Multivariate crash modeling for motor vehicle and non-motorized modes at the macroscopic level

Predicting crash frequency for multi-vehicle collision types using multivariate Poisson-lognormal spatial model: A comparative analysis

According to crash configuration and pre-crash conditions, traffic crashes are classified into different collision types. Based on the literature, multi-vehicle crashes, such as head-on, rear-end, and angle crashes, are more frequent than single-vehicle crashes, and most often result in serious consequences. From a methodological point of view, the majority of prior studies focused on multivehicle collisions have employed univariate count models to estimate crash counts separately by collision type. However, univariate models fail to account for correlations which may exist between different collision types. Among others, multivariate Poisson lognormal (MVPLN) model with spatial correlation is a promising multivariate specification because it not only allows for unobserved heterogeneity (extra-Poisson variation) and dependencies between collision types, but also spatial correlation between adjacent sites. However, the MVPLN spatial model has rarely been applied in previous research for simultaneously modelling crash counts by collision type. Therefore, this study aims at utilizing a MVPLN spatial model to estimate crash counts for four different multi-vehicle collision types, including head-on, rear-end, angle, and sideswipe collisions. To investigate the performance of the MVPLN spatial model, a two-stage model and a univariate Poisson lognormal model (UNPLN) spatial model were also developed in this study. Detailed information on roadway characteristics, traffic volume, and crash history were collected on 407 homogeneous segments from Malaysian federal roads. The results indicate that the MVPLN spatial model outperforms the other comparing models in terms of goodness-of-fit measures. The results also show that the inclusion of spatial heterogeneity in the multivariate model significantly improves the model fit, as indicated by the Deviance Information Criterion (DIC). The correlation between crash types is high and positive, implying that the occurrence of a specific collision type is highly associated with the occurrence of other crash types on the same road segment. These results support the utilization of the MVPLN spatial model when predicting crash counts by collision manner. In terms of contributing factors, the results show that distinct crash types are attributed to different subsets of explanatory variables.

Analysis of stationary and dynamic factors affecting highway accident occurrence: A dynamic correlated grouped random parameters binary logit approach

Traditional accident analysis typically explores non-time-varying (stationary) factors that affect accident occurrence on roadway segments. However, the impact of time-varying (dynamic) factors is not thoroughly investigated. This paper seeks to simultaneously identify pre-crash stationary and dynamic factors of accident occurrence, while accounting for unobserved heterogeneity. Using highly disaggregate information for the potential dynamic factors, and aggregate data for the traditional stationary elements, a dynamic binary random parameters (mixed) logit framework is employed. With this approach, the dynamic nature of weather-related, and driving- and pavement-condition information is jointly investigated with traditional roadway geometric and traffic characteristics. To additionally account for the combined effect of the dynamic and stationary factors on the accident occurrence, the developed random parameters logit framework allows for possible correlations among the random parameters. The analysis is based on crash and non-crash observations between 2011 and 2013, drawn from urban and rural highway segments in the state of Washington. The findings show that the proposed methodological framework can account for both stationary and dynamic factors affecting accident occurrence probabilities, for panel effects, for unobserved heterogeneity through the use of random parameters, and for possible correlation among the latter. The comparative evaluation among the correlated grouped random parameters, the uncorrelated random parameters logit models, and their fixed parameters logit counterpart, demonstrate the potential of the random parameters modeling, in general, and the benefits of the correlated grouped random parameters approach, specifically, in terms of statistical fit and explanatory power.

Multimodal crash frequency modeling: Multivariate space-time models with alternate spatiotemporal interactions

Enhancement of safety for all transportation mode users plays an essential role in the implementation of multimodal transportation systems. Compared with crash frequency models dedicated to motorized mode users, the use of these models has been considerably scarce in the multimodal literature. To fill this research gap, the authors aimed to develop and evaluate three multivariate space-time models with different temporal trends and spatiotemporal interactions. The model estimates justified the use of mode-varying coefficients for explanatory variables as the impact of these factors varied across different crash modes. Largely, a similar set of influential covariates was generated by the three models which indicate their robustness. However, notable differences were observed from the assessment of evaluation criteria pertaining to predictive accuracy based on criteria assessing the training and test errors. The model with time-varying spatial random effects demonstrated superior performance for training and test errors. However, due to the significant increase in number of effective parameters that were utilized for model development, this model appeared to have the largest value of deviance information criterion (DIC). In terms of the comparison between models based on site ranking performance, the time-varying spatial random effects model demonstrated the best performance in both site consistency and method consistency. In other words, the superiority of the model's predictive performance could be transferred to yield more accurate result at site ranking.

Analysis of crash proportion by vehicle type at traffic analysis zone level: A mixed fractional split multinomial logit modeling approach with spatial effects

In traffic safety literature, crash frequency variables are analyzed using univariate count models or multivariate count models. In this study, we propose an alternative approach to modeling multiple crash frequency dependent variables. Instead of modeling the frequency of crashes we propose to analyze the proportion of crashes by vehicle type. A flexible mixed multinomial logit fractional split model is employed for analyzing the proportions of crashes by vehicle type at the macro-level. In this model, the proportion allocated to an alternative is probabilistically determined based on the alternative propensity as well as the propensity of all other alternatives. Thus, exogenous variables directly affect all alternatives. The approach is well suited to accommodate for large number of alternatives without a sizable increase in computational burden. The model was estimated using crash data at Traffic Analysis Zone (TAZ) level from Florida. The modeling results clearly illustrate the applicability of the proposed framework for crash proportion analysis. Further, the Excess Predicted Proportion (EPP)-a screening performance measure analogous to Highway Safety Manual (HSM), Excess Predicted Average Crash Frequency is proposed for hot zone identification. Using EPP, a statewide screening exercise by the various vehicle types considered in our analysis was undertaken. The screening results revealed that the spatial pattern of hot zones is substantially different across the various vehicle types considered.

Exploring the safety in numbers effect for vulnerable road users on a macroscopic scale

A "Safety in Numbers" effect for a certain group of road users is present if the number of crashes increases at a lower rate than the number of road users. The existence of this effect has been invoked to justify investments in multimodal transportation improvements in order to create more sustainable urban transportation systems by encouraging walking, biking, and transit ridership. The goal of this paper is to explore safety in numbers effect for cyclists and pedestrians in areas with different levels of access to multimodal infrastructure. Data from Chicago served to estimate the expected number of crashes on the census tract level by applying Generalized Additive Models (GAM) to capture spatial dependence in crash data. Measures of trip generation, multimodal infrastructure, network connectivity and completeness, and accessibility were used to model travel exposure in terms of activity, number of trips, trip length, travel opportunities, and conflicts. The results show that a safety in numbers effect exists on a macroscopic level for motor vehicles, pedestrians, and bicyclists.

Macro-level vulnerable road users crash analysis: A Bayesian joint modeling approach of frequency and proportion

This study aims at contributing to the literature on pedestrian and bicyclist safety by building on the conventional count regression models to explore exogenous factors affecting pedestrian and bicyclist crashes at the macroscopic level. In the traditional count models, effects of exogenous factors on non-motorist crashes were investigated directly. However, the vulnerable road users' crashes are collisions between vehicles and non-motorists. Thus, the exogenous factors can affect the non-motorist crashes through the non-motorists and vehicle drivers. To accommodate for the potentially different impact of exogenous factors we convert the non-motorist crash counts as the product of total crash counts and proportion of non-motorist crashes and formulate a joint model of the negative binomial (NB) model and the logit model to deal with the two parts, respectively. The formulated joint model is estimated using non-motorist crash data based on the Traffic Analysis Districts (TADs) in Florida. Meanwhile, the traditional NB model is also estimated and compared with the joint model. The result indicates that the joint model provides better data fit and can identify more significant variables. Subsequently, a novel joint screening method is suggested based on the proposed model to identify hot zones for non-motorist crashes. The hot zones of non-motorist crashes are identified and divided into three types: hot zones with more dangerous driving environment only, hot zones with more hazardous walking and cycling conditions only, and hot zones with both. It is expected that the joint model and screening method can help decision makers, transportation officials, and community planners to make more efficient treatments to proactively improve pedestrian and bicyclist safety.

Application of Bayesian informative priors to enhance the transferability of safety performance functions

INTRODUCTION

Safety performance functions (SPFs) are essential tools for highway agencies to predict crashes, identify hotspots and assess safety countermeasures. In the Highway Safety Manual (HSM), a variety of SPFs are provided for different types of roadway facilities, crash types and severity levels. Agencies, lacking the necessary resources to develop own localized SPFs, may opt to apply the HSM's SPFs for their jurisdictions. Yet, municipalities that want to develop and maintain their regional SPFs might encounter the issue of the small sample bias. Bayesian inference is being conducted to address this issue by combining the current data with prior information to achieve reliable results. It follows that the essence of Bayesian statistics is the application of informative priors, obtained from other SPFs or experts' experiences.

METHOD

In this study, we investigate the applicability of informative priors for Bayesian negative binomial SPFs for rural divided multilane highway segments in Florida and California. An SPF with non-informative priors is developed for each state and its parameters' distributions are assigned to the other state's SPF as informative priors. The performances of SPFs are evaluated by applying each state's SPFs to the other state. The analysis is conducted for both total (KABCO) and severe (KAB) crashes.

RESULTS, CONCLUSIONS AND PRACTICAL APPLICATIONS

As per the results, applying one state's SPF with informative priors, which are the other state's SPF independent variable estimates, to the latter state's conditions yields better goodness of fit (GOF) values than applying the former state's SPF with non-informative priors to the conditions of the latter state. This is for both total and severe crash SPFs. Hence, for localities where it is not preferred to develop own localized SPFs and adopt SPFs from elsewhere to cut down on resources, application of informative priors is shown to facilitate the process.

Will higher traffic flow lead to more traffic conflicts? A crash surrogate metric based analysis

In this paper, we aim to examine the relationship between traffic flow and potential conflict risks by using crash surrogate metrics. It has been widely recognized that one traffic flow corresponds to two distinct traffic states with different speeds and densities. In view of this, instead of simply aggregating traffic conditions with the same traffic volume, we represent potential conflict risks at a traffic flow fundamental diagram. Two crash surrogate metrics, namely, Aggregated Crash Index and Time to Collision, are used in this study to represent the potential conflict risks with respect to different traffic conditions. Furthermore, Beijing North Ring III and Next Generation SIMulation Interstate 80 datasets are utilized to carry out case studies. By using the proposed procedure, both datasets generate similar trends, which demonstrate the applicability of the proposed methodology and the transferability of our conclusions.

Intersection crash prediction modeling with macro-level data from various geographic units

There have been great efforts to develop traffic crash prediction models for various types of facilities. The crash models have played a key role to identify crash hotspots and evaluate safety countermeasures. In recent, many macro-level crash prediction models have been developed to incorporate highway safety considerations in the long-term transportation planning process. Although the numerous macro-level studies have found that a variety of demographic and socioeconomic zonal characteristics have substantial effects on traffic safety, few studies have attempted to coalesce micro-level with macro-level data from existing geographic units for estimating crash models. In this study, the authors have developed a series of intersection crash models for total, severe, pedestrian, and bicycle crashes with macro-level data for seven spatial units. The study revealed that the total, severe, and bicycle crash models with ZIP-code tabulation area data performs the best, and the pedestrian crash models with census tract-based data outperforms the competing models. Furthermore, it was uncovered that intersection crash models can be drastically improved by only including random-effects for macro-level entities. Besides, the intersection crash models are even further enhanced by including other macro-level variables. Lastly, the pedestrian and bicycle crash modeling results imply that several macro-level variables (e.g., population density, proportions of specific age group, commuters who walk, or commuters using bicycle, etc.) can be a good surrogate exposure for those crashes.

Comparison of Multivariate Poisson lognormal spatial and temporal crash models to identify hot spots of intersections based on crash types

Most of the studies are focused on the general crashes or total crash counts with considerably less research dedicated to different crash types. This study employs the Systemic approach for detection of hotspots and comprehensively cross-validates five multivariate models of crash type-based HSID methods which incorporate spatial and temporal random effects. It is anticipated that comparison of the crash estimation results of the five models would identify the impact of varied random effects on the HSID. The data over a ten year time period (2003-2012) were selected for analysis of a total 137 intersections in the City of Corona, California. The crash types collected in this study include: Rear-end, Head-on, Side-swipe, Broad-side, Hit object, and Others. Statistically significant correlations among crash outcomes for the heterogeneity error term were observed which clearly demonstrated their multivariate nature. Additionally, the spatial random effects revealed the correlations among neighboring intersections across crash types. Five cross-validation criteria which contains, Residual Sum of Squares, Kappa, Mean Absolute Deviation, Method Consistency Test, and Total Rank Difference, were applied to assess the performance of the five HSID methods at crash estimation. In terms of accumulated results which combined all crash types, the model with spatial random effects consistently outperformed the other competing models with a significant margin. However, the inclusion of spatial random effect in temporal models fell short of attaining the expected results. The overall observation from the model fitness and validation results failed to highlight any correlation among better model fitness and superior crash estimation.

Multivariate poisson lognormal modeling of crashes by type and severity on rural two lane highways

In an effort to improve traffic safety, there has been considerable interest in estimating crash prediction models and identifying factors contributing to crashes. To account for crash frequency variations among crash types and severities, crash prediction models have been estimated by type and severity. The univariate crash count models have been used by researchers to estimate crashes by crash type or severity, in which the crash counts by type or severity are assumed to be independent of one another and modelled separately. When considering crash types and severities simultaneously, this may neglect the potential correlations between crash counts due to the presence of shared unobserved factors across crash types or severities for a specific roadway intersection or segment, and might lead to biased parameter estimation and reduce model accuracy. The focus on this study is to estimate crashes by both crash type and crash severity using the Integrated Nested Laplace Approximation (INLA) Multivariate Poisson Lognormal (MVPLN) model, and identify the different effects of contributing factors on different crash type and severity counts on rural two-lane highways. The INLA MVPLN model can simultaneously model crash counts by crash type and crash severity by accounting for the potential correlations among them and significantly decreases the computational time compared with a fully Bayesian fitting of the MVPLN model using Markov Chain Monte Carlo (MCMC) method. This paper describes estimation of MVPLN models for three-way stop controlled (3ST) intersections, four-way stop controlled (4ST) intersections, four-way signalized (4SG) intersections, and roadway segments on rural two-lane highways. Annual Average Daily traffic (AADT) and variables describing roadway conditions (including presence of lighting, presence of left-turn/right-turn lane, lane width and shoulder width) were used as predictors. A Univariate Poisson Lognormal (UPLN) was estimated by crash type and severity for each highway facility, and their prediction results are compared with the MVPLN model based on the Average Predicted Mean Absolute Error (APMAE) statistic. A UPLN model for total crashes was also estimated to compare the coefficients of contributing factors with the models that estimate crashes by crash type and severity. The model coefficient estimates show that the signs of coefficients for presence of left-turn lane, presence of right-turn lane, land width and speed limit are different across crash type or severity counts, which suggest that estimating crashes by crash type or severity might be more helpful in identifying crash contributing factors. The standard errors of covariates in the MVPLN model are slightly lower than the UPLN model when the covariates are statistically significant, and the crash counts by crash type and severity are significantly correlated. The model prediction comparisons illustrate that the MVPLN model outperforms the UPLN model in prediction accuracy. Therefore, when predicting crash counts by crash type and crash severity for rural two-lane highways, the MVPLN model should be considered to avoid estimation error and to account for the potential correlations among crash type counts and crash severity counts.

The effect of zonal factors in estimating crash risks by transportation modes: Motor vehicle, bicycle and pedestrian

OBJECTIVES

This paper aimed to (i) differentiate the effects of contributory factors on crash risks related to different transportation modes, i.e., motor vehicle, bicycle and pedestrian; (ii) explore the potential contribution of zone-level factors which are traditionally excluded or omitted, so as to track the source of heterogeneous effects of certain risk factors in crash-frequency models by different modes.

METHODS

Two analytical methods, i.e. negative binomial models (NB) and random parameters negative binomial models (RPNB), were employed to relate crash frequencies of different transportation modes to a variety of risk factors at intersections. Five years of crash data, traffic volume, geometric design as well as macroscopic variables at traffic analysis zone (TAZ) level for 279 intersections were used for analysis as a case study.

RESULTS

Among the findings are: (1) the sets of significant variables in crash-frequency analysis differed for different transportation modes; (2) omission of macroscopic variables would result in biased parameters estimation and incorrect inferences; (3) the zonal factors (macroscopic factors) considered played a more important role in elevating the model performance for non-motorized than motor-vehicle crashes; (4) a relatively smaller buffer width to extract macroscopic factors surrounding the intersection yielded better estimations.

Macro-level safety analysis of pedestrian crashes in Shanghai, China

Pedestrian safety has become one of the most important issues in the field of traffic safety. This study aims at investigating the association between pedestrian crash frequency and various predictor variables including roadway, socio-economic, and land-use features. The relationships were modeled using the data from 263 Traffic Analysis Zones (TAZs) within the urban area of Shanghai - the largest city in China. Since spatial correlation exists among the zonal-level data, Bayesian Conditional Autoregressive (CAR) models with seven different spatial weight features (i.e. (a) 0-1 first order, adjacency-based, (b) common boundary-length-based, (c) geometric centroid-distance-based, (d) crash-weighted centroid-distance-based, (e) land use type, adjacency-based, (f) land use intensity, adjacency-based, and (g) geometric centroid-distance-order) were developed to characterize the spatial correlations among TAZs. Model results indicated that the geometric centroid-distance-order spatial weight feature, which was introduced in macro-level safety analysis for the first time, outperformed all the other spatial weight features. Population was used as the surrogate for pedestrian exposure, and had a positive effect on pedestrian crashes. Other significant factors included length of major arterials, length of minor arterials, road density, average intersection spacing, percentage of 3-legged intersections, and area of TAZ. Pedestrian crashes were higher in TAZs with medium land use intensity than in TAZs with low and high land use intensity. Thus, higher priority should be given to TAZs with medium land use intensity to improve pedestrian safety. Overall, these findings can help transportation planners and managers understand the characteristics of pedestrian crashes and improve pedestrian safety.

Latent segmentation based count models: Analysis of bicycle safety in Montreal and Toronto

The study contributes to literature on bicycle safety by building on the traditional count regression models to investigate factors affecting bicycle crashes at the Traffic Analysis Zone (TAZ) level. TAZ is a traffic related geographic entity which is most frequently used as spatial unit for macroscopic crash risk analysis. In conventional count models, the impact of exogenous factors is restricted to be the same across the entire region. However, it is possible that the influence of exogenous factors might vary across different TAZs. To accommodate for the potential variation in the impact of exogenous factors we formulate latent segmentation based count models. Specifically, we formulate and estimate latent segmentation based Poisson (LP) and latent segmentation based Negative Binomial (LNB) models to study bicycle crash counts. In our latent segmentation approach, we allow for more than two segments and also consider a large set of variables in segmentation and segment specific models. The formulated models are estimated using bicycle-motor vehicle crash data from the Island of Montreal and City of Toronto for the years 2006 through 2010. The TAZ level variables considered in our analysis include accessibility measures, exposure measures, sociodemographic characteristics, socioeconomic characteristics, road network characteristics and built environment. A policy analysis is also conducted to illustrate the applicability of the proposed model for planning purposes. This macro-level research would assist decision makers, transportation officials and community planners to make informed decisions to proactively improve bicycle safety - a prerequisite to promoting a culture of active transportation.

Macroscopic modeling of pedestrian and bicycle crashes: A cross-comparison of estimation methods

The paper presents a cross-comparison of different estimation methods to model pedestrian and bicycle crashes. The study contributes to macro level safety studies by providing further methodological and empirical evidence on the various factors that influence the frequency of pedestrian and bicycle crashes at the planning level. Random parameter negative binomial (RPNB) models are estimated to explore the effects of various planning factors associated with total, serious injury and minor injury crashes while accounting for unobserved heterogeneity. Results of the RPNB models were compared with the results of a non-spatial negative binomial (NB) model and a Poisson-Gamma-CAR model. Key findings are, (1) the RPNB model performed best with the lowest mean absolute deviation, mean squared predicted error and Akaiki information criterion measures and (2) signs of estimated parameters are consistent if these variables are significant in models with the same response variables. We found that vehicle kilometers traveled (VKT), population, percentage of commuters cycling or walking to work, and percentage of households without motor vehicles have a significant and positive correlation with the number of pedestrian and bicycle crashes. Mixed land use is also found to have a positive association with the number of pedestrian and bicycle crashes. Results have planning and policy implications aimed at encouraging the use of sustainable modes of transportation while ensuring the safety of pedestrians and cyclist.