Assessing Spatial Associations Between Crime Exposure and Neighborhood Walkability: A Cross-Sectional Analysis of Socio-Environmental Moderators in Detroit

Abstract

Walkability is a multidimensional construct shaped by the built environment, social context, and perceived safety. Yet, most empirical studies treat walkability as spatially independent, overlooking the spatial and contextual factors that influence its relationship with neighborhood crime. This study investigates how crime affects walkability across Detroit, Michigan. Using data from 2021–2023, we developed a cross-sectional dataset of 624 census block groups. Comparing ordinary least squares (OLS), spatial lag (SLM), and spatial error (SEM) specifications, the SLM consistently provided the best fit, indicating strong spatial spillover in neighborhood walkability. Results show that higher local crime densities are positively associated with walkability, likely reflecting denser, mixed-use areas with greater pedestrian activity and exposure. Built-environment characteristics, particularly intersection density, land-use diversity, and population density, emerged as the most robust predictors of walkability, while socio-demographic factors showed weaker effects. Moderation analyses further reveal that the positive crime and walkability association diminishes in neighborhoods with higher proportions of Black residents, suggesting that structural inequities and historical segregation shape the realized benefits of walkable environments. These findings underscore the importance of accounting for spatial dependence and neighborhood context when assessing the complex interplay between safety, equity, and urban form.

1. Introduction

An urban neighborhood is a complex ecosystem in which environmental planning, social dynamics and public safety converge to determine residents’ quality of life [1]. Among the numerous key factors influencing the livability of a city, walkability, the extent to which an environment encourages safe, convenient and enjoyable pedestrian activities [2,3], has become an increasingly important factor in shaping public health and social cohesion [2,4], as well as promoting the economic development of communities [2,5]. Walkability is a multi-dimensional concept spanning the built environment, social relations and perceptual qualities that facilitate and nurture pedestrian movement [6,7].

Studies have demonstrated that enhancing the urban landscape positively influences urban walkability [4,5,8]. It includes all elements from the built environment, humanistic dimension of communities, road network planning and public facilities [4,9]. Walkability combines built environment parameters (connectivity, land-use mix, density) with its social counterparts, such as safety, accessibility, community relations and perceptual qualities, comfort level and attractiveness [10,11]. A high level of walkability has been found to be strongly linked to several urban merits, including enhanced public health by increasing physical activity [1,2,7,12], elevated social equality by providing superior access to places and opportunities [13], as well as increased urban resilience due to lower car use and environmental impacts [14,15,16].However, improvements in the built environment alone do not guarantee higher walkability; the presence of crime and perceived insecurity can override these benefits and fundamentally alter walking behavior.

Experimental studies have long shown that insecurity and crime decrease perceptions of security and discourage walking activity [17,18]. Prior literature suggests a bidirectional but asymmetric relationship between walkability and crime [19,20,21]. On one hand, crime may suppress walking by reducing perceived safety and discouraging outdoor activities [22,23,24,25]. On the other hand, highly walkable, mixed-use areas generate dense pedestrian flows and greater exposure opportunities, which can increase recorded crime incidents [26,27,28]. Several studies further argue that both crime and walkability are jointly shaped by underlying built-environment features such as land-use mix, street connectivity, and activity generators [29,30]. Given these intersecting pathways, crime can be treated as an environmental condition influencing the realized level of walkability rather than solely as an outcome of pedestrian activity. This theoretical framing supports our decision to model crime as the explanatory variable in the current study.

Factors influencing this disparity between crime and walkability include various potential moderating mechanisms [31,32]. Two overarching moderator categories are noteworthy: (1) Socioeconomic variables (SES): Greater resilience to crime in high-SES neighborhoods could manifest via enhanced community support networks, collective efficacy, and maintenance or surveillance capacity [33,34]. On the contrary, low-SES regions may suffer heightened crime effects owing to resource paucity and weakened societal bonds [33,35]. (2) Built Environmental factors: Built environment configurations including street patterning, land-usage heterogeneity, and accessibility can impact pedestrian exposure and susceptibility to crime [36,37]. Urban form attributes like visibility, active frontages, and mixed-use arrangements may either diminish or exacerbate subjective and objective hazards [36,38]. Given these mediating influences, it becomes obvious that neither the strength nor direction of the crime and walkability nexus holds universal validity across space. Rather, these effects are contingent upon local socioeconomic resilience and environmental arrangements that co-constructed how communities fare and respond to their urban contexts [33,36].

As one of America’s prominent post-industrial cities, Detroit has experienced decades of profound socioeconomic restructuring, racial segregation, and shifting crime dynamics [39,40]. Its neighborhoods exhibit sharp contrasts in population composition, economic conditions, and built-environment characteristics, resulting in concentrated crime corridors alongside highly uneven walkability and access to services [41]. Understanding how objective crime exposure intersects with these diverse neighborhood conditions is therefore essential for interpreting the spatial distribution of walkability across Detroit. Such analysis provides critical insight for planners and policymakers seeking to improve safety, equity, and mobility in a city marked by long-standing structural disparities.

Building on this context, and recognizing Detroit’s unique combination of pronounced racial segregation, spatially clustered crime patterns, and heterogeneous built-environment form, this study addresses the following site-specific research questions:

• What are the spatial patterns and correlations among walkability, crime exposure, and neighborhood socio-environmental characteristics across Detroit’s census block groups?
• How does crime intensity affect walkability after accounting for built-environment structure and socioeconomic conditions?
• How do demographic and environmental moderators condition the relationship between crime and walkability within Detroit’s historically segregated and unevenly developed neighborhoods?

This study contributes to the existing literature in three key ways. First, unlike previous research that often treats walkability as spatially independent, we explicitly incorporate spatial dependence using spatial lag models to reveal how walkability is shaped by neighborhood spillover effects. Second, while prior studies typically examine the impact of walkability on crime, this study reverses the analytical direction by investigating how objective crime densities influence walkability, providing a novel perspective for understanding the co-location of crime and pedestrian activity. Third, we introduce socio-environmental moderators to reveal structural inequities that condition the walkability–crime relationship, a dimension largely overlooked in previous investigations.

2. Methodology

2.1. Study Area and Unit of Analysis

The study is set in the City of Detroit, Michigan which is a post-industrial urban center with a population of about 640,000 in 2020 (Figure 1). Detroit’s demographic profile is characterized by a majority African American population and high levels of economic disadvantage. Decades of population decline and economic disinvestment have left Detroit with widespread vacant land and abandoned structures. The city spans a large geographic area with a low overall population density. These spatial characteristics are directly relevant to walkability: there are only a few pockets of relatively high residential density, and roughly half of Detroit’s neighborhoods have extremely low housing density. Such a sparse urban form limits the availability of amenities within walking distance and hinders the development of walkable neighborhoods. Importantly, Detroit’s walkability must be understood in the context of safety: the city has historically suffered from high crime rates, especially violent crime, which can deter pedestrian activity. Detroit’s crime rates rank among the highest in the nation for major violent offenses, although recent trends show improvement—total and violent crime rates have been declining since around 201. This backdrop makes Detroit an insightful case for studying how objective crime levels might impact the potential for walking in neighborhoods.

In this study, the primary unit of analysis is the census block group. Census block groups (CBGs) are small spatial units that approximate neighborhoods, each containing a few hundred to a few thousand residents. Detroit contains just over 600 block groups in total, allowing analysis at a fine-grained neighborhood scale.

2.2. Data Sources and Variables

2.2.1. Dependent Variable

Neighborhood walkability was measured using the Walk Score index [42], a widely adopted geospatial metric quantifying the ease of walking to nearby destinations. The index evaluates the density and accessibility of amenities and urban design characteristics. Using a distance-decay algorithm, destinations within a 5-min walk receive the highest weights, with scores decreasing up to a 30-min walk. Final scores range from 0 (car-dependent) to 100 (highly walkable). Walk Score has been validated as a reliable indicator of built environment walkability and is associated with physical activity and active travel [43,44]. In this study, Walk Scores for each census block group centroid in Detroit were retrieved via the official API (Walk Score API, https://www.walkscore.com/) and used as the dependent variable in spatial regression models.

2.2.2. Independent Variable

The primary explanatory variable is the objective crime intensity for each neighborhood, measured as the annual number of reported crime incidents per unit area. We adopt an area-based crime density rather than a population-normalized rate because Detroit contains many block groups with very low residential population but substantial commercial or daytime activity. In such contexts, population-based rates would understate the spatial concentration of crime opportunities, whereas area-based density better captures the geographic distribution of crime generators and attractors in mixed-use and high-activity corridors. This approach aligns with criminological theories emphasizing the spatial opportunity structure of crime.

We further calculated the total number of crimes and compared patterns across three categories: all crimes, violent offenses (e.g., homicide, assault, robbery), and property crimes (e.g., burglary, theft). This comparison allowed us to examine whether different types of crime show distinct associations with neighborhood walkability. All data were obtained from the City of Detroit’s Open Data Portal. Given the potential anomalies in crime reporting and mobility patterns during the COVID-19 pandemic in 2021, we use the average annual crime density from 2021 to 2023 to provide a more stable and representative estimate at the census block group level.

2.2.3. Control Variables

To isolate the net impact of crime on neighborhood walkability (

Table 1. Summary of Control Variables and Data Sources Used in the Analysis.

Category	Variable	Unit	Description (Concise)	Source
Built Environment & Land-Use Structure	Land-Use Diversity	Index (0–1)	Entropy-based index measuring the heterogeneity of residential, employment, and commercial land uses within a block group.	SLD
	Retail Employment Density	Jobs per acre	Captures the concentration of retail and service jobs such as shops and restaurants.
	Gross Population Density	People per acre	Reflects residential compactness and potential pedestrian flows.
Accessibility & Regional Structure	Distance to Nearest Transit Stop	Meters	Straight-line distance from the block-group centroid to the closest public transit stop.
	Trip Productions–Attractions Balance Index	Index (0–1)	Measures the balance between job and housing locations.
Socio-Demographic Characteristics	SES	Index (standardized)	PCA 1—derived composite index of income, education, and poverty rate.	ACS data (2023); computed via PCA 1
	Percent of Zero-Car Households	% of households	Share of households without private vehicles.	ACS data (2023)
	Occupied Housing Units	Count	Total number of occupied housing units within a block group.
	Percent Population Aged 65+	% of total population	Share of elderly residents.
	Percent Black Population	% of total population	Represents racial composition.

¹ SES index derived from 2023 ACS variables (income, education, poverty) using Principal Component Analysis (PCA).

), we incorporated a comprehensive set of control variables reflecting (1) built environment and land-use structure, (2) accessibility and regional structure, and (3) socio-demographic characteristics. Variables in categories (1) and (2) were obtained from the U.S. Environmental Protection Agency’s Smart Location Database (SLD, https://www.epa.gov/smartgrowth/smart-location-mapping), while category (3) variables were drawn from the 2023 American Community Survey (ACS).

2.3. Data Collection and Preprocessing

Data were collected from the sources and integrated through a geospatial and statistical preprocessing workflow. First, we obtained the census block group boundaries for Detroit. All location-based datasets, including walkability scores, ACS demographic indicators, and built environment measures from the SLD, are inherently aligned with standard census geographies and were therefore merged using the block group ID as a common key. Walkability data were drawn from the latest nationwide Walk Score dataset available at the block group level, from which we extracted the subset corresponding to Detroit and joined it with the GIS boundary layer for mapping and spatial analysis (processed in ArcGIS Pro v3.2; Esri Inc., Redlands, CA, USA). Similarly, ACS data were retrieved at the block group level via the U.S. Census API (https://api.census.gov/) and merged by block group code.

The SLD provides numerous built environment indicators; from these, we extracted a focused subset—covering density, land-use mix, street connectivity, and transit accessibility—for Detroit’s census block groups. The SLD variables are reported at the block group level and required minimal additional processing beyond normalization when appropriate.

We checked for missing values and outliers, and a small number of block groups with incomplete information were excluded or imputed where necessary. To avoid multicollinearity among explanatory variables, we further screened candidate predictors using the Variance Inflation Factor threshold of 10, retaining only variables below this cutoff. Finally, we compiled all processed variables into a cross-sectional dataset, in which each Detroit block group is represented by a single record used for descriptive and spatial regression analyses.

2.4. Analytical Strategy

We compiled summary statistics for all key variables which were standardized, and skewed rate variables were log-transformed to improve normality. To examine spatial patterns, we constructed a spatial weights matrix based on queen contiguity and applied both Global Moran’s I to assess spatial autocorrelation (computed in Python v3.10 using PySAL v2.6) and local clustering. Corresponding choropleth maps were generated for Walk Score, crime, SES, and land-use mix using consistent color gradients and classification schemes to ensure comparability. These descriptive and spatial analyses established the empirical foundation for subsequent regression modeling, guiding the decision to employ spatial econometric approaches and interaction tests in later stages.

To quantify the relationship between neighborhood crime and walkability, we estimated a cross-sectional regression model at the census block group level. The model incorporated both socioeconomic and built-environment characteristics to isolate the net relationship between crime and walkability:

$$Y_i= \alpha + {\beta }_1{Crime}_i+ {\delta }^{\prime }{S}_i+ {\epsilon }_i$$

(1)

where $Y_i$ denotes the Walk Score for block group; ${Crime}_i$ represents the total, violent, or property crime rate per unit area; and is a vector of socioeconomic controls including the SES index (PCA of income, education, and employment), the percentage of Black population, and the percentage of zero-car households.

Robust standard errors were used to account for heteroskedasticity. Separate models were estimated for total, violent, and property crime to identify differential relationships across crime types.

Because OLS assumes spatial independence, we next examined spatial autocorrelation in model residuals using Global Moran’s I based on a queen contiguity spatial weights matrix (W). Significant residual dependence indicated that spatial effects were present, violating OLS assumptions and suggesting the need for spatial econometric correction. To address this, we compared two common specifications: the SLM and the SEM, guided by Lagrange Multiplier (LM) and robust LM tests.

The spatial lag model incorporates spatial dependence directly in the dependent variable:

$$y= \rho Wy+X\beta +\epsilon , \epsilon ~ \mathrm{{\rm N}}(0, {\sigma }^2\mathrm{{\rm I}})$$

(2)

where $\rho$ is the spatial autoregressive coefficient measuring spillover effects across neighboring block groups.

Alternatively, the spatial error model captures unobserved spatial dependence through the disturbance term:

$$y= X\beta +u, u=\lambda Wu+ \mathrm{\xi },\mathrm{\xi }~\mathrm{{\rm N}}(0, {\sigma }^2\mathrm{{\rm I}})$$

(3)

where $\lambda$ reflects the spatial correlation in residuals. Both models were estimated via maximum likelihood using a row-standardized weight matrix. Model performance was compared based on AIC, log-likelihood, and Moran’s I of residuals to ensure spatial dependence was effectively removed.

To test whether the association between crime and walkability varies across different social and environmental contexts, we estimated a series of moderation models in which crime interacts with a single contextual moderator $x_i$. Each model takes the following general form:

$$Y_i= \alpha + {\beta }_1{Crime}_i+{\beta }_2{x}_i + {\delta }^{\prime }\left({Crime}_i \times x_i\right)+{\gamma }^{\prime }{Z}_i+ {\epsilon }_i$$

(4)

where $Y_i$ denotes the Walk Score for block group $i$; ${Crime}_i$ represents the total, violent, or property crime rate per 1000 residents; $x_i$ is the moderator variable of interest; and includes built-environment and accessibility controls. In separate estimations, $x_i$ is replaced by each of four moderators: SES, percent black population, percent zero-car households, and land-use diversity.

3. Results

3.1. Summary Statistics

Across Detroit block groups, the distribution of core built-environment and sociodemographic indicators is heterogeneous and often right-skewed (

Table 2. Descriptive statistics for explanatory and outcome variables by U.S. census block group (N = 624).

	Mean	Std	Skew	Kurtosis
Percent of Zero-Car Households	0.242137	0.162886	0.810571	0.817294
Population Density	10.3708	6.61204	1.45491	3.15758
Trip Productions and Attractions Equilibrium Index	0.256705	0.287369	0.719241	−0.793843
Land Diversity	0.179099	0.242619	1.85299	2.67096
Intersection Density	52.7298	40.575	0.689921	−0.157878
Distance to Nearest Transit Stop	183.994	4020.22	−24.9191	621.973
Occupied Housing Units	339.966	279.787	3.88788	22.2701
Percent of Population Aged 65+	0.152659	0.107401	1.35955	3.38316
Percent Black Population	0.739515	0.313972	−1.36559	0.500192
Medium Household Income	42953.6	21217.5	1.43269	3.12492
Number of Crime (2021–2023)	404.481	231.46	2.56382	19.0538
Edu Bachelor Plus Rate	0.169676	0.165834	1.95955	4.14806
Poverty Rate	0.320769	0.178857	0.374095	−0.537193
SES	33.2997	14.6746	0.991649	1.77277
Walkscore	49.4038	19.2569	−0.00969	−0.274117

). On average, 24.2% of households are zero-car, population density is 10.37 persons per acre, and intersection density is 52.7 per square mile; all three show positive skew and sizable dispersion, consistent with a small set of highly urbanized tracts amid many lower-density areas. Trip Productions & Attractions Equilibrium Index and land-use diversity are low on average (0.257 and 0.179, respectively) with long right tails, indicating that a majority of block groups exhibit limited activity balance and mixed uses, while a minority approach the upper range. The Walk Score centers near 49.4 (SD is approximately 19.3), indicating a moderate overall walkability. Median household income averages $42,954 (median: $38,963) and BA-plus attainment averages 17.0%, both of which are strongly right-skewed; poverty rates average 32.1%. The composite SES index—scaled 0–100 from the first principal component—is low on average (mean 33.3, SD 14.7), reflecting relatively disadvantaged conditions in many block groups.

Crime exposure is highly uneven (Figure 2). As shown in Table 2, the mean number of incidents per block group from 2021 to 2023 is 404.5, with a strong right skew (skew = 2.56) and very heavy tails, indicating that a small subset of areas accounts for a disproportionate share of incidents. Many variables, such as zero-car households, age 65+, and poverty, which are proportions in the range of 0 to 1, are right-skewed. In contrast, the Black population share is left-skewed (skew = −1.37), reflecting concentrations near the lower bound. Income and crime both exhibit long upper tails, and occupied housing units are likewise heavy-tailed. One variable warrants caution: the “distance to nearest transit stop” variable contained several extreme placeholder values originating from the raw SLD dataset. To retain the variable while preventing distortion in the regression models, we applied a winsorization procedure at the 1st and 99th percentiles. This correction removed the influence of erroneous extreme distances while preserving the overall distributional structure of the variable across block groups. Overall, the distributional moments align with expectations: the SES index behaves as the inverse of poverty and complements education and income; walkability is mid-range on average; and counts (crime, housing units) concentrate in a heavy-tailed minority of block groups.

Figure 2, Figure 3, Figure 4 and Figure 5 show a coherent core–corridor versus periphery pattern. Walkability (Figure 3) peaks along Detroit’s central spine, which runs from downtown/midtown and the major radial corridors, and then declines toward the predominantly residential periphery, with secondary high-scoring clusters along the east-side lakeshore and select northwest tracts. Crime surfaces share this geometry: the all-crime intensity (Figure 2) is highest in or adjacent to the same high-access belts, and the property (Figure 4) and violent (Figure 5) maps mirror those concentrations with modest variation in amplitude across space. In other words, hotspots persist in roughly the same corridors where access and mixed urban activity are most prevalent. This co-location should not be interpreted as causal, since high-access areas also tend to concentrate trips, retail, nightlife, and transient populations, which increase exposure opportunities and reporting. However, it does imply that subsequent models need to explicitly adjust for activity intensity when interpreting any association between walkability and crime.

3.2. Linear Relationship Between Walkability and Crime

Before introducing spatial models, we first tested whether the residuals from the OLS regression exhibited spatial dependence (Figure 6). The Global Moran’s I statistic for the OLS residuals was 0.358 (z = 1.04, p = 0.001), indicating significant positive spatial autocorrelation. Figure 6 presents the Moran scatterplot of the standardized residuals, which reveals a clear positive relationship between each block group’s residual value (x-axis) and the spatially lagged residuals of its neighbors (y-axis). The upward-sloping trend line confirms that nearby block groups tend to share similar levels of unexplained variation in walkability, violating the independence assumption of OLS regression. Consequently, spatial regression models were employed to explicitly account for this spatial dependence.

To assess the spatial dependence in neighborhood walkability models, we first estimated OLS regressions for overall, violent, and property crime rates. Moran’s I statistics of the OLS residuals indicated significant positive spatial autocorrelation across all three models (I = 0.38–0.40, p < 0.01), suggesting that spatially adjacent block groups shared similar levels of unexplained variation. To account for this spatial dependence, both SLM and SEM models were subsequently estimated.

Model fit statistics consistently showed that the spatial lag model provided the best performance. For all-crime, violent-crime, and property-crime models, the SLM achieved substantially lower AIC values (4983.6–4991.0) and higher pseudo-R² values (≈0.61) compared with the OLS (AIC ≈ 5277–5303, pseudo-R² ≈ 0.25–0.28) and SEM models. Moreover, the residual Moran’s I of the SLM models was close to zero and statistically insignificant (p > 0.2), indicating that spatial dependence was effectively removed. These results confirm that incorporating spatial lag dependence significantly improved model fit and reduced residual spatial autocorrelation, supporting the use of the SLM specification for subsequent interpretation.

Table 3. Comparison of OLS and spatial regression models for all, violent, and property crime rates in relation to neighborhood walkability.

Model	AIC	Log-Likelihood	Pseudo R2	Residual Moran’s I	p-Value
Model 1—All crimes
OLS	5277.23	−2626.61	0.284	0.377	0.001
SLM	4983.62	−2478.81	0.6104	−0.0182	0.234
SEM	5012.47	−2494.24	0.2512	0.5374	0.001
Model 2—Violent crimes
OLS	5302.69	−2639.34	0.254	0.4021	0.001
SLM	4990.99	−2482.50	0.6096	−0.0170	0.239
SEM	5015.85	−2495.92	0.2138	0.5468	0.001
Model 3—Property crimes
OLS	5302.69	−2639.34	0.254	0.4021	0.001
SLM	4990.99	−2482.50	0.6096	−0.0170	0.239
SEM	5015.85	−2495.92	0.2138	0.5468	0.001

Lower AIC and higher pseudo-R² indicate better model fit. Residual Moran’s I tests evaluate remaining spatial autocorrelation (values close to zero and insignificant p-values suggest adequate spatial adjustment). Boldface highlights the best-fitting models.

presents the SLM estimates examining the relationship between neighborhood walkability and different types of crime. Across all three models, the spatially lagged walkability term was positive and highly significant (p < 0.001), confirming strong spatial spillover effects, which means walkability in one block group tends to be similar to that of its neighbors.

Figure 7 illustrates the spatial distribution of model residuals across Detroit’s census block groups, comparing the OLS and the SLM. In the OLS results (left panel), residuals display distinct spatial clustering, with positive residuals concentrated in central and northwestern neighborhoods and negative residuals dominating the southern areas. This spatial pattern confirms the significant positive Moran’s I observed in the OLS residuals, implying that nearby block groups tend to share similar levels of unexplained variation in walkability. In contrast, the SLM residuals (right panel) appear more evenly distributed, showing less spatial concentration and weaker regional trends. This reduction in clustering demonstrates that incorporating a spatial lag term effectively mitigates spatial dependence, leading to a more robust and spatially consistent model fit.

Among the explanatory variables (

Table 4. Spatial lag regression results for all, violent, and property crime models.

	Coef	Std Err	p
All Crimes
Crime Rate	2.0552	0.4928	0 **
Distance to Nearest Transit Stop	0.00005	0.00012	0.6733
Intersection Density	0.0343	0.013	0.0082 **
Land Diversity	5.6457	2.6146	0.0308 *
Trip Productions and Attractions Equilibrium Index	3.6387	2.1684	0.0933 †
SES	0.0159	0.0359	0.6585
Percent of Zero-Car Households	−1.0985	3.177	0.7295
Occupied Housing Units	0.0054	0.0024	0.0237 *
Age 65+	−0.0015	0.0054	0.7735
Black Population	−0.0034	0.0012	0.0064 **
Population Density	0.2992	0.0991	0.0025 **
Spatial lag (W_walkscore)	0.7265	0.0329	<0.001 **
Violent Crimes
Crime Rate	4.2246	1.361	0.0019 **
Distance to Nearest Transit Stop	0.00005	0.00012	0.6864
Intersection Density	0.0366	0.013	0.0048 **
Land Diversity	5.8787	2.6195	0.0248 *
Trip Productions and Attractions Equilibrium Index	3.606	2.1745	0.0973 †
SES	0.0194	0.0363	0.5937
Percent of Zero-Car Households	−0.9139	3.1899	0.7745
Occupied Housing Units	0.0045	0.0024	0.0643 †
Age 65+	−0.0011	0.0054	0.833
Black Population	−0.0028	0.0012	0.0236 *
Population Density	0.3422	0.0986	0.0005 **
Spatial lag (W_walkscore)	0.7408	0.0323	<0.001 **
Property Crimes
Crime Rate	3.8655	0.8976	0 **
Distance to Nearest Transit Stop	0.00006	0.00012	0.6426
Intersection Density	0.0345	0.013	0.0079 **
Land Diversity	5.2643	2.6207	0.0446 *
Trip Productions and Attractions Equilibrium Index	3.6539	2.169	0.0921 †
SES	0.007	0.0358	0.8458
Percent of Zero-Car Households	−1.3288	3.1823	0.6763
Occupied Housing Units	0.0052	0.0024	0.027 *
Age 65+	−0.0021	0.0054	0.695
Black Population	−0.0035	0.0012	0.0051 **
Population Density	0.2873	0.0999	0.004 **
Spatial lag (W_walkscore)	0.7216	0.0332	<0.001 **

** p < 0.01, * 0.01 ≤ p < 0.05, † 0.05 ≤ p < 0.1.

), crime density showed a consistently significant and positive association with walkability (coef = 2.06 for all crimes, 4.22 for violent crimes, 3.87 for property crimes; p < 0.01). This suggests that higher local crime rates are observed in areas that are also more walkable, potentially reflecting greater pedestrian activity or mixed-use intensity rather than causality.

Indicators of the built environment, particularly intersection density and land-use diversity, were both positive and statistically significant across all models (p < 0.05), indicating that areas with finer street connectivity and more diverse land uses tend to have higher walkability scores. Population density also exerted a strong positive influence (p < 0.01), reinforcing the role of compact urban form in promoting walkability.

In contrast, socio-demographic factors such as the proportion of elderly residents, zero-car households, and SES were not statistically significant, suggesting that physical and spatial structure rather than population composition plays a more dominant role in shaping walkability at the block-group level. However, the percentage of Black residents was negatively associated with walkability (p < 0.05), pointing to potential spatial inequities in pedestrian infrastructure or neighborhood conditions.

Overall, the SLM results reveal that built environment characteristics and spatial dependence are the strongest determinants of neighborhood walkability in Detroit, while demographic variables show weaker or inconsistent effects.

3.3. Moderating Effects on the Walkability-Crime Relationship

To examine whether socio-demographic and environmental contexts condition the impact of crime on neighborhood walkability, a series of spatial lag models were estimated incorporating interaction terms between crime and four moderators: SES, percentage of Black residents, percentage of zero-car households, and land-use diversity. This approach allows us to evaluate whether crime clusters differently in walkable areas depending on neighborhood structure and resilience. Importantly, the moderation analysis also provides an opportunity to revisit long-standing theoretical claims.

Across all models, the spatial autoregressive coefficient (ρ) remained large and highly significant (ρ ≈ 0.73–0.76, p < 1 × 10⁻¹¹⁰), confirming strong spatial dependence in walkability across block groups (

Table 5. Moderation effects of socio-demographic and environmental variables in spatial lag models of walkability and crime.

Model	AIC	Pseudo-R2	ρ (Spatial Lag)	p (ρ)	Crime Main β (p)	Moderator Main β (p)
All Crime × SES	5003.882	0.6023	0.755 (p = 4.28 × 10−128)	1.301 (0.003)	0.031 (0.394)	0.039 (0.198)
All Crime × Black Population	4987.413	0.6078	0.740 (p = 5.36 × 10−119)	2.339 (<0.001)	−0.003 (0.001)	−0.001 (0.012)
All Crime × Zero-Car	5002.571	0.6023	0.753 (p = 2.98 × 10−126)	1.147 (0.008)	1.411 (0.650)	3.844 (0.077)
All Crime × Land Diversity	4989.897	0.6061	0.739 (p = 3.07 × 10−117)	1.218 (0.004)	8.677 (<0.001)	0.587 (0.684)
Violent Crime × SES	5009.68	0.601	0.763 (p = 7.44 × 10−134)	2.34 (0.063)	0.02 (0.571)	0.02 (0.825)
Violent Crime × Black Population	4996.7	0.607	0.756 (p = 2.23 × 10−130)	4.91 (0.00045)	−0.0029 (0.007)	−0.0035 (0.774)
Violent Crime × Zero-Car	5007.27	0.602	0.762 (p = 2.04 × 10−133)	1.91 (0.111)	1.43 (0.646)	9.87 (0.106)
Violent Crime × Land Diversity	4993.64	0.606	0.747 (p = 5.18 × 10−123)	2.31 (0.050)	8.93 (<0.001)	3.04 (0.482)
Property Crime × SES	5002.075	0.6025	0.752 (p = 9.64 × 10−126)	2.412 (0.0021)	0.022 (0.537)	0.081 (0.110)
Property Crime × Black Population	4984.676	0.6075	0.732 (p = 1.13 × 10−113)	4.443 (<0.001)	−0.0036 (0.0010)	−0.0027 (0.0094)
Property Crime × Zero-Car	5001.501	0.6016	0.748 (p = 1.55 × 10−122)	2.181 (0.0054)	1.274 (0.682)	6.713 (0.075)
Property Crime × Land Diversity	4989.84	0.6055	0.736 (p = 2.13 × 10−115)	2.247 (0.0041)	8.360 (0.00010)	0.297 (0.900)

). Our findings complicate these assumptions by showing that urban form alone does not universally mitigate the co-location of crime and walkability; instead, its effectiveness varies substantially across different social contexts.

For all-crime models, the baseline effect of crime on walkability was positive and statistically significant in most cases (β = 1.15–2.34, p < 0.01), indicating that higher reported crime levels tended to coincide with higher walkability, likely reflecting denser, more active urban areas where both pedestrian activity and crime incidents are concentrated. Among moderators, the percentage of Black residents significantly moderated this relationship (interaction β = −0.001, p = 0.012), suggesting that the positive association between crime and walkability weakens in predominantly Black neighborhoods. The moderation by SES, zero-car households, or land-use diversity was not statistically significant.

In the violent crime models, similar patterns emerged: the spatial lag term remained significant, and the direct effect of violent crime was positive but weaker (β = 1.9–4.9, p < 0.10). Again, racial composition exerted a consistent moderating influence (interaction β = −0.0035, p = 0.007), indicating that neighborhoods with higher proportions of Black residents, the spatial co-location of walkability and violent crime is substantially attenuated. This pattern challenges conventional theoretical standpoints suggesting that higher-density, mixed-use environments inherently produce safer pedestrian settings through natural surveillance or continuous street activity. Instead, our findings suggest that built-environment advantages, such as density, connectivity, and mixed uses, may not translate into perceived or realized safety in structurally marginalized communities. For practitioners, this underscores the importance of pairing physical design interventions with investments in social infrastructure, trust-building, community-led safety programs, and anti-disinvestment efforts. Without addressing these structural inequities, urban form alone is insufficient to ensure that walkability delivers its intended safety benefits across all neighborhoods.

For property crime, the results were the most robust and consistent with the overall pattern. Property crime exhibited significant positive main effects on walkability (β = 2.18–4.44, p < 0.01). The moderation by the share of Black residents was again significant (interaction β = −0.0027, p = 0.009), while SES and zero-car household rates showed no meaningful moderating influence. Land-use diversity remained a strong positive main predictor of walkability (β = 8.36, p < 0.001), but its interaction with crime was not significant.

Overall, these results suggest that while higher crime rates—particularly property crime—tend to occur in more walkable neighborhoods, this association is spatially structured and context-dependent. The weakening of the crime–walkability link in neighborhoods with higher percentages of Black residents points to the influence of social and structural inequalities on perceived and realized walkability.

Figure 8 presents the spatially adjusted total effects of overall crime on neighborhood walkability, conditioned by key socioeconomic and built-environment moderators. Because the moderation patterns for violent and property crime models closely resembled those of the overall crime model, only the total-crime results are illustrated here for parsimony. The figure displays both direct (dashed) and total (solid) spatial effects under low (−1.5 SD) and high (+1.5 SD) levels of each moderator: SES, percent black population, percent zero-car households, and land-use diversity.

Across all four panels, consistent moderation patterns emerge. Higher SES and greater land-use diversity systematically strengthen the positive association between walkability and crime. In neighborhoods characterized by stronger socioeconomic resources or more diverse urban forms, walkability remains relatively high even under elevated crime conditions. These findings suggest that both social capital and environmental vitality act as buffers mitigating the adverse effects of crime on pedestrian activity. In contrast, neighborhoods with lower SES, more homogeneous land use, or higher shares of zero-car households exhibit weaker or even negative relationships between crime and walkability, reflecting greater vulnerability to perceived or actual safety concerns.

Overall, the spatial lag model demonstrates that socioeconomic and built-environment contexts meaningfully condition how crime influences neighborhood walkability. Socially advantaged and physically diverse neighborhoods tend to sustain walkability despite higher crime levels, whereas mobility-limited or socioeconomically disadvantaged communities experience sharper declines. These results underscore the importance of integrated planning strategies that enhance both social resilience (through improved socioeconomic conditions and community safety) and environmental diversity (through mixed-use and pedestrian-oriented urban design) to foster walkable, resilient neighborhoods in cities facing persistent safety challenges.

To assess the stability of our findings, we conducted two robustness checks. First, we re-estimated the spatial lag models using a 6-nearest neighbors spatial-weights matrix rather than queen contiguity. The results remained consistent in direction, magnitude, and statistical significance. Second, we applied a log-transformation to the crime density variable; the transformed models produced fully comparable estimates and did not alter the substantive conclusions. These robustness checks support the reliability of the main findings.

4. Discussion

4.1. Crime and Walkability, How and Where Crime Impacts Accessibility Most Significantly

This study reveals a complex relationship between crime and walkability across Detroit neighborhoods, marked by significant spatial dependence and contextual nuances. The positive association between crime density and walkability observed in our spatial-lag models requires careful interpretation. Rather than implying that walkability is directly linked with crime, the findings reflect a spatial co-location pattern shaped by Detroit’s underlying urban form. As shown in Figure 2, Figure 3, Figure 4 and Figure 5, both walkability and crime are concentrated along Detroit’s central corridors, which are characterized by dense street networks, mixed land uses, and high levels of pedestrian movement. The spatial lag coefficient (ρ ≈ 0.72–0.74) further indicates that walkability is strongly influenced by conditions in adjacent neighborhoods, meaning that high-accessibility areas form contiguous clusters where pedestrian flows and crime opportunities naturally intensify simultaneously.

This co-location is consistent with established theories of crime generators and attractors [45,46], which posit that areas with intense economic and social activity affect both walkability and crime [28,47,48]. High-density, mixed-use environments create natural surveillance and active street life but also higher exposure opportunities, elevating recorded crime independent of actual safety deterioration [49]. The significant effects of intersection density, land-use diversity, and population density in our models (Table 4) underscore that these built-environment features—rather than crime—are the primary drivers of walkability and are simultaneously correlated with crime concentration. Thus, the positive crime–walkability association in Detroit reflects the clustering of high-activity environments rather than a behavioral or causal influence of crime on walking.

Crime does not cause walkability; rather, walkable generators attract both foot traffic and offenders [50,51,52]. Therefore, these findings demonstrate that the crime–walkability relationship is not causal but emergent from the spatial structure of Detroit’s urban form. Crime clusters where walkability is already highest, while walkability is most vulnerable to crime in low-density neighborhoods lacking the built-environment supports that promote safe and active street life.

4.2. The Role of Social and Environmental Moderators, Community Heterogeneity in Resilience or Vulnerability

An important contribution of this study lies in identifying social and environmental factors that moderate the crime and walkability relationship. Interaction models show that racial composition (not SES or car-ownership) significantly attenuates the crime and walkability link. In communities with a higher proportion of black residents, the positive correlation between crime and walkability has weakened. This indicates that historical racial segregation, structural inequality, and insufficient investment have undermined the benefits of walkable urban forms for these communities. This spatial inequality highlights the persistent racial disparities in accessing equal opportunities in safe and pleasant pedestrian environments.

Environmental moderating factors such as land use diversity and intersection density have a significant direct impact on the convenience of walking [31,53,54], but do not significantly affect the influence of crime on walking convenience. This finding emphasizes the dominant role of architectural environmental characteristics in shaping the pattern of walking convenience and also indicates that social vulnerability and historical background have complex and multi-level influences on the actual experience of pedestrians. The ability of a community to deter crime through social cohesion, collective efficacy, and local maintenance and informal monitoring can be enhanced, and these capabilities often show significant differences in diverse communities in Detroit [55]. These results are in line with broader evidence, which suggests that the strength and direction of the link between crime and walking convenience are not universally valid; instead, it depends on the local socioeconomic and environmental background that shapes the way communities respond to urban challenges.

To further strengthen the practical relevance of these findings, planning interventions should prioritize neighborhoods where social vulnerability amplifies the negative effects of crime on walkability. In areas with high proportions of Black residents or limited land-use diversity, targeted investments such as improved lighting, enhanced pedestrian infrastructure, and community-led safety programs can help ensure that built-environment improvements translate into meaningful accessibility gains.

4.3. Planning and Policy Implications for Enhancing Walkability Under Crime Exposure

The findings of this study offer actionable planning strategies tailored to the spatial imbalance relationship between crime and walkability across Detroit. Because crime and walkability co-locate primarily in high-density, mixed-use corridors yet impacts on low-density neighborhoods, policy interventions must be spatially differentiated and sensitive to the built-environment and socio-demographic context.

In Detroit’s central and radial corridors, where crime and walkability co-exist, the priority is to enhance the sense of security that people perceive and experience, while not undermining the vitality of the city. Measures such as CPTED-informed streetscape enhancements, improved lighting, transparent ground-floor facades, and safety-focused transit stop upgrades can strengthen natural surveillance while sustaining pedestrian activity [56]. These interventions preserve the benefits of high-density, mixed-use corridors by maintaining active street life while reducing crime.

Peripheral areas exhibit the greatest sensitivity to crime, where even moderate crime levels discourage walking due to sparse destinations and weak connectivity. In these areas, environmental restructuring is essential. Targeted infill development, improved sidewalk continuity, protected walking routes, and the reactivation of vacant lots can help reduce isolation and amplify the basic conditions necessary for walkability. Such changes can allow residents to reclaim outdoor spaces that currently feel unsafe or underutilized.

Moderation results indicate that racially marginalized neighborhoods experience weakened benefits of walkability under crime exposure. For these areas, community-led safety programs and participatory design approaches can be considered to help ensure proposed interventions, such as upgraded lighting, pedestrian facilities, and maintained public spaces, align with residents’ safety perceptions and mobility needs, thereby translating physical improvements into meaningful accessibility gains.

Finally, the strong spatial lag effects identified in this study highlight the need for planners to integrate crime data into walkability assessments. Spatial overlays that combine crime exposure with built-environment indicators can better identify zones where safety concerns disproportionately suppress mobility. Coordinating environmental interventions with community-based crime prevention can help cities address walkability and safety holistically rather than as separate concerns.

4.4. Limitations and Future Directions

Several limitations should be acknowledged. Firstly, the cross-sectional design fundamentally constrains causal inference and raises significant concerns. The observed positive association between crime density and walkability may suffer from multiple forms of bias. Reverse causality is plausible, that walkable neighborhoods with dense pedestrian activity and mixed land uses may generate more crime opportunities (as crime generators) rather than crime reducing walkability. Simultaneity, walkability and crime may mutually influence each other through unobserved feedback loops. Furthermore, despite comprehensive control, residual variable bias may still persist, factors such as nighttime economic activities, informal social control, or enforcement intensity that are not measured may simultaneously affect both crime reporting and walking convenience. Therefore, our findings should be interpreted as identifying spatial correlations rather than causal effects. Future research can employ longitudinal panel designs to establish temporal ordering, or quasi-experimental approaches such as difference-in-differences analysis of walkability interventions (e.g., Complete Streets policies) or instrumental variable strategies leveraging exogenous policy shocks to disentangle these complex relationships. Secondly, Crime data are susceptible to systematic reporting biases, especially across neighborhoods with varying levels of trust in police or civic institutions, and may not fully reflect residents’ perceived safety [57,58]. Prior research also shows that subjective fear of crime and objective crime rates often diverge in meaningful ways that shape walking behavior [59]. Integrating qualitative assessments of fear and safety perception will help deepen understanding. Thirdly, although multiple moderating factors have been tested, other aspects such as social cohesion, law enforcement practices, and informal social control have not been directly measured, which limits the understanding of community recovery mechanisms.

Future research should adopt a longitudinal and mixed-method approach to reveal causal relationship paths and deeper environmental influencing factors. At the same time, studying the interactions among multiple social and environmental moderating factors may help clarify the complex community dynamics. Expanding the research to other post-industrial cities or diverse urban environments will help enhance the general applicability of the research results. Finally, policy-oriented research is needed to translate these insights into strategies for achieving a fair walking environment which ensuring that all communities can benefit from a safe, convenient and vibrant pedestrian environment, despite the presence of crime issues.

5. Conclusions

This study advances understanding of how crime and socio-environmental factors jointly shape neighborhood walkability in a post-industrial context. Using a spatially explicit framework applied to Detroit’s 624 census block groups, we demonstrate that walkability and crime exhibit significant spatial clustering and are strongly interrelated. Contrary to conventional assumptions, higher objective crime densities are positively associated with Walk Score, reflecting the concentration of both pedestrian activity and crime opportunities in denser, mixed-use, and socially vibrant areas rather than a causal effect of crime itself.

SLM reveal that built-environment features, particularly intersection density, land-use diversity, and population density, are the most consistent determinants of walkability, underscoring the role of compact, well-connected urban form in sustaining pedestrian accessibility. Socio-demographic variables show weaker direct effects, yet racial composition significantly moderates the relationship between crime and walkability: in predominantly Black neighborhoods, the positive association between crime and walkability weakens, indicating that structural inequities constrain the realized benefits of otherwise walkable environments.

These findings emphasize that the interplay between safety, equity, and urban form is inherently spatial and context dependent. Planning interventions should therefore pursue integrated strategies that strengthen both physical design and social resilience—enhancing connectivity, land-use mix, and surveillance while addressing systemic disparities in neighborhood safety and investment. Future research incorporating longitudinal data and perceived-safety measures could further disentangle causal mechanisms and guide equitable, evidence-based urban design policies for safer and more walkable cities.

Despite its contributions, this study has several limitations. First, the cross-sectional design prevents causal inference and only reflects the co-location of crime and walkability at one point in time. Second, crime data capture reported incidents rather than residents’ perceived safety, which may shape walking behavior independently of objective crime. Third, although all 624 block groups were retained, several control variables required imputation due to minor missingness, and more nuanced measures of social cohesion or informal surveillance were unavailable. Finally, Detroit’s unique demographic and post-industrial context may limit generalizability to other urban settings.