatially bound to the city of Austin, a rapidly growing urban area with distinctive morphology, governance frameworks, and socioeconomic conditions. While the methodological approach is widely applicable, the findings are context-specific and may not directly translate to cities with different densities, development trajectories, or social infrastructures. Comparative studies across multiple urban contexts would be valuable for assessing the transferability of the observed patterns. Also, this study emphasizes physical and institutional factors but does not directly model social cohesion, community trust, or informal surveillance—elements that are often critical in shaping neighborhood safety. Future research could explore how these social dimensions intersect with spatial variables through community surveys, ethnographic fieldwork, or indicators of civic engagement.

Although our model centers on physical and institutional features, a substantial body of work shows that neighborhood social cohesion, informal guardianship, and perceived police effectiveness shape both the level of crime and its spatial susceptibility to environmental risks. In practice, the same POI density can carry very different implications in communities with strong social ties vs. places marked by distrust or weak collective efficacy. We did not directly model these social dimensions due to data availability and scale alignment; however, future work can integrate survey-based cohesion indices, civic engagement proxies, or co-produced community audits to test how social context moderates local coefficients. This would help distinguish place features that are risky everywhere from features that become risky only under adverse social conditions [34,35].

In the face of resource constraints, hyper-local interventions are most impactful when paired with a transparent prioritization rule. We recommend a three-step framework: (1) rank block groups by the product of local GWR coefficients and incident counts to identify high-leverage POI–crime dyads; (2) match each dyad to lowest-cost, first-best actions (e.g., sightline clearing, luminance standards, curb-use management at transit hubs) before capital-intensive works; and (3) stage delivery via short pilots with pre-specified metrics (calls-for-service, disorder proxies) and adaptive reallocation each quarter. This approach concentrates scarce funds where marginal returns are highest, protects baseline maintenance citywide, and creates an evidence loop for scaling successful hyper-local models without committing to costly, one-size-fits-all programs.

5. Conclusions

This study on crime patterns in Austin reveals several key insights that not only deepen the understanding of urban crime dynamics but also provide methodological implications for empirical research. First, the spatial distribution of crime in Austin shows significant heterogeneity, with crime hotspots associated with specific types of POIs displaying clear spatial clustering patterns. Downtown Austin, as an overlapping area of multiple high-risk POI types, forms a complex crime ecosystem, which is consistent with previous discussions in crime geography that emphasize the concentration of criminal activities at micro-spatial scales.

Transportation-related facilities throughout the city show significant spatial correlation with crime, particularly forming persistent crime hotspots around transportation hubs in areas such as Rundberg Lane, South Congress, and East Riverside. This finding is highly consistent with previous research on transportation nodes as crime generators, revealing how transportation facilities create opportunity structures for crime through increased flow of people, reduced surveillance density, and increased anonymity [58].

This study’s GWR model reveals profound spatial heterogeneity in how different POI types are associated with crime across different neighborhoods. Alcohol-related establishments show a strong positive correlation with crime in entertainment areas (such as the Sixth Street Entertainment District) (a coefficient of up to 13.5, p < 0.001), while in suburban residential areas, this association is almost negligible (a coefficient close to 0, p > 0.05). This significant spatial non-stationarity is empirically supported by Groff and Lockwood (2014), who found in their studies of Philadelphia that the crime impact of alcohol establishments is highly dependent on the surrounding land use mix, neighborhood design, and socioeconomic conditions [59].

First, theoretically, GWR highly aligns with crime pattern theory and routine activity theory, which emphasize the critical role of local conditions in shaping crime opportunities. The influence of environmental factors on crime is essentially spatially conditional, rather than universally consistent [60,61].

Second, methodologically, local models produce more accurate predictions. Our GWR model improved prediction accuracy compared to OLS, which is consistent with previous research that found that models accounting for spatial heterogeneity perform better in crime hotspot prediction [62,63,64].

Third, in terms of policy application, local models provide more actionable insights. By precisely identifying where and how environmental features influence crime, GWR provides an empirical foundation for developing targeted interventions, directly responding to previous criticisms of environmental criminology for lacking spatial specificity [65].

In summary, our research supports place-specific crime prevention strategies. In conclusion, recognizing and addressing the spatial heterogeneity of crime-environment relationships is key for cities to develop more precise, equitable, and effective safety strategies.