|
|
@@ -0,0 +1,694 @@
|
|
|
+---
|
|
|
+category: literaturenote
|
|
|
+citekey: shinfactorsassociateddifferenttypes2024
|
|
|
+title: "Factors associated with different types of freight crashes: A macro-level analysis"
|
|
|
+authors: "Shin, Eun Jin"
|
|
|
+year: 2024
|
|
|
+date: 2024-00-00 2024
|
|
|
+doi: 10.1016/j.jsr.2023.11.012
|
|
|
+publication: Journal of Safety Research
|
|
|
+url: "https://doi.org/10.1016/j.jsr.2023.11.012"
|
|
|
+zotero_key: RAGVIFDB
|
|
|
+zotero_storage: BA7TDWQA
|
|
|
+collections: transpordi_planeerimine
|
|
|
+folder: 001_artiklid/Road Safety Indicators
|
|
|
+firstAuthor: "Shin, Eun Jin"
|
|
|
+status: converted
|
|
|
+---
|
|
|
+
|
|
|
+
|
|
|
+Contents lists available at [ScienceDirect](www.sciencedirect.com/science/journal/00224375)
|
|
|
+
|
|
|
+# Journal of Safety Research
|
|
|
+
|
|
|
+journal homepage: [www.elsevier.com/locate/jsr](https://www.elsevier.com/locate/jsr)
|
|
|
+
|
|
|
+
|
|
|
+
|
|
|
+
|
|
|
+
|
|
|
+# Factors associated with different types of freight crashes: A macro-level analysis
|
|
|
+
|
|
|
+# Eun Jin Shin
|
|
|
+
|
|
|
+*Department of Public Administration and Graduate School of Governance, Sungkyunkwan University, 25-2 Sungkyunkwan-ro Hoam hall 50908, Jongno-gu, Seoul 03063, Republic of Korea*
|
|
|
+
|
|
|
+## ARTICLE INFO
|
|
|
+
|
|
|
+*Keywords:* Traffic crash Freight crash Macro-level analysis Multivariate model
|
|
|
+
|
|
|
+Bayesian conditional autoregressive model
|
|
|
+
|
|
|
+#### ABSTRACT
|
|
|
+
|
|
|
+*Introduction:* Despite evidence showing higher fatality rates in freight-related crashes, there has been limited exploration of their spatial distribution and factors associated with such distribution. This gap in the literature primarily stems from the focus of existing studies on micro-level factors predicting the frequency or severity of injuries in freight crashes. The present study delves into the factors contributing to freight crashes at the neighborhood level, particularly focusing on different types of freight crashes: collisions involving a freight vehicle and a passenger vehicle, crashes between freight vehicles, and freight vehicle–non-motorized crashes. *Method*: This study analyzes traffic crash data from the urbanized region of Seoul, collected between 2016 and 2019. To effectively deal with spatial autocorrelation and model different types of crashes in a unified framework, a Bayesian multivariate conditional autoregressive model was employed. *Results:* Findings show substantial differences in the factors associated with various types of freight crashes. The predictors for crashes between freight vehicles diverge significantly from those for freight vehicle–non-motorized crashes. Crashes between freight vehicles are relatively more influenced by road network structure, while freight crashes involving non-motorized users are relatively more affected by the built environment and freight facilities than the other crash types examined. Freight vehicle–passenger vehicle crashes fall into an intermediate category, sharing most predictors with either of the other two types of freight crashes. *Conclusions and Practical Applications*: The findings of this study offer valuable lessons for transportation practitioners and policymakers. They can guide the formulation of effective land use policies and infrastructure planning, specifically designed to address the unique characteristics of different types of freight crashes.
|
|
|
+
|
|
|
+### **1. Introduction**
|
|
|
+
|
|
|
+With the unprecedented growth in demand for road freight in numerous countries worldwide, associated freight traffic volume has increased considerably over the past decade [\(International Transport](#page-15-0) [Forum, 2023](#page-15-0)). Alongside large countries such as China and Canada, South Korea has also witnessed a growth in road freight travel, with an increase of approximately 36% recorded between 2010 and 2019. Consequently, communities with many freight vehicles passing through have encountered various problems, including traffic congestion, noise, and air pollution (Ranaiefar & [Amelia, 2011; Houston et al., 2014](#page-15-0)). Furthermore, the increased volume of freight vehicles has led to a rise in the number of freight-related crashes and their share of the total crashes in many urban areas ([Das et al., 2020; Park](#page-15-0) & Park, 2018).
|
|
|
+
|
|
|
+Since 2014, the total number of freight crashes in South Korea has
|
|
|
+
|
|
|
+been generally increasing, with an average annual increase rate of 0.8%.1 This trend is noteworthy because freight crashes have been associated with higher fatality and serious injury rates than other types of traffic crashes. Freight crashes involving non-motorized users pose a particularly high risk of causing fatal or serious injuries, as pedestrians and bicyclists usually lack physical protection. For example, [Wang et al.](#page-16-0) [\(2022\)](#page-16-0) revealed that, compared to crashes between passenger vehicles and non-motorized users, incidents involving freight vehicles and nonmotorized users have been found to increase the likelihood of severe injuries by 203%.
|
|
|
+
|
|
|
+While initially introduced before the 1990s, regulations regarding freight vehicle safety in South Korea have undergone substantial revisions over a recent decade (see Appendix A for details), reflecting the significant increase in road freight vehicle travel within the country. Notably, most of these regulations and associated policies have focused
|
|
|
+
|
|
|
+*E-mail address:* [eunjinshin@g.skku.edu](mailto:eunjinshin@g.skku.edu). 1 These statistics were sourced from the website of the Korea Road Traffic Authority, which contains data starting from 2005.
|
|
|
+
|
|
|
+<span id="page-1-0"></span>on freight vehicle drivers' behavior (e.g., ensuring adequate rest time for truck drivers) and vehicle maintenance (e.g., requiring specific safety equipment for freight vehicles). For instance, since the mid-2010s, numerous freight vehicle safety campaigns and educational policy programs have been implemented to educate and increase awareness related to safety among truck drivers. However, there has been a relative lack of policy interest on enhancing road infrastructure and the built environment to reduce freight-related crashes.
|
|
|
+
|
|
|
+This focus of safety regulations or policies may reflect the micro-level approach of most existing studies on freight crashes, which examined person-, vehicle-, or crash-related factors predicting the frequency or injury severity of freight crashes (Malczyk & [Bende, 2019; Pokorny](#page-15-0) & [Pitera, 2019; Rahimi et al., 2020](#page-15-0)). These micro-level studies have offered little insight into land use and infrastructure planning to improve road traffic safety. Although a few existing macro-level studies do exist ([Yang et al., 2021a, Yang et al., 2021b](#page-16-0)), they have not differentiated between different types of freight crashes. Given that different transportation modes exhibit unique spatial and temporal patterns of movement, the interplay of different modes involved in traffic crashes likely results in distinct spatial patterns. Furthermore, prior studies have been based on U.S. cities with a strong automobile-oriented spatial structure, thereby limiting the generalizability of their findings.
|
|
|
+
|
|
|
+Against this backdrop, this study aims to address the gaps in literature by exploring the distribution of different types of freight crashes at the neighborhood level, as well as the associated factors in the urbanized Seoul region. Specifically, it examines a range of neighborhood-level factors associated with different types of crashes, namely, freight vehicle–passenger vehicle, freight vehicle–freight vehicle, and freight vehicle–non-motorized crashes. This study employs a Bayesian multivariate conditional autoregressive model to adjust for both spatial autocorrelations in traffic crash data and correlations between different types of freight crashes. The urbanized Seoul region provides an ideal empirical setting for investigating different types of freight crashes because nonmotorized transportation trips make up a relatively high share of total trips, thus resulting in a relatively high frequency of freight vehicle–nonmotorized crashes.
|
|
|
+
|
|
|
+The remainder of this paper is organized as follows. Section 2 reviews the literature on the subject and discusses its limitations. [Section 3](#page-3-0) describes the context of the study area and explains the data sources and methodology. [Section 4](#page-8-0) presents the primary and additional analysis results. [Section 5](#page-10-0) discusses the results and their policy implications. [Section 6](#page-13-0) summarizes the main findings and proposes avenues for future research.
|
|
|
+
|
|
|
+# **2. Literature review**
|
|
|
+
|
|
|
+## *2.1. Factors influencing traffic crashes and freight-related crashes*
|
|
|
+
|
|
|
+Prior research on macroscopic traffic crash analysis has investigated various area-level crash determinants. These can be broadly categorized into four groups: transportation infrastructure, non-motorized infrastructure (in studies focusing on non-motorized crashes), built environment characteristics, and exposure variables.
|
|
|
+
|
|
|
+Transportation infrastructure, particularly road density and network measures, has been a frequent subject of investigation. Research has found that the density of various road types can significantly predict traffic crash occurrences. However, the nature of these relationships depends on the road type and crash category. For instance, while a denser network of highways or freeways tends to positively correlate with crashes between motor vehicles, likely due to increased vehicle speeds ([Yang et al., 2021a](#page-16-0)), it is associated with fewer non-motorized crashes ([Guerra et al., 2019; Barajas, 2018](#page-15-0)). However, crashes involving pedestrians or cyclists are more likely to occur on arterial ([Barajas, 2018; Dumbaugh](#page-15-0) & Li, 2010) or secondary ([Shin, 2023](#page-15-0)) roads.
|
|
|
+
|
|
|
+The influence of road network structure on traffic crashes has also garnered attention in macroscopic traffic crash analysis. Intersection density is the most frequently examined variable. Intersections, as convergence points for different transportation modes, likely increase the potential for traffic crashes [\(Abdel-Aty et al., 2013; Siddiqui et al.,](#page-15-0) [2012\)](#page-15-0). However, they can also contribute to reducing their risk by decreasing vehicle speeds [\(Merlin et al., 2020](#page-15-0)). Road connectivity, another frequently investigated element of road networks, has been measured in various ways, including the link-to-node ratio and the meshedness coefficient; among these, [Wang et al. \(2013\)](#page-16-0) demonstrated that the meshedness coefficient offers the most nuanced representation of road structure. Other road network measures frequently explored in micro-level crash analysis, such as indicators related to street centrality, appear less explored in macro-level crash analysis, although they may also play important roles in explaining spatial distribution of traffic crashes. Overall, the relationships between these road network measures and traffic crash occurrences are found to be inconsistent across studies (Choi & [Ewing, 2021; Marshall](#page-15-0) & Garrick, 2011).
|
|
|
+
|
|
|
+The role of non-motorized infrastructure networks, such as bicycle lanes and crossing facilities, has been further investigated in studies examining non-motorized crashes. The relationship between bicycle lanes and bicycle crashes remains inconclusive, partly due to varying relationships depending on the type of bicycle lane [\(Chen, 2015; Shin,](#page-15-0) [2023\)](#page-15-0). Regarding sidewalk networks, both sidewalk density and network connectivity play a crucial role in predicting the occurrence of non-motorized crashes (Osama & [Sayed, 2017](#page-15-0)). However, the effects of sidewalks on traffic crashes are often found to be counterintuitive, as many studies reported a positive correlation between sidewalk-related variables and non-motorized crashes ([Cai et al., 2016; Qiao et al.,](#page-15-0) [2020; Romanow et al., 2012](#page-15-0)).
|
|
|
+
|
|
|
+Variables associated with the built environment primarily include measures of density, such as employment and population density, along with land use factors, such as the proportions of different land types and the mix of land use ([Chen et al., 2018; Dumbaugh](#page-15-0) & Li, 2010; Wier et al., [2009\)](#page-15-0). Recent research has also employed point of interest (POI) data to delve deeper into the relationship between specific land use types and traffic crash rates. For example, [Jia et al. \(2018\)](#page-15-0) demonstrated positive correlations between the density of certain establishments, including hospitals and banks, and the frequency of traffic crashes in Suzhou, China. Numerous studies have demonstrated that areas with a greater mix of land uses tend to see a higher rate of traffic crashes, especially those involving non-motorized users ([Chen, 2015; Chen et al., 2018;](#page-15-0) [Shin, 2023\)](#page-15-0). Nonetheless, the influence of other built environment characteristics on traffic crashes has shown significant variability across studies (for a more detailed review, see [Merlin et al., 2020](#page-15-0)).
|
|
|
+
|
|
|
+These inconsistent results of the relationships between the many variables explored and traffic crashes can be attributed to differences in research design and study area contexts, as well as to variations in the inclusion of exposure variables in the crash analysis ([Merlin et al.,](#page-15-0) [2020\)](#page-15-0). An exposure variable measures a potential risk of crash by accounting for the amount of travel. Technological advancements, such as automatic traffic recorders, have significantly eased the collection of motor vehicle traffic volume data at zonal or neighborhood levels.
|
|
|
+
|
|
|
+However, accurately and comprehensively measuring nonmotorized transportation volumes remains a challenge [\(Ding et al.,](#page-15-0) [2020\)](#page-15-0). This difficulty has led some studies to exclude variables like bicycle traffic volume, even though this omission can bias estimates of other factors, such as bicycle lanes. In other studies, researchers have sought alternative data sources, including census data about bicyclist or pedestrian commuter shares, manual counts of cyclists or pedestrians at specific locations, and data from GPS-enabled devices (e.g., fitness trackers) to estimate bicycle exposure ([Barajas, 2021; Wier et al., 2009](#page-15-0)). However, as [Barajas \(2021\)](#page-15-0) emphasized, each of these data sources has its own set of limitations. The recent surge in bike-sharing services' popularity across many global cities offers an additional data source. Consequently, several recent studies have used these data to estimate bicycle traffic volume ([Ding et al., 2020; Ding et al., 2021](#page-15-0)).
|
|
|
+
|
|
|
+Despite ample research on freight crashes, area-level factors
|
|
|
+
|
|
|
+influencing them remain underexplored. That is, previous studies examining predictors of freight crashes have primarily centered around micro-level factors, ranging from person-related characteristics to vehicle- and crash-related factors. These studies have generally reported that the sociodemographic and behavioral characteristics of freight vehicle drivers are the most significant predictors of freight crashes, although other factors, such as vehicle designs and driving environment (e.g., weather), may also contribute [\(Dong et al., 2017; Duncan et al.,](#page-15-0) [1998; Taylor et al., 2018; Zhu](#page-15-0) & Srinivasan, 2011). However, this strand of research provides little insight into how land use and infrastructure planning on a larger spatial scale can help reduce freight crashes.
|
|
|
+
|
|
|
+[Yang et al., \(2021a\)](#page-16-0) is one of the few studies exploring the area-level factors associated with the occurrence of freight truck-related crashes in the Los Angeles region. Beyond the abovementioned factors, the authors investigated the role of factors uniquely related to freight crashes, such as warehouse density and airport proximity. They revealed that the locations of certain freight-related facilities and truck traffic volumes were highly correlated with freight truck crashes, while being insignificant predictors of total vehicle crashes. Other factors, including the densities of freeways and primary roads are also significant predictors of freight crashes. In a separate study, [Yang et al., \(2021b\)](#page-16-0) used machine-learning methods to explore the factors predicting the occurrence of freight crashes in the same region. Results similarly showed that freight-related facilities and road networks were both important in predicting the spatial distribution of freight-related crashes. However, as these studies focused only on total freight crashes, little is known about the distinct distribution patterns of different types of freight crashes.
|
|
|
+
|
|
|
+### *2.2. Research methods and design in macro-level traffic crash analysis*
|
|
|
+
|
|
|
+This section provides a comprehensive review on research methods and designs employed in macro-level traffic crash analysis. Poisson and negative binomial regression are classic models frequently used to estimate crash frequencies, which are non-negative integers ([Dong et al.,](#page-15-0) [2017; Dumbaugh](#page-15-0) & Li, 2010); extensions of these basic models, such as the Poisson-lognormal model [\(Heydari et al., 2022\)](#page-15-0) and zero-inflated negative binomial model [\(Shankar et al., 2003](#page-15-0)), have also been utilized to address over-dispersion and excess zero limitations. In studies where traffic crash density was the dependent variable, ordinary least square regression and log-linear models were also adopted ([Gonz](#page-15-0)´ alez [et al., 2019; Loukaitou-Sideris et al., 2007](#page-15-0)). Nevertheless, these traditional methods possess inherent limitations, particularly in addressing (1) spatial correlations observed in proximate data points and (2) the relationships among various crash types.
|
|
|
+
|
|
|
+Specifically, traffic crash occurrence in one neighborhood often correlates with adjacent areas, a fact widely acknowledged in the field (Pljaki´[c et al., 2019; Yang et al., 2021a\)](#page-15-0). This poses a challenge to traditional regression models, which are based on the assumption of independent observations. Violations of this assumption lead to inefficiency and bias in model outcomes. To counter this issue, researchers have turned to alternative approaches such as spatial lag and spatial error models ([Jia et al., 2018; Pljaki](#page-15-0)´c et al., 2019); while the spatial lag model incorporates spatial dependence into the response variable, the latter integrates it into the model's error term (Chi & [Zhu, 2019\)](#page-15-0).
|
|
|
+
|
|
|
+Another popular approach to dealing with spatial autocorrelation in traffic crash data is to use Bayesian spatial models, including the Bayesian Conditional Autoregressive model [\(Barajas, 2018, Chen,](#page-15-0) [2015\)](#page-15-0). These models handle spatial autocorrelation by introducing spatial random effects, which can capture unobserved spatially structured variability in the data, or by incorporating spatial priors. This approach also helps mitigate a boundary crash data assignment problem, which is also closely related to spatial autocorrelation. The boundary crash data assignment problem arises because traffic crashes occur on or near the boundary between two defined spatial units; the assignment of these crashes can significantly influence their spatial distribution, and thus, the observed degree of spatial autocorrelation in the data. For example, [Zhai et al. \(2018\)](#page-16-0) used an iterative approach with Bayesian spatial models to allocate boundary crashes to neighboring analysis zones, thereby enhancing model estimation and predictive performance.2
|
|
|
+
|
|
|
+Traditional count or linear regression models separately conducted for different types of crashes also overlook correlations between them. However, different types of crashes often share unobserved area-specific characteristics [\(Lee et al., 2015\)](#page-15-0). Therefore, it is problematic to conduct independent, separate analyses for each crash type. As such, some studies have adopted multivariate techniques that model various types of crashes within a unified framework, demonstrating their superior performance over univariate models (El-Basyouny & [Sayed, 2009; Lee](#page-15-0) et al., 2015; Ma & [Kockelman, 2006](#page-15-0)).
|
|
|
+
|
|
|
+With advancements in computational capabilities, recent traffic safety literature has begun applying machine learning approaches for various purposes, including handling data imbalance and predictive modeling (e.g., [Ding et al., 2022; Yang et al., 2021b\)](#page-15-0). For instance, [Ding](#page-15-0) [et al. \(2022\)](#page-15-0) used a machine learning technique—specifically, a generative approach—to manage crash data with excessive zeros by generating synthetic crash data.
|
|
|
+
|
|
|
+Finally, defining appropriate spatial units for analysis stands as a pivotal concern in macroscopic traffic crash analysis [\(Xu et al., 2018](#page-16-0)). This is underscored by the Modifiable Areal Unit Problem (MAUP), a challenge encountered in spatial analysis that manifests when identical analyses produce different results due to variations in the size or boundaries of the chosen spatial units, often attributed to differences in scale and zonal configurations. To mitigate the impact of the MAUP, several strategies have been adopted when disaggregated data or data at multiple spatial scales were available. These strategies included scale sensitivity analysis, which involves analyzing data at various spatial scales, and the use of automated zone design techniques, which allow for creating spatial units that more accurately represent the underlying data (Ziakopoulos & [Yannis, 2020; Xu et al., 2018](#page-16-0)).
|
|
|
+
|
|
|
+#### *2.3. Limitations of existing literature*
|
|
|
+
|
|
|
+Research examining freight-related crashes at the macro-level has been limited, with no previous studies specifically focusing on the different types of freight crashes. Despite increased media attention ([Choi, 2020; Kim](#page-15-0) & Park, 2021), the current understanding of the spatial distribution of freight vehicle–non-motorized crashes remains particularly limited. This is perhaps because the occurrence of this type of crash is relatively rare in previously investigated regions, such as Los Angeles, where there are relatively few non-motorized users. Consequently, performing a meaningful statistical analysis at the neighborhood level becomes challenging (Pokorny & [Pitera, 2019\)](#page-15-0). Although both freight vehicle–non-motorized and freight vehicle–passenger vehicle crashes involve freight vehicles, their spatial distributions may differ, as the interplay of different modes involved in traffic crashes likely leads to distinct spatial dynamics of collisions. Moreover, as previous area-level analyses of freight crashes have focused on U.S. cities with a strong automobile-oriented spatial structure [\(Yang et al., 2021a; Yuan](#page-16-0) & Wang, [2021\)](#page-16-0), it remains unclear whether these findings are applicable to other cities or regions with different spatial structures.
|
|
|
+
|
|
|
+Regarding methodology, despite continual advancements, limited effort has been made to address both spatial autocorrelation and correlations between different types of crashes ([Lee et al., 2015; Ma](#page-15-0) & [Kockelman, 2006\)](#page-15-0), none of which explored the spatial distribution of various types of freight crashes. Different types of freight crashes might
|
|
|
+
|
|
|
+<sup>2</sup> Other approaches to handle the boundary crash problems include a multiple membership multilevel modeling approach, which allows multiple zones to affect individual crashes [\(Park et al., 2022\)](#page-15-0). Notably, these approaches addressing the boundary crash assignment problem require data detailing the exact locations of crashes.
|
|
|
+
|
|
|
+<span id="page-3-0"></span>share common unobserved characteristics (e.g., road conditions, regulations), as they all involve a freight vehicle. Ignoring these correlations among different crash types and spatial autocorrelations can decrease the model efficiency and precision of parameter estimates.
|
|
|
+
|
|
|
+Therefore, the present study is designed to analyze the neighborhood-level distribution of different types of freight crashes in the urbanized Seoul region—a little-researched area characterized by a denser built environment and a high volume of non-motorized users. This study employs a Bayesian multivariate conditional autoregressive approach to address not only spatial autocorrelation issues but also potential correlations between different types of freight crashes.
|
|
|
+
|
|
|
+### **3. Study area context and research design**
|
|
|
+
|
|
|
+### *3.1. Selection of the study area*
|
|
|
+
|
|
|
+The study area is the urbanized Seoul region, which encompasses the city of Seoul and portions of Gyeonggi province. Many freight-related facilities are located in Gyeonggi province due to the relatively low land rent and its spatial proximity to Seoul, which can potentially affect the spatial patterns of freight movement and related crashes. This study restricts its focus to the urbanized area. The Korean government designates *haengjungdong* as a sub-administrative unit of a city for urbanized areas and uses *eup* or *myeon* for non-urbanized areas; therefore, areas classified as the latter, as well as those lacking relevant data, are excluded from the analysis. Consequently, the study area represents approximately 80% of the total population of Seoul and Gyeonggi province, encompassing 754 haengjungdongs, out of which 422 being in Seoul.
|
|
|
+
|
|
|
+### *3.2. Data and variables*
|
|
|
+
|
|
|
+The study's dependent variable is the frequency of different types of freight crashes, which include freight vehicle–passenger vehicle crashes, freight vehicle–freight vehicle crashes, and freight vehicle–non-motorized crashes. This study utilizes 2016–2019 traffic crash data, obtained from the Korea Road Traffic Authority (KoROAD). The total number of crashes from 2016 to 2019 for each type is aggregated since some crash types are too rare to analyze on an annual basis. The crash dataset comprises information from police-reported crashes. To be included in the dataset, a traffic crash must have met two criteria: (1) it should have resulted in either injury or death, and (2) it must have involved motorized or non-motorized vehicles (e.g., bicycles) traveling on public roads or trafficways that are customarily open to the public [\(KoROAD](#page-15-0) [Traffic Analysis Accident System website, 2023\)](#page-15-0). The dataset offers indepth details, including the types of vehicles involved (e.g., road freight vehicles,<sup>3</sup> passenger vehicles, pedestrians, bicyclists) and the primary purpose of freight vehicles (commercial vs. non-commercial).
|
|
|
+
|
|
|
+This study focuses exclusively on crashes involving commercial road freight vehicles4 (hereafter referred to as "freight crashes" or "freightrelated crashes"). Whereas the crash data does not distinguish between freight vehicles by weight, aggregated statistics from the National Police Agency (2015) showed that over 92% of commercial freight vehicle crashes involve medium to heavy freight vehicles (i.e., freight vehicles with gross weights greater than 3.5 tons). This definition of freight crashes aligns more closely with the definitions used in existing freight
|
|
|
+
|
|
|
+crash literature ([Yang et al., 2021a; Yuan](#page-16-0) & Wang, 2021).
|
|
|
+
|
|
|
+Table 1 presents the characteristics of crashes involving commercial freight vehicles in the study area between 2016 and 2019. Freightrelated crashes account for about 3.6% of the total crashes in the study area. Among them, the most prevalent are those between a freight and a passenger vehicle (60.3% of total freight crashes), followed by freight vehicle–non-motorized crashes (19.3%), and crashes between freight vehicles (8.6%).
|
|
|
+
|
|
|
+It has been widely recognized that not all traffic crashes, particularly minor ones or those not involving a motor vehicle, are reported to the police ([Stutts et al., 1990](#page-16-0)). Accordingly, the crash data used in this study do not represent the total number of crashes. However, given that freight-related crashes typically result in more severe injuries or fatalities compared to non-freight-related crashes [\(Desapriya et al., 2010;](#page-15-0) [Wang et al., 2022](#page-15-0)), the underreporting is relatively less concerning when compared to studies focusing on non-freight crashes. Despite this limitation, the utilization of the police-reported crash database with standardized information facilitates the analysis of the relative frequency of crash incidents across the study area. Nearly all the crash records in the dataset (99.9%) encompassed complete information related to the analysis.
|
|
|
+
|
|
|
+The smallest available geographic area for traffic crash data is the haengjungdong. Moreover, haengjungdong has often been used as a proxy for neighborhoods (e.g., [Shin, 2021](#page-15-0)), which is why it was chosen as the unit of analysis in this study. As reported in Appendix B, a haengjungdong spans an average area of 1.1 square miles, with an average population of approximately 23,800. There is a notable variation in size, with a standard deviation of 1.6 square miles. To provide perspective, the size of a typical haengjungdong (referred to as 'neighborhood' from here on) is comparable to that of a circle with a 0.6-mile radius. Its average population density is approximately 49,700 residents per square mile.
|
|
|
+
|
|
|
+The independent variables, which were identified based on a review of previous studies, can be divided into six categories: freight-related facilities, roadway network infrastructure, built environment characteristics, crash risk exposure, non-motorized infrastructure, and other variables.
|
|
|
+
|
|
|
+Specific variables related to freight-related facilities include the distance to the Incheon International Airport (ICN), distance to the nearest logistics terminal, and warehouse density. Given that warehouses are relatively more prevalent than other freight facilities in the study area, the density rather than distance is used as a measure. Spatial proximities to the Gimpo airport (secondary airport) and the Port of Incheon (largest nearby seaport) were excluded from the final analysis because of multicollinearity issues; these variables are highly correlated with the distance to the ICN.
|
|
|
+
|
|
|
+**Table 1** Frequency and percentage of freight crashes in the study area.
|
|
|
+
|
|
|
+| | Frequency | Percentage<br>among total<br>crashes | Percentage<br>among freight<br>crashes |
|
|
|
+|----------------------------------------------------------------------------------------|-----------|--------------------------------------|----------------------------------------|
|
|
|
+| Total crashes | 275,879 | 100 | – |
|
|
|
+| Total freight-related crashes<br>(crashes involving<br>commercial freight<br>vehicles) | 10,026 | 3.6 | |
|
|
|
+| Freight vehicle-passenger<br>vehicle | 6,041 | 2.2 | 60.3 |
|
|
|
+| Freight vehicle-freight<br>vehicle | 859 | 0.3 | 8.6 |
|
|
|
+| Freight vehicle-non<br>motorized user | 1,932 | 0.7 | 19.3 |
|
|
|
+
|
|
|
+Note: The "total freight-related crashes" category includes the three examined freight crash types as well as single-freight vehicle crashes and collisions involving a freight vehicle and other unexamined vehicle types, such as motorcycles.
|
|
|
+
|
|
|
+<sup>3</sup> The dataset is limited in that it does not provide more detailed categories of
|
|
|
+
|
|
|
+road freight vehicles. 4 In South Korea, the colors, starting numbers, and letters on the vehicle license plate are specific to the broad category (freight vehicle, non-freight vehicle) or purpose of vehicle (commercial, non-commercial), which is also used for police crash reports. Therefore, it allows for a clear differentiation between freight and non-freight vehicles without relying on subjective judgment.
|
|
|
+
|
|
|
+Road infrastructure includes the densities of different road types, such as highways. The data classified non-highway roads based on the number of lanes, and the density of each road type was calculated by dividing its length by neighborhood area. In terms of road network structures, three variables are considered: intersection density, closeness centrality (a measure of street accessibility), and meshedness coefficients (a measure of street connectivity). A link-node ratio was omitted from the final analysis due to its strong correlations with the meshedness coefficient. Closeness centrality measures a node's proximity to all other nodes in a network [\(Crucitti et al., 2006\)](#page-15-0). Higher closeness centrality means better street accessibility, and can be expressed as:
|
|
|
+
|
|
|
+$$CC(i) = \frac{N-1}{\sum_{j \in G} d(i,j)}$$
|
|
|
+ (1)
|
|
|
+
|
|
|
+where N is the total number of nodes in the study area G and *d* (i, j) is the shortest path distance from node *i* to all other nodes *j* in the study area. For each neighborhood, the average value of closeness centrality was computed. The meshedness coefficient is also calculated as follows:
|
|
|
+
|
|
|
+$$Meshednesscoefficient = \frac{E - N + 1}{2N - 5},\tag{2}$$
|
|
|
+
|
|
|
+where E is the number of edges and N is the number of nodes in each neighborhood. A higher meshedness coefficient indicates greater road connectivity.
|
|
|
+
|
|
|
+Regarding built environment characteristics, this study incorporates population density, employment density, the proportion of residential lands, and land use mix into the analysis. The entropy index is an indicator of land-use mixes ranging from 0 to 1, which can be expressed as
|
|
|
+
|
|
|
+$$Entropyindex = -\sum_{i}[(lu_i)(lnlu_i)]/ln(c)$$
|
|
|
+(3)
|
|
|
+
|
|
|
+where *lui* is the proportion of key land use type *i* (e.g., commercial/office, residential, public facilities, industrial), and *c* is the number of primary land use types. A higher entropy index value indicates a higher mix level. In addition to broad land use variables, this study utilizes POI data to consider detailed land use types. For example, the densities of large grocery stores and shopping malls are included because they tend to attract medium or heavy trucks due to the need for regular deliveries of diverse goods, waste management, and maintenance operations. Reflecting the local context and media reports on collisions between freight vehicles and children [\(Choi, 2020; Kim](#page-15-0) & Park, 2021), the density of elementary schools is also incorporated into the regression models; children are particularly vulnerable to truck drivers' blind spots because of their short stature (Kolliner, 2018).
|
|
|
+
|
|
|
+Crash risk exposure variables are also added to the regression models. The Korean Transport Institute provided data on the average daily traffic volume for both trucks and passenger vehicles at the neighborhood (i.e., haengjungdong) level, which were derived from GPS and navigation apps that tracked individual vehicle travel paths. This dataset, covering approximately 97% of the country's roads, has missing values for about 1.5% of the neighborhoods in the study area, which were excluded from the analysis. Based on this dataset, the densities of truck and passenger vehicle traffic volumes were computed. It is worth noting, however, that the truck traffic volume data do not distinguish trucks by weight or purpose. This study further includes the square of the passenger vehicle traffic volume density—which typically constitutes most of the total traffic volume—in its analysis to examine its potential non-linear relationship with crashes, as hinted at by some previous research (e.g., [Martin, 2002; Liu et al., 2020](#page-15-0)).
|
|
|
+
|
|
|
+Average daily bicycle traffic volume data are only available for the city of Seoul, therefore, the bicycle exposure variable is considered only in the city-focused additional analyses reported in [Section 4.2.](#page-10-0) This dataset contains information about the average daily bicycle traffic volume, as estimated by the city government using GPS data from bicycle-sharing services in May 2019. The data were provided at the hexagon level, with each hexagon representing an area of 0.104 square kilometers. Therefore, the bicycle volume count was aggregated to the neighborhood level using the area-weighted average method. A potential bias of this dataset is that it does not account for all bicycle trips, specifically those not involving bicycle-sharing services. However, considering that (1) a significant proportion of cyclists in Seoul reportedly utilize bike-sharing services,5 and (2) the bicycle volume estimation is derived from GPS data providing detailed location information on these shared bike trips, this dataset is likely to offer more representative and reliable estimates than the traditional bicycle traffic volume data discussed in [Section 2.1.](#page-1-0)
|
|
|
+
|
|
|
+Variables related to non-motorized infrastructure—sidewalks, crossing facilities, and various types of bicycle lanes (protected bicycle lane; exclusive bicycle lane; bicycle-priority lane; shared bicycle–pedestrian paths)—are further incorporated into the analysis. Regarding sidewalks, network connectivity is considered along with sidewalk density. Sidewalk connectivity, as further explained in [Table 2](#page-5-0), reflects the degree to which different sections of the sidewalk network are interconnected. However, it is worth noting that, owing to limited data availability, non-motorized infrastructure variables, except for sidewalk-related variables, are only considered in additional analyses focusing on Seoul (in [Section 4.2](#page-10-0)). [Table 2](#page-5-0) presents the definitions and data sources of the independent variables.
|
|
|
+
|
|
|
+Finally, a dummy variable indicating whether a neighborhood belongs to Seoul is added, accounting for any unobserved differences between Seoul and the outside areas. The size of neighborhoods is controlled in the model to account for substantial differences in areas between neighborhoods in the study area. Summary statistics of the variables are also reported in [Table 2.](#page-5-0)
|
|
|
+
|
|
|
+### *3.3. Study area context*
|
|
|
+
|
|
|
+As mentioned, most safety regulations for freight vehicles in South Korea have predominantly focused on the vehicles themselves or their drivers. Consequently, policies addressing other facets are relatively rare, with only a few exceptions. One long-standing regulation in the study area, for instance, prohibits freight vehicles weighing 3.5 tons or heavier from entering city centers between 7 a.m. and 10p.m. on weekdays. This likely impacts the distribution of freight crashes by restricting daytime freight vehicle movement in central neighborhoods. In line with recent pedestrianization trends, several commercial districts have established car-free zones, further limiting freight vehicle access. Additionally, there is a regulation mandating commercial freight vehicles to park only in designated garages overnight. However, due to a significant shortage of such garages and the impractical locations of those available, freight vehicles are often found illegally parked on sidewalks or road shoulders overnight ([Kang, 2021; Shin, 2023\)](#page-15-0).
|
|
|
+
|
|
|
+[Fig. 1](#page-6-0)(a)–(c) reveals the land use patterns of the study area, demonstrating notable variations across different neighborhoods. Overall, the city of Seoul dominates in terms of residential and commercial land shares compared to the rest of the study area. However, substantial differences exist in the distribution of residential and commercial lands within Seoul. Commercial areas predominantly cluster around the city center and employment subcenters, such as Gangnam
|
|
|
+
|
|
|
+<sup>5</sup> The city government has disclosed cumulative annual data for bike-sharing service membership users. This reporting format makes it difficult to pinpoint an accurate number of bike-sharing users within any specific timeframe. Nevertheless, a discernible trend in the available data illustrates a yearly average increase of approximately 516,000 in the cumulative membership of the bike-sharing service during the studied period. Given that the population of cyclists in Seoul was estimated approximately 1,707,000 based on the 2019 Seoul survey data, it is reasonable to assume that a substantial portion of these cyclists utilize the bike-sharing services.
|
|
|
+
|
|
|
+<span id="page-5-0"></span>**Table 2** Definition of variables, summary statistics, and data sources.
|
|
|
+
|
|
|
+| Variable | Description | Mean | Std. Dev. | Data source |
|
|
|
+|----------------------------------------------------------------|------------------------------------------------------------------------------------------------------------------------|--------------|--------------|------------------------------------------------|
|
|
|
+| Freight-related facilities | | | | |
|
|
|
+| Distance to ICN | Distance to the Incheon international airport (miles) | 30.09 | 6.52 | Google map |
|
|
|
+| Distance to logistics terminal | Distance to the nearest logistics terminal (miles) | 4.99 | 2.68 | National logistics information<br>center |
|
|
|
+| Warehouse density | Number of warehouses/mi2 | 0.17 | 0.81 | Local administrative permit<br>data |
|
|
|
+| Roadway characteristics | | | | |
|
|
|
+| Highway density | Highway length (in miles)/mi2 | 0.17 | 0.40 | National spatial data |
|
|
|
+| Density of roads with 8-or-more<br>lanes | Length of roads with 8-or-more lanes (in miles)/mi2 | 0.80 | 1.25 | infrastructure portal |
|
|
|
+| Density of roads with 2–7 lanes | Length of roads with 2–7 lanes (in miles)/mi2 | 6.25 | 3.46 | |
|
|
|
+| Density of 1-lane roads | Length of 1-lane roads (in miles)/mi2 | 18.75 | 14.42 | |
|
|
|
+| Road intersection density | Number of major road intersections/mi2 | 7.09 | 7.99 | |
|
|
|
+| Closeness centrality | A measure of how close a particular node in a neighborhood is, on average, | 0.00004 | 0.00001 | Open street map |
|
|
|
+| Meshedness coefficient | to all other nodes in the network<br>A measure of the interconnectedness of a road network | 0.23 | 0.24 | National spatial data<br>infrastructure portal |
|
|
|
+| Built environment | | | | |
|
|
|
+| Employment density | Number of employees (in thousands)/mi2 | 38.99 | 51.76 | Korean Statistical Information |
|
|
|
+| Population density | 2<br>Number of residents (in thousands)/mi | 49.74 | 32.48 | Service |
|
|
|
+| Land use mix | Entropy index (ranging from 0 to 1) | 0.53 | 0.18 | |
|
|
|
+| Residential land share | Proportion of residential land | 0.35 | 0.22 | National spatial data<br>infrastructure portal |
|
|
|
+| Grocery store density | Number of large grocery stores/ mi2 | 0.41 | 1.37 | Local administrative permit |
|
|
|
+| | 2 | | | |
|
|
|
+| Shopping mall density | Number of large shopping malls/ mi | 0.47 | 1.68 | data |
|
|
|
+| Elementary school density | Number of elementary schools/mi2 | 3.16 | 2.76 | |
|
|
|
+| Hospital density<br>Exposure variables | Number of hospitals/ mi2 | 0.24 | 0.82 | https://www.data.go.kr |
|
|
|
+| Truck traffic volume density | Average daily truck traffic volume (in thousands)/mi2 | 5.29 | 4.73 | Korea transport database |
|
|
|
+| Passenger vehicle traffic volume | Average daily passenger vehicle traffic volume (in thousands)/mi2 | 42.45 | 36.00 | |
|
|
|
+| density<br>Squared passenger vehicle traffic<br>volume density | Passenger vehicle traffic volume density2 | 3096.5 | 5748.6 | |
|
|
|
+| Bicycle traffic volume densitya | Average daily bicycle count (in thousands)/mi2 | 14.53 | 13.59 | Seoul policy map portal |
|
|
|
+| Non-motorized infrastructure | Length of sidewalks (in miles)/mi2 | | | |
|
|
|
+| Sidewalk density | | 10.20 | 5.62 | National spatial data |
|
|
|
+| Sidewalk connectivity | Ratio of the actual number of sidewalk links to the maximum possible<br>number of links within a specific neighborhood | 0.19 | 0.04 | infrastructure portal |
|
|
|
+| Bicycle-priority lane densitya | Length of bicycle-priority lanes (in miles) /mi2 | | | |
|
|
|
+| Exclusive bicycle lane densitya | Length of unprotected exclusive bicycle lanes (in miles)/mi2 | 0.75<br>0.14 | 1.97<br>0.73 | |
|
|
|
+| Protected bicycle lane densitya | Length of protected bicycle lanes (in miles)/mi2 | | | |
|
|
|
+| | Length of shared bicycle/pedestrian paths (in miles)/mi2 | 0.53 | 1.17 | |
|
|
|
+| Shared bicycle–pedestrian path<br>densitya | | 2.70 | 3.02 | |
|
|
|
+| Crossing facility densitya | Number of crossing facilities/mi2 | 210.25 | 110.48 | |
|
|
|
+| | | | | City of Seoul |
|
|
|
+| Other variables | | | | |
|
|
|
+| Neighborhood size<br>Seoul | Size of haengjungdong (mi2<br>)<br>Seoul (=1), Gyeonggi (=0) | 1.11<br>0.56 | 1.58<br>0.50 | Statistical Geographic<br>Information Service |
|
|
|
+
|
|
|
+\*Note: *N* = 754. The unit of analysis is haengjungdong.a Variables included only in the additional analyses reported in [Section 4.2.](#page-10-0) due to data availability (*N* = 422).
|
|
|
+
|
|
|
+and Yeouido. By contrast, residential areas are relatively concentrated in the north part of Seoul. Industrial lands are more dispersed than residential or commercial lands, including industrial corridors stretching along highways in the southern portion of the Gyeonggi region. Another large industrial cluster is found in the southwestern part of Seoul. Most freight facilities tend to be located in neighborhoods with a higher concentration of industrial lands. While logistic facilities are relatively scarce, there is a greater number of warehouses, which are also clustered within the industrial district of Seoul.
|
|
|
+
|
|
|
+[Fig. 1](#page-6-0) (a)–(c) illustrates the distribution of road networks. As highways primarily stretch beyond Seoul, the city's employment centers depend on arterial roads. In both Seoul and the Gyeonggi region, there is a noticeable spatial correlation between commercial lands and arterial roads. However, this association is not evident between residential lands and arterial roads. Outside Seoul, industrial neighborhoods are often situated near arterial roads and are surrounded by highway networks. [Fig. 1](#page-6-0) (d) depicts the spatial distribution of truck traffic volume density
|
|
|
+
|
|
|
+throughout the study area. Most truck activity is concentrated in Seoul. Interestingly, the truck traffic volume density is not significantly higher in the city center compared to the rest of Seoul. This might be due to restrictions on heavy truck entry into the city center during specific hours. Outside Seoul, truck activities are more tightly concentrated near arterial roads and warehouses.
|
|
|
+
|
|
|
+[Fig. 2](#page-7-0)(a)–(c) displays the neighborhood-level distribution of crash densities for various types of freight-related crashes, revealing pronounced differences across neighborhoods. Notably, crashes between two freight vehicles are more concentrated outside Seoul compared to the other two types of crashes. These freight vehicle–freight vehicle collisions also occur less frequently in the city center, likely mirroring the patterns of truck traffic volume density. A significant cluster of crashes between freight vehicles and passenger vehicles is evident in the Gangnam area (southern part of Seoul), an area with a dense network of arterial roads—a pattern less pronounced for the other two types of crashes. In general, logistic terminals and warehouses appear to have a
|
|
|
+
|
|
|
+<span id="page-6-0"></span>
|
|
|
+
|
|
|
+**Fig. 1.** Distribution of land use and truck traffic volume density.
|
|
|
+
|
|
|
+positive correlation with freight crashes.
|
|
|
+
|
|
|
+# *3.4. Methodology*
|
|
|
+
|
|
|
+This study utilizes a Bayesian multivariate Conditional Autoregressive (CAR) model to: (1) address the spatial autocorrelations of traffic crashes, a phenomenon widely recognized in the literature (Pljaki´[c et al., 2019; Yuan](#page-15-0) & Wang, 2021), and (2) account for the statistically significant correlations identified between different types of freight crashes and their associated errors, as shown in [Section 4.1](#page-8-0). Specifically, a multivariate model that incorporates a spatial error term within a Bayesian framework is estimated for three response variables: freight vehicle–passenger vehicle crashes, crashes between freight vehicles, and crashes between freight vehicles and non-motorized users. In CAR, "conditional" indicates that each area's value is modeled conditionally upon the values in neighboring areas.
|
|
|
+
|
|
|
+Bayesian models aim to estimate the full probability distribution of parameters. Although the simplicity of point estimates provided by a classical frequentist approach is appealing, the Bayesian approach offers notable benefits. By focusing on the posterior density of each parameter,
|
|
|
+
|
|
|
+# <span id="page-7-0"></span>(a) Freight vehicle–Passenger vehicle crashes (b)Freight vehicle–freight vehicle crashes
|
|
|
+
|
|
|
+
|
|
|
+
|
|
|
+# (c) Freight vehicle–non-motorized crashes
|
|
|
+
|
|
|
+
|
|
|
+
|
|
|
+**Fig. 2.** Neighborhood-level distribution of the density of different types of freight crashes in the study area.
|
|
|
+
|
|
|
+it allows for a more nuanced and direct interpretation of uncertainty. Specifically, a Bayesian credible interval defines a specific range or interval where an unobserved parameter value is likely to fall with a given probability. Due to these benefits, the Bayesian approach has been gaining popularity in recent traffic safety research ([Barajas, 2018, Chen,](#page-15-0) 2015; Ma & [Kockelman, 2006\)](#page-15-0).
|
|
|
+
|
|
|
+Specifically, the model used in this study assumes that *Yik* and *μik*
|
|
|
+
|
|
|
+represent the observed and expected crashes of type *k* in neighborhood *i*, respectively. *Yik* is expected to follow the Poisson distribution, which can be expressed as follows:
|
|
|
+
|
|
|
+$$Y_{ik} \sim Poisson(\mu_{ik}), i = 1, 2, \dots, n$$
|
|
|
+ (4)
|
|
|
+
|
|
|
+$$\log(\mu_{ik}) = \beta_{k0} + \sum_{m=1}^{M} \beta_{km} X_{ikm} + \varphi_i + \varepsilon_{ik}$$
|
|
|
+(5)
|
|
|
+
|
|
|
+<span id="page-8-0"></span>where *βk*0 is the constant term for freight crash type *k* (*k* = 1,2,3); *βkm* is the coefficient of the *m*th covariate for freight crash type *k*; *Xikm* denotes the *m*th covariate for the neighborhood *i* and crash type *k*; *φi*indicates the spatial random effect for neighborhood *i*. Spatial dependence was incorporated into these random effects via *n* by *n* neighborhood matrix **W**, where *wij* refers to whether neighborhood *i* and *j* are spatially close.
|
|
|
+
|
|
|
+$$\varphi_i | \varphi_{(-i),\rho}, \Sigma MVN(\frac{\rho \sum_{j=1}^{n} w_{i,j} \varphi_j}{\rho \sum_{j=1}^{n} w_{i,j} + 1 - \rho}, \frac{\Sigma}{\rho \sum_{j=1}^{n} w_{i,j} + 1 - \rho})$$
|
|
|
+ (6)
|
|
|
+
|
|
|
+That is, *φi*has a CAR distribution, as follows:
|
|
|
+
|
|
|
+where *φ*(− *<sup>i</sup>*) is the full set of random effect vectors except for those related to *i*th neighborhood. *wij* denotes the weight between neighborhood *i* and *j*. Here, *wij* equals 1 if neighborhood *i* and *j* are among the four nearest neighbors based on adjacency, and *wij* equals 0, otherwise. The strength of spatial correlation in the data is controlled by *ρ*, with *ρ* close to one corresponding to strong spatial dependence in the data, and *ρ* equals 0 corresponding to independence in the data. Σ denotes the variance–covariance matrix that shows the covariance/correlation structure of the random effects vectors, which were omitted from the univariate models dealing with a single variance parameter.
|
|
|
+
|
|
|
+To estimate the Bayesian multivariate spatial model, prior distributions for the regression parameters and variance–covariance structure need to be specified, which reflects prior knowledge about the parameters of interest. Due to the lack of sufficient prior knowledge related to the distributions of these parameters, weakly informative prior distributions are employed. This choice allows the data to have a stronger influence on the inference, as opposed to any initial assumptions or beliefs. Specifically, for regression parameters, normal distributions with a mean of 0 and a variance of 1000 are selected as prior distributions. The parameter ρ is given a uniform distribution between 0 and 1. Σ is assumed to follow an Inverse Wishart distribution with 4 degrees of freedom and the identity matrix *I* as the scale matrix. To acquire the posterior estimates of the multivariate and univariate model parameters, the Markov chain Monte Carlo sampling procedure was executed for 30,000 iterations, with the first 10,000 iterations being discarded as burn-in. The trace plots confirmed that the convergence of the parameters had been reached.
|
|
|
+
|
|
|
+This study also compares the fitting performance of the multivariate models with that of the univariate models using a widely recognized criterion for Bayesian models: the Deviance Information Criterion (DIC). The DIC is a comprehensive indicator of model fitting and complexity, which is calculated as:
|
|
|
+
|
|
|
+$$DIC = \bar{D} + pD \tag{6}$$
|
|
|
+
|
|
|
+where D ¯ indicates the posterior mean deviance, which is a measure of how well the model fits the data, and *p*D indicates the effective number of parameters, which measures model complexity. In general, a lower DIC value is indicative of better model performance. Considering the additive characteristic of the DIC ([El-Basyouny](#page-15-0) & Sayed, 2009), the combined DIC values of the univariate spatial models were computed for a direct comparison with the DIC value of the multivariate spatial model.
|
|
|
+
|
|
|
+# **4. Results**
|
|
|
+
|
|
|
+# *4.1. Main results*
|
|
|
+
|
|
|
+To begin with, the correlations between different types of crashes and between their associated errors were examined. Table 3 indicates that crashes between freight vehicles and freight vehicle–passenger vehicle crashes are highly correlated (correlation coefficient: 0.75). The correlation between freight vehicle–non-motorized crashes and the other two types of crashes is relatively moderate, yet statistically significant. Table 3 shows that, even after accounting for measurable variables, correlations of errors across crash types remain. These
|
|
|
+
|
|
|
+**Table 3** Correlations between different types of crashes (a) and correlations of errors across crash types (b).
|
|
|
+
|
|
|
+| (a) | | | | |
|
|
|
+|-------------------------------------------------|-------------------------------------------------|-----------------------------------------------|---------------------------------------------|--|
|
|
|
+| | Freight vehicle<br>passenger vehicle<br>crashes | Freight vehicle<br>freight vehicle<br>crashes | Freight vehicle<br>non-motorized<br>crashes | |
|
|
|
+| Freight vehicle<br>passenger<br>vehicle crashes | 1 | – | – | |
|
|
|
+| Freight vehicle<br>freight vehicle<br>crashes | 0.75** | 1 | – | |
|
|
|
+| Freight vehicle<br>non-motorized<br>crashes | 0.32** | 0.21** | 1 | |
|
|
|
+| (b) | | | | |
|
|
|
+| | Freight vehicle<br>passenger vehicle<br>crashes | Freight vehicle<br>freight vehicle<br>crashes | Freight vehicle<br>non-motorized<br>crashes | |
|
|
|
+| Freight vehicle<br>passenger<br>vehicle crashes | 1 | – | – | |
|
|
|
+| Freight vehicle<br>freight vehicle<br>crashes | 0.27** | 1 | – | |
|
|
|
+| Freight vehicle<br>non-motorized<br>crashes | 0.25** | 0.11** | 1 | |
|
|
|
+
|
|
|
+Notes: \*\* *p <* 0.05.
|
|
|
+
|
|
|
+correlations are not as high (ranging from 0.11 to 0.27) but remain statistically significant. These results therefore justify the application of multivariate CAR models.
|
|
|
+
|
|
|
+To further assess the appropriateness of multivariate models over their univariate counterparts, the DIC was computed for both models, the results of which are provided in the footnotes of [Table 4](#page-9-0) and Appendix C. Given that the DIC value for a multivariate model is considerably lower than that of the univariate models—with a difference exceeding 150—the multivariate model is, again, preferred.
|
|
|
+
|
|
|
+Overall, the results of both multivariate [\(Table 4\)](#page-9-0) and univariate models (Appendix C) appear to be similar in terms of statistical significance and magnitude of coefficients, with a few exceptions. For example, the share of residential land (mean (of the posterior distribution): 0.537, 95% Bayesian Credible Interval (BCI): 0.052, 1.037) is a statistically significant predictor for freight vehicle-non-motorized crashes exclusively in the univariate models, while the squared passenger vehicle traffic volume density (mean: 2.1E-05, 95% BCI: 1.7E-05, 2.3E-05) significantly predicts the freight vehicle–passenger vehicle crashes solely in the multivariate model. This discrepancy can be attributed to the error correlations across crash types considered in the multivariate models, which yield more robust estimations. Therefore, the remainder of this section focuses on the results from the multivariate CAR models [\(Table 4](#page-9-0)), with [Table 5](#page-10-0) illustrating the average marginal effects of each explanatory variable.
|
|
|
+
|
|
|
+Specifically, the Bayesian multivariate CAR model shows that distinct sets of explanatory variables are correlated with different type of freight crashes. Overall, the factors correlated with crashes between two freight vehicles and freight vehicle–non-motorized crashes are largely different, whereas freight vehicle–passenger vehicle crashes occupy a more intermediate position, sharing almost all predictors with one of the other two types of freight crashes.
|
|
|
+
|
|
|
+Focusing on predictors common to all types of freight crashes examined, only two variables—distance to logistics terminals and population density—are consistently significant for every crash type, aside from neighborhood size. Specifically, neighborhoods closer to logistics terminals and those characterized by lower population densities are shown to experience higher frequencies of freight crashes, irrespective
|
|
|
+
|
|
|
+<span id="page-9-0"></span>**Table 4** Bayesian **multivariate** conditional autoregressive model for different types of freight crashes.
|
|
|
+
|
|
|
+| | | | Freight vehicle-passenger vehicle crashes | | Freight vehicle-freight vehicle crashes | | Freight vehicle-non-motorized crashes | | |
|
|
|
+|---------------------------------------------------|---------|---------|-------------------------------------------|---------|-----------------------------------------|---------------------|---------------------------------------|---------|---------------------|
|
|
|
+| | Mean | S.D. | 95% BCI | Mean | S.D. | 95% BCI | Mean | S.D. | 95% BCI |
|
|
|
+| Freight-related facilities | | | | | | | | | |
|
|
|
+| Distance to ICN | − 0.002 | 0.008 | (− 0.019,<br>0.013) | − 0.008 | 0.012 | (− 0.030,<br>0.027) | − 0.010 | 0.007 | (− 0.023,<br>0.005) |
|
|
|
+| Distance to logistics terminal | − 0.053 | 0.015 | (¡0.084,<br>¡0.024) | − 0.058 | 0.021 | (¡0.099,<br>¡0.020) | − 0.041 | 0.020 | (¡0.076,<br>¡0.001) |
|
|
|
+| Warehouse density | 0.055 | 0.031 | (− 0.011,<br>0.110) | 0.073 | 0.042 | (− 0.004,<br>0.154) | 0.105 | 0.027 | (0.051,<br>0.154) |
|
|
|
+| Roadway characteristics | | | | | | | | | |
|
|
|
+| Highway density | 0.414 | 0.074 | (0.272,<br>0.558) | 0.623 | 0.100 | (0.429, 0.846) | 0.122 | 0.093 | (− 0.077,<br>0.308) |
|
|
|
+| Density of roads with 8-or-more<br>lanes | 0.120 | 0.025 | (0.072,<br>0.169) | 0.047 | 0.047 | (− 0.050,<br>0.135) | 0.069 | 0.026 | (0.017,<br>0.120) |
|
|
|
+| Density of roads with 2–7 lanes | 0.019 | 0.012 | (− 0.001,<br>0.052) | 0.023 | 0.023 | (− 0.021,<br>0.069) | 0.030 | 0.011 | (0.009,<br>0.054) |
|
|
|
+| Density of 1-lane roads | − 0.002 | 0.004 | (− 0.009,<br>0.004) | 0.004 | 0.009 | (− 0.012,<br>0.024) | 0.016 | 0.004 | (0.009,<br>0.023) |
|
|
|
+| Road intersection density | 0.006 | 0.005 | (− 0.005,<br>0.015) | − 0.030 | 0.012 | (¡0.051,<br>¡0.006) | 0.007 | 0.005 | (− 0.003,<br>0.018) |
|
|
|
+| Closeness centrality | − 6948 | 7583.2 | (− 21917,<br>7679) | − 25841 | 11612.9 | (¡48588,<br>¡5553) | − 4484.3 | 8679.1 | (− 21991,<br>11275) |
|
|
|
+| Meshedness coefficient | 0.112 | 0.191 | (− 0.270,<br>0.478) | − 0.091 | 0.421 | (− 0.910,<br>0.753) | − 0.404 | 0.281 | (− 1.032,<br>0.136) |
|
|
|
+| Built environment | | | | | | | | | |
|
|
|
+| Employment density | 0.002 | 0.001 | (0.001,<br>0.004) | 0.002 | 0.001 | (− 0.001,<br>0.005) | 0.003 | 0.001 | (0.002,<br>0.004) |
|
|
|
+| Population density | − 0.012 | 0.002 | (¡0.015,<br>¡0.009) | − 0.016 | 0.005 | (¡0.025,<br>¡0.007) | − 0.006 | 0.002 | (¡0.009,<br>¡0.002) |
|
|
|
+| Land use mix | 0.205 | 0.214 | (− 0.155,<br>0.627) | 0.429 | 0.377 | (− 0.421,<br>1.112) | 0.861 | 0.234 | (0.386,<br>1.338) |
|
|
|
+| Residential land share | 0.046 | 0.238 | (− 0.376,<br>0.473) | − 0.420 | 0.464 | (− 1.332,<br>0.375) | 0.400 | 0.250 | (− 0.095,<br>0.829) |
|
|
|
+| Large grocery store density | − 0.008 | 0.021 | (− 0.050,<br>0.032) | − 0.035 | 0.044 | (− 0.120,<br>0.050) | 0.024 | 0.020 | (− 0.018,<br>0.062) |
|
|
|
+| Large shopping mall density | 0.016 | 0.019 | (− 0.019,<br>0.054) | 0.010 | 0.040 | (− 0.075,<br>0.088) | 0.004 | 0.018 | (− 0.035,<br>0.039) |
|
|
|
+| Elementary school density | − 0.034 | 0.013 | (¡0.062,<br>¡0.009) | − 0.077 | 0.033 | (¡0.135,<br>¡0.006) | − 0.018 | 0.014 | (− 0.046,<br>0.009) |
|
|
|
+| Hospital density | 0.006 | 0.032 | (− 0.057,<br>0.062) | 0.024 | 0.081 | (− 0.137,<br>0.175) | 0.046 | 0.035 | (− 0.022,<br>0.112) |
|
|
|
+| Exposure variable | | | | | | | | | |
|
|
|
+| Truck traffic volume density | 0.048 | 0.016 | (0.021,<br>0.084) | 0.008 | 0.026 | (− 0.043,<br>0.058) | 0.026 | 0.020 | (− 0.012,<br>0.069) |
|
|
|
+| Passenger vehicle traffic volume<br>density | − 0.011 | 0.016 | (¡0.014,<br>¡0.007) | − 0.003 | 0.005 | (− 0.011,<br>0.004) | − 0.017 | 0.003 | (¡0.023,<br>¡0.011) |
|
|
|
+| Passenger vehicle traffic volume | 2.1E | 1.5E-05 | (1.7E-05, | 1.3E-05 | 3.3E-05 | (− 5.6E-05, | 5.0E-05 | 1.3E-05 | (2.4E-05, |
|
|
|
+| density (squared)<br>Non-motorized infrastructure | 05 | | 2.3E-05) | | | 6.7E-05) | | | 7.5E-05) |
|
|
|
+| Sidewalk density | 0.017 | 0.006 | (0.006,<br>0.028) | 0.031 | 0.012 | (0.009, 0.059) | − 0.005 | 0.007 | (− 0.020,<br>0.010) |
|
|
|
+| Sidewalk connectivity | − 0.257 | 0.818 | (− 1.767,<br>1.141) | 1.043 | 1.307 | (− 1.549,<br>3.488) | − 0.156 | 0.933 | (− 2.083,<br>1.844) |
|
|
|
+| Other variables | | | | | | | | | |
|
|
|
+| Square mile | 0.178 | 0.019 | (0.139,<br>0.216) | 0.245 | 0.027 | (0.188, 0.298) | 0.103 | 0.026 | (0.053,<br>0.158) |
|
|
|
+| Seoul (yes = 1, no = 0) | 0.114 | 0.089 | (− 0.044,<br>0.280) | 0.173 | 0.146 | (− 0.084,<br>0.466) | 0.388 | 0.116 | (0.128,<br>0.590) |
|
|
|
+| Intercept | 2.156 | 0.473 | (1.398,<br>3.206) | 0.696 | 0.662 | (− 0.477,<br>1.955) | 0.781 | 0.524 | (¡0.166,<br>1.811) |
|
|
|
+
|
|
|
+Notes: S.D.: Standard deviation, BCI.: Bayesian credible interval; *N* = 754; DIC = 8129.844. BCI values in bold were statistically significant at the 0.05 level.
|
|
|
+
|
|
|
+of crash type. A one-mile increase in distance to logistics terminals corresponds to an average decrease in the frequencies of freight vehicle–passenger vehicle crashes, freight vehicle–freight vehicle crashes, and freight vehicle–non-motorized crashes by 0.412, 0.064, and 0.103, respectively [\(Table 5\)](#page-10-0). Furthermore, the respective average marginal effects of population density for freight vehicle–passenger vehicle crashes, freight vehicle–freight vehicle crashes, and freight vehicle–nonmotorized crashes are − 0.095, − 0.018, and − 0.016.
|
|
|
+
|
|
|
+Three predictors are exclusively shared between freight vehicle–passenger vehicle and freight vehicle–freight vehicle crashes. Highway density positively correlates with the occurrence of both the
|
|
|
+
|
|
|
+former (mean: 0.414, 95% BCI: 0.272, 0.558) and the latter (mean: 0.623, 95% BCI: 0.429, 0.846); however, it does not significantly predict freight vehicle–non-motorized crashes. A one-unit increase in highway density, on average, leads to approximately four more freight vehicle–passenger vehicle crashes (marginal effect: 4.113) and one additional freight vehicle–freight vehicle crash (marginal effect: 0.979). Sidewalk density is also significantly and positively associated with freight vehicle–passenger vehicle crashes (mean: 0.017, 95% BCI: 0.006, 0.028) and freight vehicle–freight vehicle crashes (mean: 0.031, 95% BCI: 0.009, 0.059). However, neighborhoods with higher elementary school density see fewer occurrences of these two types of crashes. For
|
|
|
+
|
|
|
+<span id="page-10-0"></span>**Table 5** Estimation results of marginal effects (based on the results reported in [Table 4\)](#page-9-0).
|
|
|
+
|
|
|
+| | Freight vehicle<br>passenger vehicle<br>crashes | Freight vehicle<br>freight vehicle<br>crashes | Freight vehicle<br>non-motorized<br>crashes |
|
|
|
+|---------------------------------------------------------------------------|-------------------------------------------------|-----------------------------------------------|---------------------------------------------|
|
|
|
+| | M.E. | M.E. | M.E. |
|
|
|
+| Freight-related | | | |
|
|
|
+| facilities | | | |
|
|
|
+| Distance to ICN | − 0.016 | − 0.009 | − 0.025 |
|
|
|
+| Distance to logistics<br>terminal | ¡0.412 | ¡0.064 | ¡0.103 |
|
|
|
+| Warehouse density | 0.451 | 0.086 | 0.285 |
|
|
|
+| Roadway<br>characteristics | | | |
|
|
|
+| Highway density | 4.113 | 0.979 | 0.331 |
|
|
|
+| Density of roads with<br>8-or-more lanes | 1.019 | 0.055 | 0.182 |
|
|
|
+| Density of roads with<br>2–7 lanes | 0.156 | 0.027 | 0.077 |
|
|
|
+| Density of 1-lane<br>roads | − 0.019 | 0.005 | 0.043 |
|
|
|
+| Road intersection<br>density | 0.047 | ¡0.034 | 0.018 |
|
|
|
+| Closeness centrality | − 8.011 | ¡1.134 | − 2.6 |
|
|
|
+| Meshedness<br>coefficient | 0.952 | − 0.099 | − 0.851 |
|
|
|
+| Built environment | | | |
|
|
|
+| Employment density | 0.020 | 0.002 | 0.008 |
|
|
|
+| Population density | ¡0.095 | ¡0.018 | ¡0.016 |
|
|
|
+| Land use mix | 1.820 | 0.607 | 3.496 |
|
|
|
+| Residential land share | 0.375 | − 0.389 | 1.261 |
|
|
|
+| Large grocery store<br>density | − 0.064 | − 0.039 | 0.062 |
|
|
|
+| Large shopping mall<br>density | 0.131 | 0.012 | 0.011 |
|
|
|
+| Elementary school<br>density | ¡0.267 | ¡0.084 | − 0.045 |
|
|
|
+| Hospital density | 0.047 | 0.028 | 0.120 |
|
|
|
+| Exposure variable | | | |
|
|
|
+| Truck traffic volume<br>density | 0.397 | 0.009 | 0.068 |
|
|
|
+| Passenger vehicle<br>traffic volume<br>density | ¡0.089 | − 0.004 | ¡0.044 |
|
|
|
+| Passenger vehicle<br>traffic volume<br>density (squared)<br>Non-motorized | 5.62E-05 | 1.47E-05 | 1.28E-04 |
|
|
|
+| infrastructure | | | |
|
|
|
+| Sidewalk density | 0.136 | 0.036 | − 0.013 |
|
|
|
+| Sidewalk connectivity | − 1.812 | 2.082 | − 0.369 |
|
|
|
+| Other variables | | | |
|
|
|
+| Square mile | 1.565 | 0.315 | 0.277 |
|
|
|
+| Seoul (yes = 1, no =<br>0) | 0.963 | 0.214 | 1.214 |
|
|
|
+| Intercept | 61.204 | 1.139 | 3.028 |
|
|
|
+
|
|
|
+Notes: M.E.: marginal effects; M.E. values presented in bold are statistically significant at the 0.05 level.
|
|
|
+
|
|
|
+every one unit rise in elementary school density, the frequencies of freight vehicle–passenger vehicle crashes and crashes between freight vehicles decline by 0.267 and 0.084, respectively.
|
|
|
+
|
|
|
+Freight vehicle–passenger vehicle crashes and freight vehicle–nonmotorized crashes also share several predictors exclusively. Areas with higher employment density and a greater concentration of roads with eight or more lanes are more likely to witness these two types of crashes. For every increase of one thousand job per square mile in neighborhoods, the frequency of freight vehicle–passenger vehicle and freight vehicle–non-motorized crashes rises by 0.020 and 0.008, respectively. Similarly, each unit increase in the density of roads with eight or more lanes corresponds to an average increase of 1.019 and 0.182 for the same crash types. Notably, passenger vehicle traffic volume density-—including the original variable and its squared term—also serves as a significant predictor of freight vehicle–passenger vehicle crashes and
|
|
|
+
|
|
|
+freight vehicle–non-motorized crashes. According to the signs of these variables, the frequencies of both crash types present a u-curve relationship with passenger vehicle traffic volume density.
|
|
|
+
|
|
|
+Lastly, a set of variables emerge as significant predictors exclusive to a single type of crashes. Specifically, the warehouse density (mean: 0.105, 95% BCI: 0.051, 0.154), densities of roads with 2–7 lanes (mean: 0.030, 95% BCI: 0.009, 0.054) and 1 lane (mean: 0.016, 95% BCI: 0.009, 0.023), as well as mixed land use (mean: 0.861, 95% BCI: 0.386, 1.338) are each significantly associated with the frequency of freight vehicle–non-motorized crashes only. Moreover, after adjusting for other factors, the difference in crash frequency between Seoul and Gyeonggi province remains significant only for freight vehicle–non-motorized crashes. Regarding crashes between two freight vehicles, some variables related to the road network structure, such as intersection density (mean: − 0.030, 95% BCI: − 0.051, − 0.006) and closeness centrality (mean: − 25841, 95% BCI: − 48588, − 5553), are found to be significant predictors; both of these variables exhibit negative average marginal effects (− 0.034 for intersection density, − 1.134 for closeness centrality), implying that their increase is associated with a decline in the crash frequency. Finally, truck traffic volume density (mean: 0.048, 95% BCI: 0.021, 0.084) significantly predicts freight vehicle-passenger vehicle crashes only, with each one-unit increase in truck traffic volume density resulting in a 0.397 increase in their frequency.
|
|
|
+
|
|
|
+## *4.2. Additional analyses: The roles of non-motorized infrastructure and bicycle traffic volume*
|
|
|
+
|
|
|
+This section presents the results of additional analyses, with a focus on the city of Seoul. As previously noted, this focus is due to constraints posed by limited data availability; variables related to some nonmotorized infrastructure and bicycle traffic volume density were added to the base models after restricting samples to neighborhoods within the city of Seoul.
|
|
|
+
|
|
|
+[Table 6](#page-11-0) first outlines the estimation results related to freight vehicle–non-motorized crashes, before introducing new variables. This approach allows for determining whether any observed changes in the results of other explanatory variables are due to the incorporation of additional variables, or simply the result of a reduced sample size. The results, including those related to other two types of freight crashes (not reported to save space), suggest that despite the smaller sample size, all explanatory variables maintain the same signs as presented in [Table 4](#page-9-0). However, a few variables lose their statistical significance, likely due to reduced statistical power. Moreover, [Table 6](#page-11-0) illustrates that, regardless of whether variables related to non-motorized transportation infrastructure and bicycle traffic volume are included, the statistical significance and signs of other explanatory variables remain unaffected.
|
|
|
+
|
|
|
+As expected, the analysis reveals a positive correlation between bicycle traffic volume density and crashes between freight vehicles and non-motorized vehicles (mean: 0.009, 95% BCI: 0.002, 0.017). As indicated in the marginal effect column, each unit increase in bicycle traffic volume density (one thousand counts per square mile) corresponds with an additional 0.027 freight vehicle–non-motorized crash. By contrast, the impact of non-motorized facilities on this type of crash is observed to be limited. The only non-motorized facility significantly associated with freight vehicle–non-motorized collisions is the density of shared bike-pedestrian paths (mean: − 0.006, 95% BCI: − 0.011, − 0.001). Specifically, a unit increase in the density of these shared paths results in a reduction of freight vehicle–non-motorized collisions by 0.018. However, other types of on-road bicycle lanes and crossing facilities appear to be insignificant predictors of freight crashes involving non-motorized users.
|
|
|
+
|
|
|
+### **5. Discussion and policy implications**
|
|
|
+
|
|
|
+This section discusses the main results grouped by variable category and provides the associated policy implications.
|
|
|
+
|
|
|
+<span id="page-11-0"></span>**Table 6** Bayesian multivariate conditional autoregressive model: focusing on Seoul.
|
|
|
+
|
|
|
+| | Freight vehicle-non-motorized crashes without additional<br>variables | | | | | Freight vehicle-non-motorized crashes with additional<br>variables | | | | |
|
|
|
+|----------------------------------------------------|-----------------------------------------------------------------------|---------|------------------|---------|---------|--------------------------------------------------------------------|--------------------|---------|--|--|
|
|
|
+| | Mean | S.D. | 95% BCI | M.E. | Mean | S.D. | 95% BCI | M.E. | | |
|
|
|
+| Freight-related facilities | | | | | | | | | | |
|
|
|
+| Distance to ICN | 0.005 | 0.012 | (− 0.021, 0.028) | 0.014 | 0.001 | 0.010 | (− 0.018, 0.021) | 0.003 | | |
|
|
|
+| Distance to logistics terminal | − 0.060 | 0.032 | (¡0.125, ¡0.001) | ¡0.170 | − 0.067 | 0.028 | (¡0.124, ¡0.016) | ¡0.192 | | |
|
|
|
+| Warehouse density | 0.040 | 0.037 | (− 0.033, 0.111) | 0.119 | 0.052 | 0.037 | (− 0.024, 0.124) | 0.155 | | |
|
|
|
+| Roadway characteristics | | | | | | | | | | |
|
|
|
+| Highway density | 0.058 | 0.158 | (− 0.262, 0.362) | 0.176 | 0.100 | 0.151 | (− 0.191, 0.393) | 0.310 | | |
|
|
|
+| Density of roads with 8-or-more lanes | 0.059 | 0.037 | (− 0.017, 0.132) | 0.178 | 0.054 | 0.034 | (− 0.009, 0.122) | 0.164 | | |
|
|
|
+| Density of roads with 2–7 lanes | 0.035 | 0.014 | (0.007, 0.061) | 0.104 | 0.026 | 0.014 | (0.0003, 0.052) | 0.078 | | |
|
|
|
+| Density of 1-lane roads | 0.015 | 0.004 | (0.006, 0.024) | 0.044 | 0.015 | 0.004 | (0.006, 0.023) | 0.043 | | |
|
|
|
+| Road intersection density | 0.009 | 0.009 | (− 0.009, 0.025) | 0.026 | 0.007 | 0.009 | (− 0.01, 0.024) | 0.022 | | |
|
|
|
+| Closeness centrality | − 16486 | 16,227 | (− 46576, 16728) | − 2.936 | − 18942 | 14,209 | (− 47820, 9261) | − 2.936 | | |
|
|
|
+| Meshedness coefficient | − 0.263 | 0.366 | (− 0.998, 0.405) | − 0.678 | − 0.234 | 0.344 | (− 0.901, 0.488) | − 0.612 | | |
|
|
|
+| Built environment | | | | | | | | | | |
|
|
|
+| Employment density | 0.003 | 0.001 | (0.002, 0.005) | 0.010 | 0.003 | 0.001 | (0.001, 0.004) | 0.008 | | |
|
|
|
+| Population density | − 0.005 | 0.002 | (¡0.009, ¡0.001) | ¡0.015 | − 0.005 | 0.002 | (¡0.009, ¡4.0E-04) | ¡0.014 | | |
|
|
|
+| Land use mix | 1.123 | 0.321 | (0.531, 1.778) | 6.092 | 1.093 | 0.335 | (0.440, 1.707) | 5.819 | | |
|
|
|
+| Residential land share | 0.393 | 0.291 | (− 0.151, 1) | 1.414 | 0.482 | 0.295 | (− 0.099, 1.088) | 1.816 | | |
|
|
|
+| Large grocery store density | 0.020 | 0.022 | (− 0.024, 0.062) | 0.059 | 0.015 | 0.021 | (− 0.026, 0.055) | 0.045 | | |
|
|
|
+| Large shopping mall density | − 0.006 | 0.024 | (− 0.055, 0.041) | − 0.019 | − 0.008 | 0.023 | (− 0.05, 0.035) | − 0.024 | | |
|
|
|
+| Elementary school density | − 0.012 | 0.016 | (− 0.045, 0.019) | − 0.034 | − 0.010 | 0.016 | (− 0.04, 0.020) | − 0.029 | | |
|
|
|
+| Hospital density | 0.056 | 0.039 | (− 0.022, 0.13) | 0.168 | 0.052 | 0.037 | (− 0.022, 0.122) | 0.155 | | |
|
|
|
+| Exposure variable | | | | | | | | | | |
|
|
|
+| Truck traffic volume density | 0.033 | 0.030 | (− 0.028, 0.09) | 0.097 | 0.028 | 0.028 | (− 0.029, 0.071) | 0.082 | | |
|
|
|
+| Passenger vehicle traffic volume density | − 0.012 | 0.005 | (¡0.021, ¡0.002) | ¡0.034 | − 0.015 | 0.005 | (¡0.026, ¡0.005) | ¡0.043 | | |
|
|
|
+| Passenger vehicle traffic volume density (squared) | 1.8E-05 | 2.0E-05 | (− 2.2E-05, | 5.3 | 3.4 | 2.1 | (− 1.7E-05, | 1.0 | | |
|
|
|
+| | | | 5.5E-05) | E-05 | E-05 | E-05 | 7.1E-05) | E-04 | | |
|
|
|
+| Bicycle traffic volume density* | | | | | 0.009 | 0.004 | (0.002, 0.017) | 0.027 | | |
|
|
|
+| Non-motorized infrastructure | | | | | | | | | | |
|
|
|
+| Sidewalk density | − 0.002 | 0.011 | (− 0.023, 0.02) | − 0.005 | − 0.012 | 0.015 | (− 0.042, 0.014) | − 0.035 | | |
|
|
|
+| Sidewalk connectivity | − 0.735 | 1.162 | (− 3.062, 1.498) | − 1.528 | − 0.241 | 1.228 | (− 2.453, 2.243) | − 0.628 | | |
|
|
|
+| Bicycle-priority lane density* | | | | | 0.001 | 0.022 | (− 0.043, 0.041) | 0.002 | | |
|
|
|
+| Exclusive bicycle lane density* | | | | | 0.043 | 0.055 | (− 0.065, 0.153) | 0.129 | | |
|
|
|
+| Protected bicycle lane density* | | | | | 0.002 | 0.038 | (− 0.076, 0.079) | 0.006 | | |
|
|
|
+| Shared bike-pedestrian path density* | | | | | − 0.006 | 0.003 | (¡0.011, ¡0.001) | ¡0.018 | | |
|
|
|
+| Crossing facility density* | | | | | 0.001 | 0.001 | (− 3.0E-04, 0.002) | 0.002 | | |
|
|
|
+| Other variables | | | | | | | | | | |
|
|
|
+| Square mile | 0.235 | 0.100 | (0.032, 0.428) | 0.777 | 0.202 | 0.100 | (0.021, 0.407) | 0.656 | | |
|
|
|
+| Intercept | 0.831 | 0.943 | (− 1.101, 2.579) | 3.806 | 1.042 | 0.815 | (− 0.507, 2.641) | 5.383 | | |
|
|
|
+| DIC | 4363.64 | | | | 4373.96 | | | | | |
|
|
|
+
|
|
|
+Notes: *N* = 422; M.E.: marginal effects, S.D.: standard deviation, BCI: Bayesian credible interval; Credible intervals and marginal effects presented in bold are statistically significant at the 0.05 level; \* variables newly added.
|
|
|
+
|
|
|
+### *5.1. Freight-related facilities*
|
|
|
+
|
|
|
+The results from the Bayesian multivariate CAR model reveal the distinct roles that different types of freight-related facilities play in shaping the spatial distribution of crashes involving freight vehicles. Notably, distance to logistics terminals shows a consistently negative association with all types of freight crashes examined, whereas warehouse density is positively correlated with freight vehicle–non-motorized crashes only. Potential explanations for these different effects could be attributed to the inherent differences in operational and locational characteristics between these two types of facilities. Logistics terminals typically manage a broader spectrum of operations, which may include transportation, warehousing, inventory management, and packaging. However, warehouses are primarily concerned with storage ([Heitz,](#page-15-0) [et al., 2019\)](#page-15-0). Consequently, logistics terminals may attract a broader range of vehicle activities, thereby increasing the probability of collisions between freight vehicles or crashes involving freight vehicles and passenger vehicles. Meanwhile, distance to the international airport appears to be an insignificant predictor for any type of freight crash. This may be due to its location, which is outside the study area, and therefore wields less influence. Overall, these findings align with the results reported by [Yang et al. \(2021a\)](#page-16-0) from the Los Angeles region despite considerable differences in urban spatial structures between Los Angeles ([Schuetz et al., 2018a,b; Giuliano et al., 2022\)](#page-15-0) and the study area.
|
|
|
+
|
|
|
+These results suggest that it is crucial to consider the potential road safety implications of proposed developments of freight facilities, such as logistics terminals and warehouses, during the planning stages. This can be achieved, for instance, through comprehensive traffic impact analysis. Moreover, policy measures need to be implemented to more effectively regulate and manage freight-related activities near existing logistics facilities. Potential strategies could include banning certain operations during peak traffic hours or separating truck flow from the main traffic flow. Furthermore, it is imperative for policymakers to introduce measures that enhance safety for non-motorized users in the vicinity of logistic facilities and warehouses, with initiatives such as improved signage and lighting.
|
|
|
+
|
|
|
+### *5.2. Roadway network infrastructure*
|
|
|
+
|
|
|
+Crashes between freight vehicles, as well as collisions involving both freight vehicles and passenger vehicles, are more likely to occur in areas characterized by a higher concentration of highways. Conversely, crashes between freight vehicles and non-motorized users are more prevalent in neighborhoods with a denser network of non-highway roads, including narrower ones. This pattern aligns with prior findings on crashes between passenger vehicles and non-motorized users or total non-motorized crashes [\(Barajas, 2018; Guerra et al., 2019; Shin, 2023](#page-15-0)). Consequently, it could be inferred that crashes involving non-motorized users, regardless of whether they involve freight vehicles or not, are less frequent on highways, likely due to the limited access points available for them. The positive effects of highway density on crashes between freight vehicles and freight vehicle-passenger vehicle crashes are also in line with the patterns of total freight crashes found in the Los Angeles region, where most freight crashes are between a freight vehicle and passenger vehicle ([Yang et al., 2021a](#page-16-0)).
|
|
|
+
|
|
|
+Road network structure variables appear to be correlated only with crashes between freight vehicles. Road intersection density is a negative predictor for these crashes. This result is consistent with previous research examining the role of street network design in the distribution of total traffic crashes (Marshall & [Garrick, 2011](#page-15-0)), as well as nonmotorized crashes [\(Zhang et al., 2012](#page-16-0)). Similarly, closeness centrality, a measure of street accessibility, negatively correlates with crashes between freight vehicles. The meshedness coefficient, however, does not significantly predict any freight crashes, possibly reflecting the ambiguous and contradictory role of street connectivity in traffic crash predictions. While greater connectivity may raise collision risk due to increased conflicts and complexity (Marshall & [Garrick, 2011\)](#page-15-0), it may also reduce crash risks by improving road performance and lowering traffic congestion levels (Choi & [Ewing, 2021\)](#page-15-0).
|
|
|
+
|
|
|
+Results suggest that improvements in highway safety for freight vehicles, such as providing more rest areas for truck drivers along highways [\(Bunn et al., 2019\)](#page-15-0), are necessary. Safety measures should also be established to protect non-motorized users from freight traffic on nonhighway roads, including narrower ones. For example, local media reports suggest that illegally parked cars on smaller roads can compromise road safety by obstructing drivers' visibility, which can significantly increase the frequency of freight crashes involving non-motorized users (e.g., [Jung, 2016](#page-15-0)). Therefore, the strong enforcement of parking regulations needs to be prioritized along with the development of designated parking zones in appropriate locations. The results related to road network patterns further suggest that policymakers should prioritize designing street networks that enhance street accessibility to reduce crashes between freight vehicles.
|
|
|
+
|
|
|
+# *5.3. Built environment characteristics*
|
|
|
+
|
|
|
+Among the built environment characteristics, population density stands out as the only factor that is statistically significant across all examined freight crash types. Specifically, neighborhoods with higher population density tend to witness fewer freight crashes, regardless of specific crash type. Conversely, areas with higher employment density tend to experience more collisions between a freight vehicle and both passenger vehicles and non-motorized users. This pattern aligns closely with [Yuan and Wang](#page-16-0)'s (2021) analysis of injured freight-related crashes. They postulated that the differing impacts of employment and population densities on freight crashes could arise from a combination of heightened conflicts and increased truck driver caution in denser areas. However, it is worth noting that the identified negative correlation between population density and freight crashes involving non-motorized users diverges from the commonly observed positive relationship between population density and passenger vehicles–non-motorized crashes (Dumbaugh & [Li, 2010; Loukaitou-Sideris et al., 2007](#page-15-0)).
|
|
|
+
|
|
|
+Mixed land use is significantly and positively correlated only with crashes between freight vehicles and non-motorized users. Among the POI variables studied—including those related to retail—none demonstrably predict any kind of freight crash, with the exception of elementary school density. This observation aligns with the findings of [Jia et al.](#page-15-0) [\(2018\),](#page-15-0) who also identified an insignificant correlation between commercial-related POI variables and overall traffic crashes in Suzhou, China. Interestingly, the impacts of elementary school density on freight crashes are either negative or insignificant, depending on the crash type, which contrasts with the prevailing narratives in local media [\(Choi,](#page-15-0) [2020; Kim](#page-15-0) & Park, 2021).
|
|
|
+
|
|
|
+These results suggest the potential need for implementing safety
|
|
|
+
|
|
|
+measures in employment centers to reduce freight crashes. While strategies such as restriction of heavy freight vehicle access during a certain time of day are currently only implemented in city centers, similar tactics could be considered for other employment subcenters, provided they do not greatly compromise the efficiency of freight deliveries. Furthermore, the findings hint that existing policies in the study area, such as car-free streets in commercial areas and school zones, may somewhat mitigate the risk of certain types of freight crashes because none of related variables are positively associated with crashes. However, further in-depth investigation is needed to assess the effectiveness of these policies.
|
|
|
+
|
|
|
+## *5.4. Crash risk exposure*
|
|
|
+
|
|
|
+The effect for truck traffic volume density, as expected, is positive for all three types of freight crashes, but it is statistically significant only for freight vehicle–passenger vehicle crashes. Although [Yang et al., \(2021a\)](#page-16-0) also reported the relatively marginal impacts of truck volume density on total freight crashes, the definition of trucks utilized in the present study could further contribute to these results. Despite the focus being on commercial freight vehicles in this study, the truck traffic volume data does not differentiate between commercial and non-commercial trucks, nor does it distinguish between truck weights. This might impact the results, as a significant proportion of commercial trucks are medium or heavy, while most non-commercial trucks are light trucks. Given that light trucks likely exhibit spatial patterns more similar to passenger vehicles than to medium or heavy trucks, truck traffic volume density may more significantly predict freight crashes involving passenger vehicles than other crash types.
|
|
|
+
|
|
|
+Another notable finding is that passenger vehicle traffic volume density has a non-linear relationship with freight vehicle–passenger vehicle and freight vehicle–non-motorized crashes. The signs of the related variable indicate that these two types of crashes decrease with an increase in passenger vehicle traffic volume, but only up to a certain point. After this threshold, the frequency of crashes increases along with passenger vehicle traffic volume density, corroborating some earlier findings (e.g., [Martin, 2002; Liu et al., 2020](#page-15-0)). One potential explanation for this relationship is that drivers or non-motorized users have less space and time to react to sudden changes in highly congested areas, leading to more frequent collisions.
|
|
|
+
|
|
|
+Additional analyses focused on Seoul further confirm the positive association between bicycle traffic volume density and freight crashes involving non-motorized users, echoing earlier findings related to nonmotorized crashes in general (e.g., [Ding et al., 2021; Ding et al., 2020](#page-15-0)).
|
|
|
+
|
|
|
+The results related to crash risk exposure provide important lessons for policymakers, including a need for better data collection for traffic volume that distinguish between different types of trucks to improve crash prediction and prevention efforts. Additionally, results suggest that managing the volume of traffic and congestion could be crucial for preventing certain types of freight crashes. It may be helpful to monitor areas with high passenger vehicle traffic and implement measures such as re-routing to reduce peak traffic volume and maintain optimal traffic volume density. Additionally, neighborhoods with higher bicycle traffic volume density need to be prioritized for measures to enhance bicyclist safety.
|
|
|
+
|
|
|
+### *5.5. Non-motorized infrastructure*
|
|
|
+
|
|
|
+Results show that sidewalk density is positively correlated with both freight vehicle–freight vehicle and freight vehicle–passenger vehicle crashes. Given that sidewalk connectivity was controlled for in the analysis, the positive association between sidewalk density and freight crashes likely stems from illegal parking on sidewalks, rather than sidewalks intersecting with driveways or crossings. Local media reports corroborate this, suggesting that the absence of designated parking garages for freight vehicles leads to illegal parking on sidewalks. This can <span id="page-13-0"></span>obstruct sightlines, potentially raising the likelihood of crashes between motor vehicles on sidewalks ([Kang, 2021;](#page-15-0) Shin, D.M., 2023). However, sidewalk density does not significantly increase the frequency of freight vehicle–non-motorized crashes, maybe because sidewalks are primarily designed for pedestrian safety.
|
|
|
+
|
|
|
+Similarly, despite the functional goals of crossing facilities to enhance pedestrian safety, their density does not significantly predict crashes between freight vehicles and non-motorized users. This could be attributed to the large blind spot on the right side of freight vehicles when making a right turn ([Kim, 2022\)](#page-15-0).
|
|
|
+
|
|
|
+The results also demonstrate that among the various types of bicycle lanes, only the shared bicycle-pedestrian paths show a negative correlation with freight vehicle–non-motorized crashes. This suggests that other forms of on-road bicycle lanes might be susceptible to encroachment by vehicles, a concern spotlighted by local media sources [\(Park](#page-15-0) & [Kim, 2020\)](#page-15-0), which in turn compromises their effectiveness in enhancing road safety.
|
|
|
+
|
|
|
+Therefore, these results suggest that shared bicyclist–pedestrian paths could potentially improve road safety, particularly in areas where freight–non-motorized crashes are more prevalent, such as those near freight facilities and neighborhoods with mixed land use. Moreover, enforcing regulations that prohibit the illegal parking of freight vehicles on sidewalks may help reduce the occurrence of crashes between freight vehicles, as well as freight vehicle–passenger vehicle crashes. This approach may be particularly effective near logistics terminals and employment centers, again when complemented by designated parking garages or zones for freight vehicles in suitable locations.
|
|
|
+
|
|
|
+### **6. Conclusion**
|
|
|
+
|
|
|
+This study valuably contributes to the literature by being the first to explore and compare the spatial patterns of different types of freight crashes and the neighborhood-level factors associated with them. Given the rising trends of crashes involving freight vehicles worldwide, the methodologies and findings from the present study can serve as a foundation for future investigations. The primary takeaways are fourfold.
|
|
|
+
|
|
|
+First, the results highlight the importance of area-level determinants of freight-related crashes. It becomes evident that the incidence of freight crashes is influenced not just by road network and general built environment characteristics, but also by the distribution of freightrelated facilities, such as logistic facilities and warehouses, at the neighborhood level. Therefore, this study adds to prior research that primarily focused on micro-level factors associated with freight crashes. While the relationships between variables and freight crashes observed in this study generally align with those reported for total traffic or nonfreight crashes in previous research, there are noticeable discrepancies. For instance, unlike most studies showing a positive correlation between population density and traffic crashes, this research consistently revealed a negative correlation across all freight-related crashes, even for freight vehicle–non-motorized collisions.
|
|
|
+
|
|
|
+Second, the analysis reveals considerable variation in the factors associated with different types of freight crashes. Only three factors—distance to logistics terminals, population density, and neighborhood size—significantly predict all three types of freight crashes investigated. Notably, the predictors of freight vehicle–freight vehicle crashes and freight vehicle–non-motorized crashes exhibit significant divergence. The built environment and freight facility-related factors (e.g., land use mix and warehouse density) more heavily influence crashes between freight vehicles and non-motorized users, whereas crashes between freight vehicles are more influenced by road network structure variables. Several variables, including land use mix and density of one-lane roads, are also found to be exclusively associated with freight
|
|
|
+
|
|
|
+vehicle–non-motorized crashes. These results emphasize the need to differentiate between different types of freight crashes, which can aid policymakers in directing resources more effectively to areas at higher risk for certain freight crash types.
|
|
|
+
|
|
|
+Third, despite the noticeable differences in spatial structure and examined freight crashes, findings related to crashes between freight vehicles and passenger vehicles align significantly with existing research from Los Angeles [\(Yang et al., 2021a; Yang et al., 2021b\)](#page-16-0). Los Angeles is known for an automobile-oriented urban spatial structure, in contrast to Seoul's transit- and non-motorized user-oriented layout. In both Seoul and Los Angeles, freight crashes are found to be significantly correlated with highways and freight-related facilities other than international airports. Additionally, a consistent negative correlation between population density and freight crashes exists in both regions ([Yang et al.,](#page-16-0) [2021b\)](#page-16-0). Yang (2021a) revealed a negligible association between the proportion of retail use and freight crashes, which aligns with findings from the current study. These parallels hint at the potential for generalizing some outcomes regardless of the local context. However, reinforcing this perspective requires further research based on other regions, as emphasized below.
|
|
|
+
|
|
|
+Lastly, this study underscores the usefulness of a multivariate analytical framework when examining the multiple types of traffic crashes with interconnected errors. In addition to the better performance of a multivariate modeling technique, subtle yet significant discrepancies between univariate and multivariate models were identified. Therefore, these highlight the importance of multivariate analyses of different types of crashes.
|
|
|
+
|
|
|
+Despite the valuable contributions of the present study, limitations exist, such as a lack of detailed classification for freight vehicles in both the traffic crash and truck traffic volume data. Additionally, this study did not include a variable related to pedestrian traffic volume, a shortfall attributed to the absence of comparable data. Therefore, future research could benefit from a more nuanced examination of freight-related crashes, considering freight vehicle subtypes (by weight). Integrating pedestrian exposure variables and truck traffic volume data with better classification of trucks will also help better estimate the freight crashes. Furthermore, this study did not address the modifiable areal unit problem and the boundary crash assignment problem, due in part to crash and some other data sources available only at the heangjungdong level. Therefore, further research could address these methodological limitations. Finally, considering the geographical scope of the present study, conducting additional research in other cities or regions, especially East Asian regions with similar built environments, could serve to enhance the generalizability of this study's findings.
|
|
|
+
|
|
|
+### **Funding**
|
|
|
+
|
|
|
+This work was supported by the Ministry of Education of the Republic of Korea and the National Research Foundation of Korea (NRF-2022S1A5A8052305).
|
|
|
+
|
|
|
+### **Declaration of competing interest**
|
|
|
+
|
|
|
+The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
|
|
|
+
|
|
|
+### **Acknowledgment**
|
|
|
+
|
|
|
+The author would like to thank the participants of the 2022 Association of Collegiate Schools of Planning annual meeting for their valuable comments. The author is also grateful to KoROAD for providing the traffic crash data.
|
|
|
+
|
|
|
+**Appendix A. Summary of major changes in laws and regulations related to freight vehicle safety in South Korea**
|
|
|
+
|
|
|
+| Year | Revisions of laws and<br>regulations | Target | Revised content |
|
|
|
+|------|--------------------------------------|---------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
|
|
|
+| 1995 | Motor Vehicle Safety<br>Standard act | Vehicle | Freight vehicles (weighing more than 8 tons) are required to have a tachograph installed. |
|
|
|
+| 2004 | Trucking Transport<br>Business act | Driver | The qualification requirements for freight transport operators are strengthened. |
|
|
|
+| 2005 | Motor Vehicle Safety<br>Standard act | Vehicle | The installation of speed limiting devices on freight vehicles with a gross vehicle weight rating (GVWR) of 16 tons or more, or a<br>maximum payload capacity of 8 tons or more, is mandatory. |
|
|
|
+| 2011 | Trucking Transport<br>Business act | Driver | Traffic accidents, traffic law violations, and criminal records under Article 9–2(1) of freight vehicle drivers must be recorded and<br>managed |
|
|
|
+| 2012 | Motor Vehicle Safety<br>Standard act | Vehicle | The mandatory installation of a speed limiter applies to freight vehicles with a total weight exceeding 3.5 tons. |
|
|
|
+| 2012 | Trucking Transport<br>Business act | Driver | Individuals who have violated the prohibition on drunk driving three or more times within the past five years are ineligible to<br>obtain a qualification for professional freight vehicle drivers. |
|
|
|
+| 2017 | Traffic Safety act | Vehicle | The installation of LDWS (Lane Departure Warning System) is mandatory for freight vehicles weighing 20 tons or more. |
|
|
|
+| 2017 | Trucking Transport<br>Business act | Driver | Commercial freight vehicle drivers are required to take a 30-minute break after driving continuously for 4 h. |
|
|
|
+| 2018 | Trucking Transport<br>Business act | Driver | The government ensures a minimum transportation fee for freight vehicle owners. |
|
|
|
+| 2018 | Road act | Road | Freight vehicles are prohibited from using the left lanes on roads with more than two lanes (one-way). |
|
|
|
+
|
|
|
+Source: [Han et al. \(2021\).](#page-15-0)
|
|
|
+
|
|
|
+**Appendix B. Descriptive statistics of haengjungdong**
|
|
|
+
|
|
|
+| | Mean | Standard deviation | 25th percentile | 50th percentile | 75th percentile | 90th percentile |
|
|
|
+|-------------------------------------|---------|--------------------|-----------------|-----------------|-----------------|-----------------|
|
|
|
+| Area (square mile) | 1.1 | 1.6 | 0.3 | 0.5 | 1.1 | 2.8 |
|
|
|
+| Total population | 23819.0 | 10585.7 | 16702.7 | 22622.3 | 29748.7 | 37084.7 |
|
|
|
+| Population density(per square mile) | 49738.8 | 32482.7 | 21220.2 | 47662.6 | 73804.2 | 94137.9 |
|
|
|
+
|
|
|
+**Appendix C. Bayesian univariate conditional autoregressive model for different types of freight crashes**
|
|
|
+
|
|
|
+| | | | Freight vehicle-passenger vehicle crash | | Freight vehicle-freight vehicle crashes | | | Freight vehicle-non-motorized crashes | | |
|
|
|
+|----------------------------------------------------|--------------|-------|-----------------------------------------|---------|-----------------------------------------|--------------------------------------------------------------------------------|---------|---------------------------------------|----------------------------|--|
|
|
|
+| | Mean | S.D. | 95% BCI | Mean | S.D. | 95% BCI | Mean | S.D. | 95% BCI | |
|
|
|
+| Freight-related facilities | | | | | | | | | | |
|
|
|
+| Distance to ICN | − 0.006 | 0.007 | (− 0.022, 0.008) | − 0.005 | 0.012 | (− 0.028, 0.016) | − 0.010 | 0.006 | (− 0.022, 0.0005) | |
|
|
|
+| Distance to logistics terminal | − 0.043 | 0.018 | (¡0.08, ¡0.008) | − 0.049 | 0.024 | (¡0.095, ¡0.011) | − 0.033 | 0.016 | (¡0.063, ¡0.001) | |
|
|
|
+| Warehouse density | 0.060 | 0.036 | (− 0.008, 0.128) | 0.092 | 0.052 | (− 0.006, 0.205) | 0.112 | 0.029 | (0.055, 0.173) | |
|
|
|
+| Roadway characteristics | | | | | | | | | | |
|
|
|
+| Highway density | 0.404 | 0.081 | (0.239, 0.558) | 0.648 | 0.147 | (0.329, 0.917) | 0.120 | 0.095 | (− 0.063, 0.315) | |
|
|
|
+| Density of roads with 8-or-more lanes | 0.130 | 0.024 | (0.083, 0.177) | 0.050 | 0.052 | (− 0.052, 0.148) | 0.065 | 0.027 | (0.015, 0.117) | |
|
|
|
+| Density of roads with 2–7 lanes | 0.019 | 0.011 | (− 0.002, 0.04) | 0.021 | 0.025 | (− 0.025, 0.069) | 0.030 | 0.011 | (0.007, 0.052) | |
|
|
|
+| Density of 1-lane roads | − 0.002 | 0.004 | (− 0.01, 0.006) | 0.001 | 0.009 | (− 0.02, 0.017) | 0.016 | 0.004 | (0.008, 0.024) | |
|
|
|
+| Road intersection density | 0.004 | 0.005 | (− 0.008, 0.014) | − 0.031 | 0.012 | (¡0.057, ¡0.008) | 0.006 | 0.005 | (− 0.005, 0.016) | |
|
|
|
+| Closeness centrality | − 16512 8989 | | (− 33609, 1893) | − 40518 | 14,983 | (¡68975, ¡15857) − 13899 8378 | | | (− 29964, 2289) | |
|
|
|
+| Meshedness coefficient | 0.137 | 0.214 | (− 0.283, 0.541) | − 0.131 | 0.439 | (− 0.981, 0.725) | − 0.305 | 0.268 | (− 0.824, 0.207) | |
|
|
|
+| Built environment | | | | | | | | | | |
|
|
|
+| Employment density | 0.002 | 0.001 | (0.001, 0.004) | 0.001 | 0.002 | (− 0.003, 0.004) | 0.003 | 0.001 | (0.001, 0.004) | |
|
|
|
+| Population density | − 0.012 | 0.002 | (¡0.015, ¡0.008) | − 0.016 | 0.005 | (¡0.024, ¡0.006) | − 0.006 | 0.002 | (¡0.01, ¡0.002) | |
|
|
|
+| Land use mix | 0.177 | 0.222 | (− 0.27, 0.586) | 0.752 | 0.394 | (− 0.102, 1.418) | 0.924 | 0.234 | (0.45, 1.373) | |
|
|
|
+| Residential land share | 0.197 | 0.244 | (− 0.258, 0.699) | − 0.191 | 0.568 | (− 1.29, 0.918) | 0.537 | 0.245 | (0.052, 1.037) | |
|
|
|
+| Grocery store density | − 0.006 | 0.021 | (− 0.047, 0.037) | − 0.035 | 0.049 | (− 0.131, 0.06) | 0.032 | 0.020 | (− 0.008, 0.07) | |
|
|
|
+| Shopping mall density | 0.016 | 0.017 | (− 0.019, 0.047) | 0.014 | 0.043 | (− 0.075, 0.093) | 0.001 | 0.018 | (− 0.034, 0.035) | |
|
|
|
+| Elementary school density | − 0.039 | 0.015 | (¡0.067, ¡0.01) | − 0.106 | 0.040 | (¡0.181, ¡0.026) | − 0.023 | 0.014 | (− 0.051, 0.003) | |
|
|
|
+| Hospital density | 0.015 | 0.034 | (− 0.049, 0.084) | 0.026 | 0.092 | (− 0.16, 0.208) | 0.052 | 0.034 | (− 0.014, 0.119) | |
|
|
|
+| Exposure variable | | | | | | | | | | |
|
|
|
+| Truck volume density | 0.050 | 0.019 | (0.012, 0.086) | 0.002 | 0.033 | (− 0.061, 0.068) | 0.027 | 0.020 | (− 0.016, 0.066) | |
|
|
|
+| Passenger vehicle volume density | − 0.011 | 0.004 | (¡0.018, ¡0.002) | 0.002 | 0.004 | (− 0.005, 0.012) | − 0.016 | 0.004 | (¡0.023, ¡0.007) | |
|
|
|
+| Passenger vehicle volume density (squared) 3.2E-06 | | | | | | 1.68E-05 (− 3.3E-05, 3.1E-05) − 1.10E-05 3.04E-05 (− 6.4E-05, 5.1E-05) 4.7E-05 | | | 1.8E-05 (1.2E-05, 7.8E-05) | |
|
|
|
+| Non-motorized infrastructure | | | | | | | | | | |
|
|
|
+| Sidewalk density | 0.020 | 0.007 | (0.007, 0.033) | 0.039 | 0.014 | (0.016, 0.066) | − 0.002 | 0.007 | (− 0.016, 0.011) | |
|
|
|
+| Sidewalk connectivity | − 0.266 | 0.860 | (− 1.878, 1.685) | 0.693 | 1.512 | (− 2.666, 3.68) | − 0.049 | 0.854 | (− 1.669, 1.641) | |
|
|
|
+| Other variables | | | | | | | | | | |
|
|
|
+| Square mile | 0.188 | 0.028 | (0.13, 0.243) | 0.257 | 0.026 | (0.208, 0.312) | 0.112 | 0.026 | (0.06, 0.163) | |
|
|
|
+| Seoul (yes = 1, no = 0) | 0.186 | 0.111 | (− 0.034, 0.396) | 0.266 | 0.171 | (− 0.066, 0.587) | 0.489 | 0.116 | (0.25, 0.705) | |
|
|
|
+| Intercept | 2.490 | 0.503 | (1.438, 3.421) | 0.950 | 0.778 | (− 0.402, 2.417) | 0.900 | 0.459 | (0.047, 1.803) | |
|
|
|
+
|
|
|
+<span id="page-15-0"></span>Notes: S.D." Standard deviation, BCI: Bayesian credible interval; *N* = 754; DIC = 8285.62. BCI values in bold were statistically significant at the 0.05 level.
|
|
|
+
|
|
|
+#### **References**
|
|
|
+
|
|
|
+- [Abdel-Aty, M., Lee, J., Siddiqui, C., & Choi, K. \(2013\). Geographical unit based analysis](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0005) [in the context of transportation safety planning.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0005) *Transportation Research Part A: [Policy and Practice, 49](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0005)*, 62–75.
|
|
|
+- [Barajas, J. M. \(2018\). Not all crashes are created equal.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0010) *Journal of Transport and Land Use, 11*[\(1\), 865](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0010)–882.
|
|
|
+- [Barajas, J. M. \(2021\). Biking where Black: Connecting transportation planning and](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0015) [infrastructure to disproportionate policing.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0015) *Transportation Research Part D: Transport [and Environment, 99](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0015)*, Article 103027.
|
|
|
+- [Bunn, T. L., Slavova, S., & Rock, P. J. \(2019\). Association between commercial vehicle](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0020) [driver at-fault crashes involving sleepiness/fatigue and proximity to rest areas and](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0020) truck stops. *[Accident Analysis](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0020) & Prevention, 126*, 3–9.
|
|
|
+- [Cai, Q., Lee, J., Eluru, N., & Abdel-Aty, M. \(2016\). Macro-level pedestrian and bicycle](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0025) [crash analysis: Incorporating spatial spillover effects in dual state count models.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0025) *[Accident Analysis](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0025) & Prevention, 93*, 14–22.
|
|
|
+- [Chen, P. \(2015\). Built environment factors in explaining the automobile-involved bicycle](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0030) [crash frequencies: A spatial statistic approach.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0030) *Safety Science, 79*, 336–343.
|
|
|
+- [Chen, P., Sun, F., Wang, Z., Gao, X., Jiao, J., & Tao, Z. \(2018\). Built environment effects](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0035) [on bike crash frequency and risk in Beijing.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0035) *Journal of Safety Research, 64*, 135–143.
|
|
|
+- [Chi, G., & Zhu, J. \(2019\). Spatial regression models for the social sciences.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0040) *SAGE*. [publications](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0040).
|
|
|
+- [Choi, D. A., & Ewing, R. \(2021\). Effect of street network design on traffic congestion and](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0045) traffic safety. *[Journal of Transport Geography, 96](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0045)*, Article 103200.
|
|
|
+- Choi, S.E. (November 17, 2020). "2-year-old girl died as truck crashed into school zone; residents lamented the absence of traffic light.". *Seoulsinmun*. [https://www.seoul.co.](https://www.seoul.co.kr/news/newsView.php?id=20201117500234) [kr/news/newsView.php?id](https://www.seoul.co.kr/news/newsView.php?id=20201117500234)=20201117500234 (in Korean).
|
|
|
+- [Crucitti, P., Latora, V., & Porta, S. \(2006\). Centrality measures in spatial networks of](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0055) urban streets. *[Physical Review E, 73](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0055)*(3), Article 036125.
|
|
|
+- [Das, S., Islam, M., Dutta, A., & Shimu, T. H. \(2020\). Uncovering deep structure of](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0060) [determinants in large truck fatal crashes.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0060) *Transportation Research Record, 2674*(9), 742–[754](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0060).
|
|
|
+- [Desapriya, E., Subzwari, S., Sasges, D., Basic, A., Alidina, A., Turcotte, K., & Pike, I.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0065) [\(2010\). Do light truck vehicles \(LTV\) impose greater risk of pedestrian injury than](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0065) [passenger cars? A meta-analysis and systematic review.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0065) *Traffic Injury Prevention, 11* [\(1\), 48](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0065)–56.
|
|
|
+- [Ding, H., Lu, Y., Sze, N. N., Chen, T., Guo, Y., & Lin, Q. \(2022\). A deep generative](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0070) [approach for crash frequency model with heterogeneous imbalanced data.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0070) *Analytic [Methods in Accident Research, 34](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0070)*, Article 100212.
|
|
|
+- [Ding, H., Sze, N. N., Guo, Y., & Li, H. \(2021\). Role of exposure in bicycle safety analysis:](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0075) [Effect of cycle path choice.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0075) *Accident Analysis & Prevention, 153*, Article 106014.
|
|
|
+- [Ding, H., Sze, N. N., Li, H., & Guo, Y. \(2020\). Roles of infrastructure and land use in](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0080) [bicycle crash exposure and frequency: A case study using Greater London bike](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0080) sharing data. *[Accident Analysis](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0080) & Prevention, 144*, Article 105652.
|
|
|
+- [Dong, C., Dong, Q., Huang, B., Hu, W., & Nambisan, S. S. \(2017\). Estimating factors](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0085) [contributing to frequency and severity of large truck](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0085)–involved crashes. *Journal of [Transportation Engineering, Part A: Systems, 143](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0085)*(8), Article 040170.
|
|
|
+- [Dumbaugh, E., & Li, W. \(2010\). Designing for the safety of pedestrians, cyclists, and](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0090) motorists in urban environments. *[Journal of the American Planning Association, 77](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0090)*(1), 69–[88](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0090).
|
|
|
+- [Duncan, C. S., Khattak, A. J., & Council, F. M. \(1998\). Applying the ordered probit model](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0095) [to injury severity in truck-passenger car rear-end collisions.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0095) *Transportation Research [Record, 1635](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0095)*(1), 63–71.
|
|
|
+- [El-Basyouny, K., & Sayed, T. \(2009\). Collision prediction models using multivariate](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0100) [Poisson-lognormal regression.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0100) *Accident Analysis & Prevention, 41*(4), 820–828.
|
|
|
+- [Giuliano, G., Hou, Y., Kang, S., & Shin, E. J. \(2022\). Polycentricity and the evolution of](http://refhub.elsevier.com/S0022-4375(23)00180-9/optTsnjxt70LO) [metropolitan spatial structure.](http://refhub.elsevier.com/S0022-4375(23)00180-9/optTsnjxt70LO) *Growth and Change, 53*(2), 593–627.
|
|
|
+- Gonzalez, ´ [S. R., Loukaitou-Sideris, A., & Chapple, K. \(2019\). Transit neighborhoods,](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0105) [commercial gentrification, and traffic crashes: Exploring the linkages in Los Angeles](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0105) and the Bay Area. *[Journal of Transport Geography, 77](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0105)*, 79–89.
|
|
|
+- [Guerra, E., Dong, X., & Kondo, M. \(2019\). Do denser neighborhoods have safer streets?](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0110) [Population density and traffic safety in the Philadelphia region.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0110) *Journal of Planning [Education and Research, 0739456X19845043](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0110)*.
|
|
|
+- [Han, S. J., Kang, S. G., & Kim, S. H. \(2021\). Analysis on risk factors of commercial freight](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0115) [truck crashes to identify safety measures.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0115) *Korea Transport Institute report.*
|
|
|
+- [Heitz, A., Launay, P., & Beziat, A. \(2019\). Heterogeneity of logistics facilities: An issue for](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0120) [a better understanding and planning of the location of logistics facilities.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0120) *European [Transport Research Review, 11](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0120)*(1), 5.
|
|
|
+- [Heydari, S., Forrest, M., & Preston, J. \(2022\). Investigating the association between](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0125) [neighbourhood characteristics and e-scooter safety.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0125) *Sustainable Cities and Society, 83*, [Article 103982](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0125).
|
|
|
+- [Houston, D., Li, W., & Wu, J. \(2014\). Disparities in exposure to automobile and truck](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0130) [traffic and vehicle emissions near the Los Angeles-Long Beach port complex.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0130) *[American Journal of Public Health, 104](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0130)*(1), 156–164.
|
|
|
+- International Transport Forum. (2023). Freight transport: Road freight transport. Available at: [https://stats.oecd.org/Index.aspx?](https://stats.oecd.org/Index.aspx?%26datasetcode=ITF_GOODS_TRANSPORT)&datasetcode=ITF\_GOODS\_ [TRANSPORT](https://stats.oecd.org/Index.aspx?%26datasetcode=ITF_GOODS_TRANSPORT).
|
|
|
+- [Jia, R., Khadka, A., & Kim, I. \(2018\). Traffic crash analysis with point-of-interest spatial](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0140) clustering. *[Accident Analysis](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0140) & Prevention, 121*, 223–230.
|
|
|
+
|
|
|
+- Jung, S.T. (April 8, 2016). ""Illegal parking up to the crosswalk... Startled by a pedestrian who suddenly appeared" *Dongailbo.* [https://www.donga.com/news/Society/ar](https://www.donga.com/news/Society/article/all/20160408/77466037/1) [ticle/all/20160408/77466037/1](https://www.donga.com/news/Society/article/all/20160408/77466037/1) (in Korean).
|
|
|
+- Kang, M. S. (November 11, 2021). "Illegally parked truck on sidewalks backed up without seeing the child." *Segyeilbo.* <https://segye.com/view/20211110517676> (in Korean).
|
|
|
+- Kim, H. S. (April 22, 2022). "Why there were so many right-turn accidents with large trucks" *Autoherald*. [http://www.autoherald.co.kr/news/articleView.html?idxn](http://www.autoherald.co.kr/news/articleView.html?idxno=43695) o=[43695](http://www.autoherald.co.kr/news/articleView.html?idxno=43695) (in Korean).
|
|
|
+- Kim, T. Y., & Park, H. J. (March 23, 2021). "Elementary school student killed in traffic accident... Safety concerns persist in Incheon school zones. Six-lane road in front of school... Trucks constantly speeding dangerously." *Kyeonginilbo.* [http://www.](http://www.kyeongin.com/main/view.php?key=20210322010004650) [kyeongin.com/main/view.php?key](http://www.kyeongin.com/main/view.php?key=20210322010004650)=20210322010004650 (in Korean).
|
|
|
+- KoROAD Traffic Analysis Accident System website. (2023). Description of traffic crash database (in Korean). <https://taas.koroad.or.kr/sta/acs/exs/wordArngPopup.do>.
|
|
|
+- [Lee, J., Abdel-Aty, M., & Jiang, X. \(2015\). Multivariate crash modeling for motor vehicle](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0170) [and non-motorized modes at the macroscopic level.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0170) *Accident Analysis & Prevention, 78*[, 146](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0170)–154.
|
|
|
+- [Liu, J., Khattak, A. J., Li, X., Nie, Q., & Ling, Z. \(2020\). Bicyclist injury severity in traffic](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0175) [crashes: A spatial approach for geo-referenced crash data to uncover non-stationary](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0175) correlates. *[Journal of Safety Research, 73](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0175)*, 25–35.
|
|
|
+- [Loukaitou-Sideris, A., Liggett, R., & Sung, H. G. \(2007\). Death on the crosswalk: A study](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0180) [of pedestrian-automobile collisions in Los Angeles.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0180) *Journal of Planning Education and [Research, 26](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0180)*(3), 338–351.
|
|
|
+- [Ma, J., & Kockelman, K. M. \(2006\). Bayesian multivariate Poisson regression for models](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0185) of injury count, by severity. *[Transportation Research Record, 1950](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0185)*(1), 24–34.
|
|
|
+- Malczyk, A., & Bende, J. (2019). Heavy truck crashes involving pedestrians in comparison to bicyclists. In *26th International Technical Conference on the Enhanced Safety of Vehicles (ESV): Technology: Enabling a Safer Tomorrow National Highway Traffic Safety Administration* (No. 19-0080).
|
|
|
+- [Marshall, W. E., & Garrick, N. W. \(2011\). Does street network design affect traffic safety?](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0195) *[Accident Analysis](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0195) & Prevention, 43*(3), 769–781.
|
|
|
+- [Martin, J. L. \(2002\). Relationship between crash rate and hourly traffic flow on](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0200) [interurban motorways.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0200) *Accident Analysis & Prevention, 34*(5), 619–629.
|
|
|
+- [Merlin, L. A., Guerra, E., & Dumbaugh, E. \(2020\). Crash risk, crash exposure, and the](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0205) [built environment: A conceptual review.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0205) *Accident Analysis & Prevention, 134*, Article [105244](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0205).
|
|
|
+- [Park, H. C., Park, B. J., & Park, P. Y. \(2022\). A multiple membership multilevel negative](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0210) [binomial model for intersection crash analysis.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0210) *Analytic Methods in Accident Research, 35*[, Article 100228](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0210).
|
|
|
+- Park, S. M., & Park, J. Y. (2018). *[September\). Analyzing the severity of freight vehicle crashes](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0215) [using order probit model](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0215)*. South Korea.
|
|
|
+- Park, S.H., & Kim, J. H. (July 20, 2020). "Chaos on Day 2 of Bike Lane Opening: Taxi and Motorcycle 'Invasion' Leaves Bikers 'Dazed" *Chosunilbo.* [https://www.chosun.com/](https://www.chosun.com/site/data/html_dir/2018/04/10/2018041000996.html) [site/data/html\\_dir/2018/04/10/2018041000996.html](https://www.chosun.com/site/data/html_dir/2018/04/10/2018041000996.html) (in Korean).
|
|
|
+- Pljaki´c, M., Jovanovi´c, D., Matovi´c, B., & Mi´ci´[c, S. \(2019\). Macro-level accident modeling](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0225) [in Novi Sad: A spatial regression approach.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0225) *Accident Analysis & Prevention, 132*,
|
|
|
+- [Pokorny, P., & Pitera, K. \(2019\). Truck-bicycle safety: An overview of methods of study,](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0230) risk factors and research needs. *[European Transport Research Review, 11](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0230)*(1), 1–14.
|
|
|
+- [Qiao, S., Yeh, A. G. O., Zhang, M., & Yan, X. \(2020\). Effects of state-led suburbanization](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0235) [on traffic crash density in China: Evidence from the Chengdu City Proper.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0235) *Accident Analysis & Prevention, 148*[, Article 105775](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0235).
|
|
|
+- [Osama, A., & Sayed, T. \(2017\). Evaluating the impact of connectivity, continuity, and](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0240) [topography of sidewalk network on pedestrian safety.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0240) *Accident Analysis & Prevention, 107*[, 117](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0240)–125.
|
|
|
+- [Rahimi, A., Azimi, G., Asgari, H., & Jin, X. \(2020\). In](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0245) *August). Injury severity of pedestrian [and bicyclist crashes involving large trucks](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0245)* (pp. 110–122). Reston, VA: American [Society of Civil Engineers](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0245).
|
|
|
+- [Ranaiefar, F., & Amelia, R. \(2011\). Freight-transportation externalities.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0250) *Logistics [Operations and Management](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0250)*, 333–358.
|
|
|
+- [Romanow, N. T., Couperthwaite, A. B., McCormack, G. R., Nettel-Aguirre, A.,](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0255) [Rowe, B. H., & Hagel, B. E. \(2012\). Environmental determinants of bicycling injuries](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0255) in Alberta, Canada. *[Journal of Environmental and Public Health, 2012](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0255)*.
|
|
|
+- [Schuetz, J., Giuliano, G., & Shin, E. J. \(2018a\). Can a car-centric city become transit](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0260) [oriented? Evidence from Los Angeles.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0260) *Cityscape, 20*(1), 167–190.
|
|
|
+- [Schuetz, J., Giuliano, G., & Shin, E. J. \(2018b\). Does zoning help or hinder transit](http://refhub.elsevier.com/S0022-4375(23)00180-9/opti5GUiq6MIu)[oriented \(re\) development?](http://refhub.elsevier.com/S0022-4375(23)00180-9/opti5GUiq6MIu) *Urban Studies, 55*(8), 1672–1689.
|
|
|
+- [Shankar, V. N., Ulfarsson, G. F., Pendyala, R. M., & Nebergall, M. B. \(2003\). Modeling](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0265) [crashes involving pedestrians and motorized traffic.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0265) *Safety Science, 41*(7), 627–640.
|
|
|
+- Shin, D.M. (February 23, 2023). "Large freight vehicles illegally parked on sidewalks overnight pose great threat: will authorities crack down on this dangerous trend?" *AnsanTimes.* [http://www.isntimes.co.kr/news/articleView.html?idxno](http://www.isntimes.co.kr/news/articleView.html?idxno=25298)=25298 (in Korean).
|
|
|
+- [Shin, E. J. \(2021\). Spatial segregation of Chinese immigrants in Seoul, South Korea,](http://refhub.elsevier.com/S0022-4375(23)00180-9/optEfKR3FTkRt) [during the COVID-19 pandemic: Evidence from population data derived from mobile](http://refhub.elsevier.com/S0022-4375(23)00180-9/optEfKR3FTkRt) phone signals. *[The Social Science Journal](http://refhub.elsevier.com/S0022-4375(23)00180-9/optEfKR3FTkRt)*, 1–22.
|
|
|
+- [Shin, E. J. \(2023\). Decomposing neighborhood disparities in bicycle crashes: A Gelbach](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0275) [decomposition analysis.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0275) *Transport Policy, 131*, 156–172.
|
|
|
+- [Siddiqui, C., Abdel-Aty, M., & Choi, K. \(2012\). Macroscopic spatial analysis of pedestrian](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0280) [and bicycle crashes.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0280) *Accident Analysis & Prevention, 45*, 382–391.
|
|
|
+
|
|
|
+- <span id="page-16-0"></span>[Stutts, J. C., Williamson, J. E., Whitley, T., & Sheldon, F. C. \(1990\). Bicycle accidents and](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0285) [injuries: A pilot study comparing hospital-and police-reported data.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0285) *Accident Analysis & [Prevention, 22](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0285)*(1), 67–78.
|
|
|
+- [Taylor, S. G., Russo, B. J., & James, E. \(2018\). A comparative analysis of factors affecting](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0290) [the frequency and severity of freight-involved and non-freight crashes on a major](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0290) freight corridor freeway. *[Transportation Research Record, 2672](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0290)*(34), 49–62.
|
|
|
+- [Wang, J., Parajuli, S., Cherry, C. R., McDonald, N. C., & Lyons, T. \(2022\). Vulnerable](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0295) [road user safety and freight vehicles: A case study in North Carolina and Tennessee.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0295) *[Transportation research interdisciplinary perspectives, 15](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0295)*, Article 100650.
|
|
|
+- [Wang, X., Wu, X., Abdel-Aty, M., & Tremont, P. J. \(2013\). Investigation of road network](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0300)
|
|
|
+- [features and safety performance.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0300) *Accident Analysis & Prevention, 56*, 22–31. [Wier, M., Weintraub, J., Humphreys, E. H., Seto, E., & Bhatia, R. \(2009\). An area-level](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0305) [model of vehicle-pedestrian injury collisions with implications for land use and](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0305)
|
|
|
+- [transportation planning.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0305) *Accident Analysis & Prevention, 41*(1), 137–145. [Xu, P., Huang, H., & Dong, N. \(2018\). The modifiable areal unit problem in traffic safety:](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0310) [Basic issue, potential solutions and future research.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0310) *Journal of Traffic and [Transportation Engineering, 5](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0310)*(1), 73–82.
|
|
|
+- [Yang, C., Chen, M., & Yuan, Q. \(2021a\). The geography of freight-related accidents in the](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0315) [era of E-commerce: Evidence from the Los Angeles metropolitan area.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0315) *Journal of [Transport Geography, 92](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0315)*, Article 102989.
|
|
|
+- [Yang, C., Chen, M., & Yuan, Q. \(2021b\). The application of XGBoost and SHAP to](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0320) [examining the factors in freight truck-related crashes: An exploratory analysis.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0320) *[Accident Analysis](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0320) & Prevention, 158*, Article 106153.
|
|
|
+
|
|
|
+- [Yuan, Q., & Wang, J. \(2021\). Goods movement, road safety, and spatial inequity:](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0325) [Evaluating freight-related crashes in low-income or minority neighborhoods.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0325) *Journal [of Transport Geography, 96](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0325)*, Article 103186.
|
|
|
+- [Zhai, X., Huang, H., Gao, M., Dong, N., & Sze, N. N. \(2018\). Boundary crash data](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0330) [assignment in zonal safety analysis: An iterative approach based on data](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0330) [augmentation and Bayesian spatial model.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0330) *Accident Analysis & Prevention, 121*, 231–[237](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0330).
|
|
|
+- Zhang, Y., Bigham, J., Li, Z., Ragland, D., & Chen, X. (2012). Associations between road network connectivity and pedestrian-bicyclist accidents (No. 12-0478).
|
|
|
+- [Zhu, X., & Srinivasan, S. \(2011\). A comprehensive analysis of factors influencing the](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0340) [injury severity of large-truck crashes.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0340) *Accident Analysis & Prevention, 43*(1), 49–57. [Ziakopoulos, A., & Yannis, G. \(2020\). A review of spatial approaches in road safety.](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0345) *[Accident Analysis](http://refhub.elsevier.com/S0022-4375(23)00180-9/h0345) & Prevention, 135*, Article 105323.
|
|
|
+
|
|
|
+**Eun Jin Shin** obtained her doctoral degree in Urban Planning and Development from the University of Southern California. Currently, she is an assistant professor in the Department of Public Administration and the Graduate School of Governance at Sungkyunkwan University, South Korea. Her research interests include transportation policy and transportation safety. Her recent publications appeared in Transportation Research Part A: Policy and Practice and Transport Policy.
|
