Crash Severity Distributions for Life-Cycle Benefit-Cost Analysis of Safety-Related Improvements on Utah Roadways
Conor J Seat, Avenue ConsultantsShow Abstract
B Wyatt Clegg, Los Alamos National Laboratory
Mitsuru Saito, Saito Research and Consulting
Grant Schultz, Brigham Young University
The Utah Department of Transportation (UDOT) developed life-cycle benefit-cost analysis spreadsheets that allow engineers and analysts to evaluate multiple countermeasures. The spreadsheets have included the functionality to evaluate a roadway based on the 11 facility types from the Highway Safety Manual (HSM) with the use of crash severity distributions. The HSM recommends that local agencies should develop crash severity distributions based on their local crash data. UDOT had only one severity distribution for all the facility types. The primary objective of this research was to utilize available roadway characteristic and crash data to develop crash severity distributions for the 11 facility types in the HSM. This objective was accomplished by segmenting the roadway data based on homogeneity and developing statistical models to determine the distributions. Due to insufficient data, the facility types 10 and 11 (freeway speed change lanes and freeway ramps, respectively) were excluded from the scope of this research. In order to accommodate more roadway segments for the analysis the facility type definitions were expanded. The statistical models that were developed for this research include multivariate regression, frequentist binomial regression, frequentist multinomial, and Bayesian multinomial regression models. A cross-validation study was conducted to determine the models that described the data the best. Bayesian Information Criterion, Deviance Information Criterion, and Root-Mean-Square Error values were used to compare model reliabilities. The Bayesian multinomial regression model was found to be the most effective model to describe the crash severity distributions for the nine facility types evaluated.
Safety Evaluation of Pedestrian Countdown Signals - Definitive Results from Two Cities in the United States
Raghavan Srinivasan, University of North Carolina, Chapel HillShow Abstract
Bo Lan, UNC Highway Safety Research Center
Daniel Carter, UNC Highway Safety Research Center
Sarah Smith, University of North Carolina
Bhagwant Persaud, Ryerson University
Kari Signor, UNC Highway Safety Research Center
Taha Saleem, UNC Highway Safety Research Center
The pedestrian countdown signals (PCS) treatment involves the display of a numerical countdown that shows how many seconds are left in the flashing DON’T WALK interval. Although many studies have attempted to evaluate the safety of PCS, the results have been inconsistent due to many reasons including inadequate samples, and the inability to control for possible bias due to regression to the mean, and exposure. This study performed a before-after empirical Bayes (EB) analysis using data from 115 treated intersections in Charlotte, North Carolina and 218 treated intersections in Philadelphia, Pennsylvania to evaluate the safety effects of PCS. The evaluation also included 136 reference intersections in Charlotte, and 597 reference intersections from Philadelphia. Following the implementation of PCS, total crashes decreased by approximately 8 percent and rear-end crashes decreased approximately 12 percent, and these reductions were statistically significant at the 95-percent confidence level. Pedestrian crashes decreased by about 9 percent and this reduction was statistically significant at the 90-percent confidence level. The economic analysis revealed a benefit-cost ratio of 23 with a low of 13 and a high of 32.
Estimating Safety with the Multiresolution HAAR Wavelet Method: Comparison with the Safety Performance Function Approach
Samer Katicha, Virginia Polytechnic Institute and State UniversityShow Abstract
John Khoury, Lebanese American University
Gerardo Flintsch, Virginia Polytechnic Institute and State University
This paper presents the multiresolution Haar wavelet (MHW) approach to estimate the expected number of crashes at roadway sections. The MHW is similar to kernel density estimation or sliding window (moving average) estimation with the additional benefit of being spatially adaptive. This means that the window size (or bandwidth) can be different at different locations allowing different averaging length (or amount of smoothing). Furthermore, the optimal window size (at each location) is determined solely based on the data. The MHW approach is compared to the current state of the practice Safety Performance Function (SPF) approach on the entire State of Virginia interstate network. The results of the comparison showed that the MHW better predicts future crashes. The MHW approach, unlike the SPF approach, does not require any data other than the crash counts to estimate the expected number of crashes. This makes it easy to implement and simple to use. We have implemented the approach in an Excel spreadsheet that is freely available for use.
Empirical Bayes Safety Evaluation of a Modified Right Turn Lane Design at Intersections
Kerrie Schattler, Bradley UniversityShow Abstract
Trevor Hanson, Illinois Department of Transportation
From 2006 to 2014, ten right-turn approaches in Peoria, Illinois, were reconstructed with a modified right-turn lane design. The major purpose of the modified design was to improve the line of sight for passenger vehicles turning right, while also accommodating semi-tractor trailer trucks. While the actual changes varied among intersections, the final result at each was an improvement to the approach angle for right-turning vehicles stopped to view cross traffic. The authors of this paper were involved in a two-part research study in which they (1) performed site-specific effectiveness evaluations of the modified sites, and (2) conducted a crash causation analysis of right-turn crashes at 116 sites in Illinois to identify geometric design variables that correlate with right-turn crashes. In the site-specific evaluation, traffic crash–based safety analyses were performed to assess the safety performance of the modified right-turn lane design using the empirical Bayes method. Statistically significant reductions in crashes were observed at the subject approaches (59.0%) after the modifications were made. The results of the crash causation analyses revealed significantly higher right-turn crashes for approaches with head-turn angles greater than 140, right-turn angles less than 45, and acute intersection angle less than 75. Recommendations on the characteristics of good candidate sites for the installation of the modified right-turn lane design in Illinois are presented in this paper. This research received a 2017 AASHTO Research Advisory Committee High-Value Research Maintenance and Safety Project designation.
Safety Performance of Median U-Turn Conversions in Michigan
Jonathan J. Kay, Michigan State UniversityShow Abstract
Timothy Gates, Michigan State University
Meghna Chakraborty, Michigan State University
Steven Stapleton, Michigan State University
Alternative intersection designs can offer safety and operational benefits with potentially lower costs than conventional intersections when used in the proper setting. One such alternative intersection design that has been used extensively across Michigan for decades is the median U-turn (MUT), which accommodates left turns via a U-turn crossover within the median. This evaluation examined 28 stop-controlled MUT intersections as well as 100 signalized MUT intersections in order to help quantify the safety benefits of implementing MUTs. The percentage of angle collisions at stop-controlled intersections was substantially lower in the post-conversion period (5.7 percent) compared to the pre-conversion period (50.3 percent). The decrease in the proportion of angle collisions was offset by a notable increase in rear-end collisions (25.8 percent in the pre-conversion sites and 75.7 percent in the post-conversion sites). There was also a decrease in the proportion of head-on left-turn collisions in the post-conversion period (0.6 percent) compared to the pre-conversion period (2.8 percent) for signalized MUTs. Stop-controlled MUT intersections exhibited superior safety performance in terms of fatal and injury (FI) crashes over traditional intersections but experienced more property damage only (PDO) crashes as major approach volumes exceed 15,000 entering vehicles per day. Signalized MUT and traditional intersections performed similarly up to approximately 20,000 entering vehicles per day along the major approach, at which point MUTs experienced more FI crashes but fewer PDO crashes. Ultimately, fully-specified negative binomial regression models were developed to estimate the FI and PDO crash frequencies for both stop-controlled and signalized MUT intersections.
Predicting real-time crash risk on urban expressways using Recurrent Neural Network
Kui Yang, Tongji UniversityShow Abstract
Xuesong Wang, Tongji University
Mohammed Quddus, Loughborough University
Rongjie Yu, Tongji University
Real-time crash risk prediction is an important area of research that focuses on identifying hazardous traffic conditions as part of proactive traffic safety management. Although there is a plethora of classification algorithms applied to predict an unsafe traffic condition, they cannot capture spatio-temporal variability in traffic dynamics and are not transferable. In this paper, a state-of-the-art approach based on supervised machine learning - recurrent neural network (RNN) is developed and implemented to address the challenges of predictability of crash risk models. In relation to existing techniques, one of the unique features of RNN is to employ feedback loops where the output from each of the steps is feedback to the RNN to affect the outcome of the current step. It also has a self-updating ability of model parameter via a time sequence, which is helpful for the model adaptability by overcoming the spatial-temporal variability of traffic dynamics. Historical crash data and real-time traffic data from Shanghai Urban Expressway System were matched and split into a training dataset and a test dataset: the training dataset was designed in the matched case-control study and used to develop the crash risk prediction models; the test dataset was a full set including all cases and employed to evaluate the performance of the models via the area under ROC curve (AUC) and sensitivity. In addition, the prediction results were compared with those given by other frequently used classification algorithms, including logistic regression and support vector machine (SVM). The results proved that RNN had a better prediction performance. It could increase the crash prediction accuracy by an average of 13.3% and 7.9% compared to the SVM and logistic regression model, respectively. Furthermore, the optimal ratio of crashes to non-crashes has found to be 1:4 for the model development.
Comparison of Calibration Methods for Improving the Transferability of Safety Performance Functions
Xuesong Wang, Tongji UniversityShow Abstract
Dongjie Tang, Tongji University
Saijun Pei, Tongji University
Safety performance functions (SPFs) are critical for traffic safety management. They have been applied for identifying significant risk factors, estimating crash frequencies, and screening potentially hazardous locations. Since SPFs proposed by Highway Safety Manual (HSM) are developed based on certain states in the United States, regions without jurisdiction-specific SPFs need model calibrations for the localization of SPFs. The main objective of this study is to compare the typical calibration methods that used in the literature and identify the appropriate ones. Random effects Negative Binomial (NB) models were established for urban arterials in Shanghai and Guangzhou during peak hours and off-peak hours separately. Four calibration methods, including the calibration factor, empirical Bayes (EB) method, K Nearest Neighbor (KNN) regression method, and pooled data, were applied. The performance in improving model transferability was measured by transfer index and the adaptability to insufficient data was assessed by necessary data collected for each method. Based on the modeling results, pooled data approach that composed of the entire Shanghai dataset and 50% proportion of the Guangzhou dataset provides the best performance. And EB method and KNN regression method are preferable to the calibration factor. Furthermore, modeling and calibrating for different time periods should be considered when average speed is incorporated in the model.
Transferability of Safety Performance Functions and Hotspot Identification for Freeways of the United States and China
Mingjie Feng, Tongji UniversityShow Abstract
Xuesong Wang, Tongji University
Jaeyoung Lee, University of Central Florida
Mohamed Abdel-Aty, University of Central Florida
Safety performance functions have been a vital tool in traffic safety evaluation including finding contributing factors to crashes, identifying hotspots, and assessing safety effects of countermeasures. In the United States, the Highway Safety Manual has provided a series of SPFs for a variety of road facilities. Due to the limited availability of traffic data in many regions, the transferability of SPFs has been an important topic in the traffic safety field and several studies have been conducted to evaluate the transferability of SPFs. Nevertheless, no study has investigated the international transferability of freeway SPFs and the consistency in hotspot identification has been rarely investigated. Using data from Shanghai and three U.S. states: Florida, Texas and New York, we examine the transferability of freeway SPFs between China and the United States. SPFs were developed separately for total crashes, single-vehicle and multi-vehicle crashes. According to the estimated transfer indices (TIs), all Shanghai SPFs are reasonably transferable to U.S. data, but no U.S. SPFs can be transferred to Shanghai data. The modeling results suggest that this discrepancy might be due to the greater sensitivity of U.S. crashes to annual average daily traffic (AADT), and the likelihood that Shanghai crashes are influenced by factors other than segment length and AADT. The method consistency test (MCT) shows that both Shanghai and U.S. SPFs can identify quite consistent hotspots in the other country. The findings from study are expected to be a good reference for researchers and practitioners in developing countries who want to understand the transferability and applicability of SPFs in the international context.
Comparing HSM Calibrated and Local Developed SPFs for Rural Two Way Intersections
Deo Chimba, Tennessee State UniversityShow Abstract
Chacha Wambura, Tennessee State University
Asad Khattak, University of Tennessee, Knoxville
Jim Waters, Tennessee Department of Transportation
This study developed HSM calibration factors for Rural Two-Lane, Two-Way Intersections in Tennessee for three leg stop controlled intersections (3ST), four leg stop controlled intersections (4ST), and four leg signalized intersections (4SG). Utilizing crash data from 2011 to 2015, and by applying crash modification factors (CMFs), corresponding statewide and regional calibrations factors for 2010 HSM Safety Performance Functions (SPFs) were developed as 0.633 for 3ST intersections, 0.980 for 4ST intersections and 0.730 for 4SG intersections. The calibration factors changed slightly without applying CMFs (using HSM default values) as 0.514 for 3ST, 0.747 for 3ST and 0.461 for 4SG. Overall, the developed statewide calibration factors for 3ST, 4ST and 4SG intersections were less than 1.0 indicating that Tennessee has fewer crashes than those predicted using 2010 HSM SPFs. Comparing to findings from other states, the Tennessee-developed Rural Two-Lane, Two-Way intersections calibration factors are comparable to, but slightly higher than, those developed in most states. Using Tennessee crash and traffic data, the study developed local safety performance functions (SPFs) reflecting those developed in 2010 HSM. The sign and magnitude of the model constants and variable coefficients of the locally developed Tennessee SPFs were very close to those in 2010 HSM. For three leg stop controlled intersection (3ST), the Tennessee-developed SPF has a constant term of -9.25 (-9.86 in HSM), the major road AADT coefficient of 0.71 (0.76 in HSM) and the minor road AADT coefficient is 0.41 (0.49 in HSM).
Rural Intersection SPFS – Slip Lanes and Seagulls
Shane Alan Turner, Stantec Consulting Services, Inc.Show Abstract
Graham Wood, Consultant
Fergus Tate, NZ Transport Agency
In New Zealand the majority of rural intersection fatal and serious crashes occur at rural priority T-intersections. While most intersections have a standard layout higher volume intersections often have alternative layouts that include auxiliary lanes and/or channelization. Two alternative intersection layouts are reviewed in this research: 'priority controlled seagull (channelized) intersections' and ‘intersections with slip lanes’. Seagull intersections are used on roads to reduce traffic delays. However, some do experience high crash rates. Slip lanes (left turn for left-hand drive and right turn for right-hand drive) allow turning traffic to move clear of the through traffic before decelerating. Although there is debate about the safety problems that occur at Seagull intersections and slip lanes there has been very little research to quantify the safety impact of different layouts. In this study, safety performance functions have been developed for standard rural T-intersections and the two alternative intersection layouts for the key crash types. A point of difference in the modelling is that a design index has been developed for road layout variables, rather than including each layout variable separately in the models, along with exposure and speed variables.
Examining the Safety Performance and Injury Severity Characteristics of Rural County Roadways
Srinivas Geedipally, Texas A&M Transportation InstituteShow Abstract
Timothy Gates, Michigan State University
Steven Stapleton, Michigan State University
Anthony Ingle, Michigan State University
Raul Avelar, Texas A&M Transportation Institute
Much of the earlier work on rural safety focused on state-maintained roadways and little is known about the safety performance of low volume county-maintained roads. This study involved the estimation of safety performance for rural county roadways (paved and gravel). This was accomplished through the development of safety performance functions (SPFs) to estimate the number of annual crashes at a given highway segment, crash modification factors to determine the impacts associated with various roadway and geometric characteristics, and severity distribution functions (SDFs) to predict the crash severity. County road segment data were collected across a sample of 30 counties representing all regions of Michigan. Due to overwhelming proportion of deer crashes, only non-deer related crashes were considered. To minimize the influence of variability among counties, the random effect negative binomial model was used to develop SPFs. In addition, a multinomial logit model was used to develop SDFs. Paved county roadways showed approximately double the crash occurrence rate of typical state-maintained two-lane rural highways, while gravel roadways showed a substantially greater crash occurrence rate than paved county roadways across the equivalent range of traffic volumes. The economic analysis showed that it is beneficial to pave a gravel road when the traffic volume is greater than 600 vehicles per day. The random effect variable is significant in all the calibrated models which shows that there is a considerable variability among counties that cannot be captured with the available variables. Not considering the random effects will result in biased estimation of crashes.
Developing Rural Four Lane Freeway Crash Prediction Models Using Hourly Flow Parameters
Nancy Dutta, University of VirginiaShow Abstract
Michael Fontaine, Virginia Transportation Research Council
Most past crash prediction research has examined the relationship between crashes, traffic volumes, and other factors at the annual level, due to the rare and random nature of crash occurrence and data availability. For example, the current functional form of safety performance functions in the Highway Safety Manual is based on annual average daily traffic (AADT). Less attention has been given to explicitly modeling the safety effects of vehicle density, volume-to-capacity ratio, and speed distribution at a sub-daily level. This research used continuous count station data from 4 lane rural freeway segments in Virginia and developed crash prediction models using traffic and geometric information provided at hourly aggregation intervals. The results showed that using average hourly volume along with average speed and selected geometric variables improved predictions compared to models that used AADT. When comparing an AADT-based model to an average hourly volume model, the mean absolute prediction error improved by 15% for total crashes. This value improved by 20% after including geometric variables, and by 30% after adding speed to the volume and geometry model. Similar improvements were observed for injury crashes. These results provide a strong indication that crash predictions could be improved using more disaggregate data and justifies further exploration of these relationships using larger datasets and other statistical methodologies. The findings from this research also indicate that inclusion of quality of flow variables, like speed, could create improvements in the quality of crash prediction models.
Developing Safety Performance Functions for North Carolina Low-Volume Roadways
Subasish Das, Texas A&M Transportation InstituteShow Abstract
Ioannis Tsapakis, Texas A&M Transportation Institute
Songjukta Datta, Texas A&M Transportation Institute
The Moving Ahead for Progress in the 21st Century Act (MAP-21) mandates for a Highway Safety Improvement Program (HSIP) for all states that “emphasizes a data-driven, strategic approach to improving highway safety on all public roads that focuses on performance”. To determine the predicted crashes on a specific roadway facility, the most convenient and widely used tool is the first edition of Highway Safety Manual (HSM), which provides predictive models (known as safety performance functions, SPFs) of crash frequencies for different roadways. Low-volume roads are defined as roads located in rural areas with daily traffic volumes of less than or equal to 400 vehicles per day (vpd). Low-volume roadways cover a significant portion of the roadways in the U.S. While much work has been done to develop SPFs for high-volume roads, less effort has been devoted to low-volume road safety issues. This study used 2013-2017 traffic count, roadway network, and crash data to develop six SPFs for three low-volume roadways, which can be used to predict total crashes, fatal, and injury crashes. These SPFs will provide state and local agencies with the means to quantify safety impacts on low-volume roadway networks.
The Safety Implications of the Conversion of Continuous Green T-Intersections Back to Conventional T-Intersections
Jaeyoung Lee, University of Central FloridaShow Abstract
Mohamed Abdel-Aty, University of Central Florida
Alan El-Urfali, Florida Department of Transportation
A continuous green T-intersection (CGT) is an innovative intersection that could improve the through traffic capacity by allowing major-leg vehicles on the top side of T-intersection to pass through without stopping. Recently, traffic engineers decided to stop CGT operations at several T-intersections in Florida because of traffic safety concerns, conversion to four-legged intersection, pedestrians’ demand, and non-compliance with the latest Manual on Uniform Traffic Control Devices. In this study, safety effects of recent conversions of CGTs back to conventional T-intersections in Florida are explored. A before-and-after study with the comparison group method are adopted. The results indicate significant reductions in total, fatal-and-injury, rear-end, and CGT-related crashes by 40% to 60% after the conversion. In order to validate the results, a cross-sectional analysis was conducted with new data from four states. The results are consistent for total, fatal-and-injury, and CGT-related crashes with those from the before-and-after study. The results also show that crashes at CGTs could be minimized by providing a physical separation between the acceleration lane for the merging vehicles and the CGT through lane, along with other factors. Because Florida’s T-intersections that were converted back to the conventional design from CGT had no physical separation, and the results showed a significant safety improvement after the conversion. Therefore, the decision to stop CGT operations at the Florida’s study sites was supported from the safety aspect. The study concluded that safety at CGTs could be a concern compared to non-CGTs; however, it could be significantly improved by providing appropriate countermeasures.
Safety Effects of Pavement Roughness for Freeways: A Comparative Analysis of Interstate Highways in Five States
Jaeyoung Lee, University of Central FloridaShow Abstract
Mohamed Abdel-Aty, University of Central Florida
Traffic crashes occur usually because of the contribution of a combination of human, roadway/ environment, and vehicle factors. Pavement condition is closely related to the three factors as it is one of the most important roadway/environment factors and it affects driving behavior and vehicle performance at the same time. Previous studies have shown that pavement conditions have played an important role in safety. In this study, we develop four different safety performance functions to evaluate the effect of pavement roughness, which is measured by the International Roughness Index (IRI), on the number of crashes using the interstate highway data from five states representing different geographical and weather regions in the US: Arizona, Colorado, Florida, Maryland, and Michigan. The modeling results identify many significant variables including traffic volume and proportion of trucks, through lane count, shoulder type, median width, high-occupancy vehicle lane operation and HOV lane count, speed limit, area type along with IRI-related factors. The results indicate that the increased IRI (deterioration of pavement quality) contribute to larger numbers of total crashes. On interstate highways with speed limit of 70 mph and higher, the effects of IRI are relatively smaller. On the other hand, the effects of IRI increase with a larger traffic volume. Based on the modeling results, seven crash modification functions of IRI values by crash type and speed limit were estimated. The findings from this study are expected to be useful for both pavement and safety engineers to understand the relationship between IRI and safety on freeways.
International Transferability of Macro-Level Safety Performance Functions: A Case Study of the United States and Italy
Jaeyoung Lee, University of Central FloridaShow Abstract
Mohamed Abdel-Aty, University of Central Florida
Maria Rosaria De Blasiis, Roma Tre University
Xuesong Wang, Tongji University
Ilaria Mattei, Ferrovie dello Stato Italiano
Safety performance functions (SPFs) or crash prediction models have played an important role in identifying the contributing factors of crashes, predicting crash counts, identifying hotspots, etc. Because it needs a lot of time and efforts to estimate a SPF, previous studies have evaluated if a SPF could be applied to data from other regions, i.e., transferability. Although many efforts have been made for micro-level SPF transferability, not many have been done for macro-level SPF transferability. Transferability analysis of macro-level SPFs in the international context, especially between western countries, has not been conducted. Therefore, we evaluate the transferability of SPFs of several states in the United States (i.e., Illinois, Florida, and Colorado) and Italy in this study. The SPFs were developed using data from counties in the United States and provincias in Italy, and the results show that there are multiple common significant variables across the countries. Subsequently, transfer indexes are calculated between the developed SPFs, and the indexes show that the Italian SPFs for total and bicyclists crashes are transferable to U.S. data after calibration factors are applied while the U.S. total and bicycle SPFs, except for the Colorado SPF, cannot be transferred to the Italian data. On the other hand, none of the developed pedestrian SPFs are transferable to other countries. This paper provides insights into the applicability of macro-level SPFs between the U.S. and Italy, and shows a good potential of international SPFs’ transferability. Nevertheless, further investigation is needed for the SPF transferability between more countries.
Functional Forms of the Negative Binomial Models in Safety Performance Functions for Rural Two-Lane Intersections
Kai Wang, University of ConnecticutShow Abstract
Shanshan Zhao, Connecticut Transportation Safety Research Center
Eric Jackson, Connecticut Transportation Safety Research Center
Safety Performance Functions (SPFs) play a prominent role in estimating intersection crashes, and identifying the sites with the highest potential for safety improvement. To maximize the crash prediction accuracy, this paper describes the application of different functional forms of the negative binomial models (i.e. NB-1, NB-2 and NB-P) in estimating safety performance functions by crash type for rural two-lane intersections. Crash types were aggregated into same-direction, opposite-direction, intersecting-direction and single-vehicle crashes. Major and minor road AADT were used as predictors in the SPF estimation. The over-dispersion parameter of the NB models was estimated by the AADT to account for the crash data heterogeneity. The models were compared based on both the model estimation goodness-of-fit and the prediction performance. The model goodness-of-fit indicates that the NB-P model outperforms the NB-1 and NB-2 for most crash types and intersection types, by providing a flexible variance structure to the NB approaches. The parameterization of the over-dispersion factor verifies that the over-dispersion parameter of the NB models highly depends on how the variance structure is defined in the model, and the over-dispersion parameter is shown to vary among different crashes and can be estimated using both the major and minor road AADT at rural two-lane intersections. The prediction performance comparison illustrates that the NB-P model slightly improves the crash prediction accuracy compared with the other two models. Therefore, the NB-P model with parameterized over-dispersion factor is recommended to provide more unbiased parameter estimates when estimating SPFs by crash type for rural two-lane intersections.
A Novel Approach for Estimating Crash Modification Factors: Jointly Modeling Crash Counts and Time Intervals between Crashes
Lingtao Wu, Texas A&M Transportation InstituteShow Abstract
Yi Meng, Texas A&M University
Xiaoqiang Kong, Texas A&M Transportation Institute
Yajie Zou, Tongji University
Crash modification factors (CMFs) play important roles in roadway safety management. Safety analysts have proposed various methods for developing CMFs, and nearly all of them are crash count-based. Intervals between crashes are not considered, while survival theory has been widely used in other fields. The objective of this study is to incorporate survival models into the estimation of CMFs and to examine if it increases estimation accuracy. To accomplish the objective, this study proposed a joint modeling approach, simultaneously analyzing crash counts and time intervals between crashes, for estimating CMFs. In order to assess the performance, this study developed CMFs for a dummy treatment at 90 sites on rural two-lane highways in Texas, with the standard empirical Bayes (EB) method and the joint model, separately. The findings are very interesting: (1) The standard EB method tends to over-estimate the CMFs for the treatment, and under-estimate the standard error of the CMFs. Most of the cases, the results are biased; (2) The CMFs developed with the joint model have greater standard errors, but their values are closer to the true effects, which is more realistic; (3) Temporary instability in traffic crashes are also observed in this study. Increasing the duration of study period does not always increase the accuracy of CMF estimates. Roadway agencies are encouraged to deploy the joint model for dynamical monitoring safety effects of treatments by flexible feedback.
Seasonal Crash Prediction Model for Urban Signalized Intersections: Wisconsin Southeast Region
Boris Claros, University of Wisconsin, MadisonShow Abstract
Madhav Chitturi, University of Wisconsin, Madison
Glenn Vorhes, University of Wisconsin, Madison
Andrea Bill, University of Wisconsin, Madison
David Noyce, University of Wisconsin, Madison
The Highway Safety Manual (HSM) provides methods to quantitatively evaluate safety for a vast range of roadway transportation facilities. The Negative Binomial has been traditionally used for modeling crashes (i.e. crashes per year). Highly aggregated cross-sectional data omits natural time dependent variations leading to important loss of information and introducing error in model predictions. Furthermore, traffic conditions and weather vary over time and space. An alternative approach with seasonal crash estimates is proposed in this paper. Local crashes, traffic, geometry, signal type, and weather data of urban signalized intersections in the Southeast region of Wisconsin were used. Four seasons were considered: Winter, Spring, Summer, and Fall. The Negative Multinomial was used for modeling to account for seasonal variations. The functional form for each predictor variable was optimized. Measures of log-likelihood, inverse overdispersion, cumulative residual (CURE) plots, and Akaike information criterion (AIC) showed adequate model prediction accuracy. Seasonal estimates for fatal and injury (FI) crashes showed that during the Spring season, crash estimates were the lowest and during the Summer were the highest. In contrast, model crash estimates for Property Damage Only (PDO) crashes peaked during the Winter season and remain below annual estimates for the rest of the seasons. Magnitude of fluctuations and accuracy of crash estimates contribute to managerial decisions and allocation of resources for the implementation of treatments, safety programs, minimize safety impacts, and reduce risk of crashes—contributing to reduction in costs associated with crashes, property damage, maintenance, and emergency services.
Modeling Safety Performance of the New Super DDI Design in terms of Vehicular Traffic and Pedestrian
Amirarsalan Mehrara Molan, University of WyomingShow Abstract
Joseph Hummer, North Carolina Department of Transportation
Khaled Ksaibati, University of Wyoming
Most existing interchanges in the United States were built more than 50 years ago based on old design policies. Many of these designs are not consistent with current traffic and pedestrian demands anymore. This paper models the safety performance of a new design called a super diverging diamond interchange (super DDI) using VISSIM simulation and the Surrogate Safety Assessment Model (SSAM). Six other interchange designs were also considered for comparing to the new super DDI design. Also, the same number of tests were conducted to evaluate pedestrian performance of the designs considered in this study. Based on the results, the super DDI showed a high potential either in terms of traffic safety and pedestrian safety. In comparison to other designs, the super DDI had the minimum number of simulated conflicts as well as the lowest mean speed and time to collision (TTC) of simulated conflicts. Reviewing the geometry of the super DDI, lower traffic volumes involved in each conflict point should be one of the main reasons for the promising traffic safety performance of the design. Regarding pedestrian performance, there is no free-flowing conflict between vehicles and pedestrians in a super DDI. Therefore, pedestrian paths of the super DDI are predicted to be safer than the paths in a typical DDI design.
Planning-Level Crash Modification Factors and Data-Driven Safety Approaches to Long Range Transportation Planning
Jeffrey Gooch, VHBShow Abstract
Ian Hamilton, VHB
John W Moore, Kentucky Transportation Cabinet
Elissa Goughnour, VHB
In 2016, FHWA published the Safety Performance Management Final Rule requiring State Departments of Transportation (DOT) and Metropolitan Planning Organizations (MPO) to set annual targets for their Highway Safety Improvement Program (HSIP), focused on reducing fatalities and serious injuries on their roadways. Agencies usually apply a short-term strategy, analyzing alternatives for projects being considered in the near-term. However, legislation incentivizes agencies to incorporate safety performance into long-term project programming prioritization. Some State agencies have adopted programs which take a data-driven approach to mid- and long-term project programming. The North Carolina Department of Transportation (NCDOT), Virginia Department of Transportation (VDOT), and Kentucky Transportation Cabinet (KYTC) have adopted programs to quantify potential performance and prioritize candidate projects; the scores consider various project characteristics and estimate future benefits, such as congestion, operations, and safety. The three agencies developed planning-level crash modification factors (CMF) to standardize the anticipated safety benefit of a project. These planning-level CMFs are developed for common project types considered within the agency. Two strategies have been used for estimating planning-level CMFs: 1) a project-level, statistical before and after treatment evaluation of crash data, and 2) identifying high-quality CMFs from the FHWA CMF Clearinghouse for treatments usually included in common project types. Planning-level CMFs have been adopted because the specific design details usually required to apply CMFs are typically not available during the project programming stage. Since planning-level CMFs are a relatively new approach to incorporating safety into longer-term project programming, their performance will need to be monitored in the future.
Comparison of the Highway Safety Manual Predictive Method with Jurisdiction-Specific Safety Performance Functions and Effects of Geometric Design Consistency
David Llopis Castello, Universitat Politècnica de ValènciaShow Abstract
Daniel J Findley, North Carolina State University
García Alfredo, Universitat Politècnica de València
Road safety is a major public health concern in our society. Effective road design and accurate safety analyses must be a component of programs focused on reducing and eliminating roadway injuries and deaths. Various methodologies exist to determine the expected number of crashes on rural two-lane, two-way roadway segments with a goal of improving road safety. This research compares different procedures which allow for the estimation of the number of crashes on entire homogeneous road segments. In this effort, a total of 27 two-lane rural road sections located in North Carolina were considered, resulting in 59 homogeneous road segments composed of 350 horizontal curves and 375 tangents along 150 km (90 miles) of road. Four methods were applied to the selected roadways: the HSM predictive method, two jurisdiction-specific Safety Performance Functions (SPFs), and a SPF which includes a consistency parameter. This research found that the use of SPFs which incorporate a consistency parameter allows practitioners and highway engineers to consider human factor impacts on road safety assessment. The use of a consistency parameter can also simplify the crash estimation process. Analysis methods which only included local geometric variables provided unreliable results due to the calibration of only the specific road elements instead of their relationship with other road elements along homogeneous road segments.
IMPACT EVALUATION OF WIRE ROPE INSTALLATION ON TWO-WAY TWO-LANE EXPRESSWAYS
Jian Xing, Nippon Expressway Research InstituteShow Abstract
Tadahisa Muramatsu, Central Nippon Expressway Company Limited
Hidenori Goto, Oriental Consultants Co., Ltd.
Daisuke Yamaguchi, Oriental Consultants Co., Ltd.
In order to develop a road network under budget and short time constraints, high standard arterial expressways in Japan incorporate dual four-lane roads with provisional two-way two-lane (hereinafter referred to as TWTL) sections in areas where the traffic volume is low. In the majority of TWTL sections, rubber poles are used to separate inbound and outbound lanes, but the safety risks, including head-on collisions resulting from lane departures of vehicles into opposite lane, are considered problematic. In order to urgently address the situation, wire ropes are now being installed to trial road sections to verify their effectiveness in preventing head-on collisions of vehicles. This paper conducted a multifaceted study and evaluation of the effectiveness of wire ropes on TWTL sections, including drivability and maintainability in addition to effectiveness in preventing accidents.
Integrated Approach to the Network Screening of Urban Intersections
Bahar Dadashova, Texas A&M Transportation InstituteShow Abstract
Karen Dixon, Texas A&M Transportation Institute
Ioannis Tsapakis, Texas A&M Transportation Institute
Jing Li, Texas A&M Transportation Institute
As the first and one of the most important steps of Highway Safety Improvement Program (HSIP), network screening aims to identify sites with the highest potential for improvement. Network screening is not a trivial process and depends on several factors such as crash frequency and severity, traffic volume and roadway characteristics, and crash history of similar sites. The reliability of network screening is based on the safety performance measure selected for conducting the analysis. In this paper, the authors propose an integrated approach that incorporates a weighted ranking to rank the sites with higher potential for improvement. The results of the pilot study show that the proposed methodology is more reliable than using individual performance measures and could be implemented by transportation agencies that identify highway safety improvement projects.
Transferability and Calibration of Highway Safety Manual Safety Performance Function for Two Lane Highways in Brazil
Karla Cristina Rodrigues Silva, Centro Federal de Educação Tecnológica de Minas GeraisShow Abstract
Antonio Clovis Pinto Ferraz, USP
The present study focused on evaluating HSM crash prediction model for two lane highways on Brazilian conditions. Also, the transferability of the model was considered, specifically by means of a comparison between Brazil and HSM conditions. The analysis of two lane highways crash prediction models was promising when the road characteristics were well known and there was not much difference from base conditions. This conclusion was attained regarding the comparison of results for all segments, non-curved segments and curved segments, confirming that a transferred model can be used with caution. Finally, there are many factors that could not be measured by these models and reflects road safety various condition. Even so, the study of crash predict models in Brazilian context could provide a better start point in safety road analysis.
Short Segment Statewide Screening of Midblock Crashes in South Carolina
Afshin Famili, Clemson UniversityShow Abstract
Wayne Sarasua, Clemson University
Adika Iqbal, Clemson University
Devesh Kumar, Clemson University
Jennifer Ogle, Clemson University
The AASHTO Highway Safety Manual (HSM) presents a variety of methods for quantitatively estimating crash frequency or severity at a variety of locations. The HSMpredictive methods require the roadway network to be divided into homogeneous segments and intersections, or sites populated with a series of attributes. It recommends a minimum segment length of 0.1 miles. This research focuses on segment lengths of less than 0.1 miles for statewide screening of midblock crash locations to identify site specific locations with high crash incidence. The paper makes an argument that many midblock crashes can be concentrated along a very short segment due to an undesirable characteristic of a specific site. The use of longer segments may “hide” the severity of a single location if the rest of the segment has few or no additional crashes. In actuality, this research does not divide sections of roads into short segments. Instead, a short window approach is used. The underlying road network is used to create a layer of segment polygons using GIS buffering. Crash data are then overlaid and aggregated to the segment polygons for further analysis. The paper makes a case for the use of short fixed segments to do statewide screening and how accurately geocoded crash data is key to its use. A comparison is made with a sliding window approach (Network Kernel Density). The benefits of using fixed segments is that they are much less complex then using the sliding window approach.
A National-Level Safety Evaluation of Diverging Diamond Interchanges
Timothy Scott Nye, North Carolina Department of TransportationShow Abstract
Christopher Cunningham, North Carolina State University
Elizabeth Byrom, North Carolina State University
A national-level safety evaluation of Diverging Diamond Interchanges (DDIs) in the United States was completed. This study aimed to update previous evaluations and to expand the treatment group size of previous studies to provide a more robust and reliable safety assessment of DDI deployments. For this particular treatment, it was determined that, of the observational before-and-after evaluation methodologies, the comparison group approach yields the best evaluation results. The naïve method can be influenced by outside factors that cannot be accounted for (weather, crash reporting tendencies, etc.). The empirical Bayes methods is unnecessary as DDIs are installed for operational benefits, meaning that risk of selection bias and regression-to-the-mean is minimal. This study recommends a total crashes CMF of 0.633 based on the comparison group analysis of 26 DDIs in 11 states. The comparison group method was also applied to a variety of crash variables for this study. Angle, rear-end, and sideswipe crashes were found to have CMFs of 0.441, 0.549, and 1.139, respectively. Fatal-and-injury crashes provided a CMF of 0.461. Daytime and nighttime crashes provided CMFs of 0.648 and 0.638, respectively.
Clustering the effects of traffic control type, functional class and spatial distributions to intersections traffic safety
Brionne Henderson, Tennessee State UniversityShow Abstract
Deo Chimba, Tennessee State University
This paper contributes to the literature by examining the effects of traffic control type, spatial distribution and functional class to the traffic safety at intersections. The goal is to correlate crash occurrences to intersection types based on the amount of traffic volume entering the intersection, functional classes of intersecting streets, traffic control type, and the location with respect to CBD areas. Using data from Davidson County in Nashville Tennessee, the study evaluated the intersections considering crashes occurring within 50 feet and 250 feet from the intersection. The study found that signal controlled intersection crash rates are high within the CBD areas but lower in non-CBD areas. However, all-way stop controlled intersections have high crash rates than signalized intersections in non-CBD areas. The overall finding is that the signal controlled intersections are more hazardous within CBD areas relative to non CBD areas compared to stop controlled intersections. Considering the functional class of intersecting streets, the study found that signal controlled intersections crash rates are lowest when the municipal roads are intersecting state roads. Stop controlled intersections have high crash rates when two municipal roads are crossing compared to other combination of functional classes. The intersection of municipal and state roads showed the lowest crash rates for non CBD areas. The statistical modeling validated the findings by quantifying the effect of these variables and their direction of impact (increasing or decreasing probability of crashes).
Calibration of Highway Safety Manual Predictive Models for Kansas Freeway Segments
Imalka Matarage, Kansas State UniversityShow Abstract
Sunanda Dissanayake, Kansas State University
Prediction models in the Highway Safety Manual (HSM) are used to quantify the potential safety experience of existing and new roadways. Safety Performance Functions (SPFs) in the HSM predictive method are statistical formulas developed based on limited data gathered from selected few states. Therefore, HSM recommends to modify SPFs for a certain jurisdiction by following a calibration methodology or develop local SPFs to enhance the accuracy of predicted crash frequencies. This paper demonstrates the calibration procedure and quality assessment of the calibration process for freeway segments in Kansas utilizing crash data from 2013-2015. Most of the required data were collected from two main databases maintained by Kansas Department of Transportation and the remaining were gathered using Google Earth and ArcGIS tools. A sampling technique was applied and a minimum sample size of 446 freeway segments was calculated corresponding to 95% confidence level and 5% error. Consequently, data for 521 freeway segments were collected and utilized in this freeway calibration. Estimated calibration factors were 0.952, 0.936, 1.982 and 1.843 for multiple vehicle fatal and injury, single vehicle fatal and injury, multiple vehicle property damage only and single vehicle property damage only models respectively. Results indicated that HSM methodology overpredicts crashes for fatal and injury freeway segment models and underpredicts crashes for property damage only freeway segment models in Kansas. Results of quality assessment of the calibration process showed that estimated calibration factors were satisfactory for all freeway facilities considered in this study.
A Meta-Analysis of Collision Expectations at Signalized and Stop-Controlled Intersections in North America
Andrew Northmore, University of New BrunswickShow Abstract
Eric Hildebrand, University of New Brunswick
Safety performance functions (SPFs) have been developed for specific jurisdictions and road authorities across North America, but there are practical applications for national average SPFs. Some examples include use by jurisdictions lacking resources to develop their own SPFs and for developing national guidelines such as traffic signal warrants. The only work on average collision expectation models to date are those presented in the Highway Safety Manual (HSM), but there are questions as to how representative the HSM equations are of a national average due to the scope of the studies that developed those models. This study developed models for average intersection collision expectation across Canada and the United States based on a diverse set of published jurisdiction-specific SPFs and HSM calibrations. The models focused on the effects of traffic volume, region fixed-effects, and local jurisdiction random-effects on intersection collision expectation. In general, it was found that the models that included a jurisdiction random-effect provided the best fit. These results were compared to the HSM models and there was substantial variation between the two in terms of predicting collision expectation and collision modification factors (CMFs) for signalization, suggesting that the HSM models do not adequately represent a national average. CMFs based on this research suggest that collision rates tend to increase due to signalization, whereas most published CMFs suggest a decrease. This finding suggests that jurisdiction-specific CMFs for signalization may not be transferable for use outside of the jurisdictions where they are developed.
Jurisdiction-specific versus SafetyAnalyst-default Safety Performance Functions: A Case Study on Two-lane and Multi-lane Arterials
Hector Vargas, Florida International UniversityShow Abstract
MD Asif Raihan, Florida International University
Priyanka Alluri, Florida International University
Albert Gan, Florida International University
Network screening is the most important step in the highway safety management process. Screening criteria based on the Empirical Bayes (EB) approach are considered to be most reliable as it accounts for the regression-to-the-mean (RTM) bias. However, the EB approach requires Safety Performance Functions (SPFs), preferably calibrated to local conditions, which are often unavailable. The SafetyAnalyst software, developed by the Federal Highway Administration (FHWA), automates the EB approach using the default SPFs which were developed using multiple states’ data. Local agencies are encouraged to develop jurisdiction-specific SPFs to better reflect the local conditions. However, the benefits of developing local SPFs for rural and urban two-lane and multi-lane highway facilities are unclear and may vary from state to state. This research compares the performance of Florida-specific SPFs with SafetyAnalyst-default SPFs calibrated to Florida data using mean absolute deviation, mean squared predicted error, and Freeman-Tukey R-square goodness-of-fit measures. The results showed that Florida-specific SPFs generally produced better-fitted models than the calibrated SafetyAnalyst-default SPFs. In contrast, when the crash prediction capabilities of the already-available local SPFs calibrated to the latest time period for which they will be applied are compared to the calibrated SafetyAnalyst-default SPFs, the calibrated SafetyAnalyst-default SPFs in general were found to better predict crash frequencies compared to the existing Florida-specific SPFs calibrated to the latest data. Therefore, the local SPFs are recommended when developed using present data; however, the calibrated SafetyAnalyst-default SPFs could be used if local SPFs developed from present data are not available.
Comparison and analysis of crash frequency and rate in cross-river tunnels using random-effect models
Yao Chen, Tongji UniversityShow Abstract
Yingying Xing, Tongji University
Jian Lu, Tongji University
Tao Xu, Tongji University
Yujie Liu, Tongji University
Underground road systems are becoming popular in cites as it can overcome urban space constraints and increase capacity and accessibility for urban transport systems. For cities with rivers and seas, the construction of cross-river tunnel can preserve land resources and reduce traffic congestion without affecting navigation. However, tunnel traffic safety has become an increasing concern due to frequent and serious tunnel traffic crashes. The severity of crashes and the difficulty of rescue in tunnels are higher than those of other road sections. In order to improve the safety of tunnel operation, this paper analyzes the crash data of 14 river-crossing tunnels in Shanghai from 2015 to 2016. A negative binomial (NB) model and a random-effect negative binomial (RENB) model were developed to investigate the relationship between crash frequency and potential influence factors, including tunnel geometry characteristics, traffic volume and crash location. The results show that AADT, speed limit, grade, grade differences and RGR) are likely to increase the crash frequency in cross-river tunnels while horizontal curve radius, vertical curve radius and long tunnel are associated with less crashes. This study also explored the use of crash rate instead of crash frequency as dependent variable by using random-effect Tobit model. The results indicate that the significance of most independent variables is consistent with the results found upon the RENB model based on crash frequency.
Transferability of Crash Modification Factors via Graphical Causal Models: An Introduction
Gary Davis, University of Minnesota, Twin CitiesShow Abstract
Jingru Gao, University of Minnesota, Twin Cities
This paper describes an exploratory analysis of how to transfer a crash modification factor, estimated for one set of conditions, to a different set of conditions. Such situations are likely to become important as automated vehicles improve their capabilities and increase their market share. Our starting point is a graphical model describing the dependencies among the variables in a crash mechanism, and we focus on (1) identifying sufficient conditions for taking causal information determined in one situation and applying to another, and (2) deriving expressions for computing the transferred quantities. Three simplified but plausible scenarios are proposed. For each scenario transportability analyses developed by Pearl and his associates are used to develop a re-calibration formula with which an existing CMF can be adjusted to reflect new conditions. Computation examples are used to illustrate these results.
Calibration of the Highway Safety Manual Predictive Methods for Unsignalized Intersections at Urban and Suburban Areas in Kansas.
Rijesh Karmacharya, Kansas State UniversityShow Abstract
Sunanda Dissanayake, Kansas State University
The Highway Safety Manual (HSM) provides predictive methodologies which help predict crashes on various facility types based on traffic and geometric characteristics, incorporated through Safety Performance Functions (SPFs) and Crash Modification Factors (CMFs). Since the SPFs were developed using data from the states of Minnesota and North Carolina for three-leg unsignalized intersections (3ST) and four-leg unsignalized intersections (4ST), the calibration of the predictive methodologies would increase the accuracy of the prediction for Kansas. In this study, a total of 234 3ST intersections (128 having minor AADT data and 106 intersections using estimated minor AADT data) were used for the calibration, to satisfy the HSM criteria of at least 100 crashes per year for the selected set of sites. Multiple linear regression, with log10 transformation of the continuous variables was carried out to develop a minor AADT estimation model, for which the R-squared value was 0.3281. A calibration factor of 0.51 was calculated when considering all crashes, and 0.40 for fatal and injury (FI) crashes. For 4ST, 167 intersections were used as the sample sites resulting in the calibration factor of 0.61 when considering all crashes and 0.72 for FI crashes. For both facility types, the calibration factors were less than 1.00, implying that the HSM predictive methodology overpredicted the number of crashes for the state of Kansas. The effectiveness of the developed calibration factors were checked with the help of CURE plots and Coefficient of Variation, which showed that the obtained calibration factors are acceptable for application.
An Enhanced Methodology for the Identification of Locations with High Risk of Wet Crashes
Kenneth Velez-Rodriguez, Virginia Polytechnic Institute and State UniversityShow Abstract
Samer W. Katicha, Virginia Polytechnic Institute and State University
Gerardo Flintsch, Virginia Polytechnic Institute and State University
About 18% of crashes on Virginia’s interstates from 2014 to 2016 were reported to be wet crashes. Although extensive research on crashes has been conducted, limited attention has been devoted to the prediction of wet crashes. The ratio of wet over dry crashes (wet over dry ratio, WDR) has traditionally been the parameter of interest. In this paper, negative binomial regression is used to quantify the relationship between WDR and traffic and road parameters. One issue with the WDR is the handling of sites with zero dry crash counts. This was addressed by numerically replacing the zeros with 0.5 or by using an empirical Bayes estimate of the expected number of dry crashes instead of the dry crash counts. The empirical Bayes approach resulted in a better model fit as measured using Akaike’s Information Criterion (AIC). The negative binomial model developed for wet crashes was used to identify parameters that affect the pavement water film thickness and the expected number of wet crashes. The approach identified the longitudinal grade difference as an important parameter.
Evaluating Performance of Safety Countermeasures: Applied Benefit Cost Analysis
Alan Hachey, AICP, CDM SmithShow Abstract
Michael Lawrence, Jack Faucett Associates, Inc.
Frank Gross, VHB
Geni Bahar, NAVIGATS Inc.
Karen Scurry , Federal Highway Administration (FHWA)
A Benefit Cost Analysis (BCA) is a key component of a comprehensive project or program development process that considers quantitative and qualitative impacts of highway investments. This research developed methods and procedures that transportation agencies can use to identify, quantify, and assign value to the economic benefits and costs of highway projects and programs over multiyear timeframes. This paper introduces fundamental concepts of BCA and the safety management process, defines economic measures for BCA, provides an overview of BCA in the safety management process and project development process, and identifies several related resources. The paper describes a BCA tool developed by the FHWA, Safety BCA, that supports transportation professionals in applying BCA to safety countermeasure project performance evaluation. Finally, the paper applies the process to a safety countermeasure project example: converting a rural arterial four-way stop intersection into a signalized intersection or a roundabout.
Zone-based Modeling of Time-dependent Safety Performance Using Smartphone-based Connected Vehicle Data
Di Yang, New York UniversityShow Abstract
Kun Xie, University of Canterbury
Kaan Ozbay, New York University
Hong Yang, Old Dominion University
Noah Budnick, Zendrive Inc
Safety performance functions (SPFs) are generally used to correlate risk factors with crash counts aggregated over a long time (e.g. a year), and to identify hotspots that have excessive crashes regardless of different time periods. However, it is highly likely that the relationship between risk factors and crash occurrence can vary across different times of day. This study aims to characterize time-dependent safety performance in urban areas by modeling crash counts for different times of day. Anonymized and aggregated driving data collected by the Zendrive’s smartphone-based technology is used to capture time-dependent dangerous driving events. Multivariate conditional autoregressive (MVCAR) models are developed to jointly account for spatial and temporal dependence of crash observations. Results of two-sample Kolmogorov-Smirnov tests affirm the heterogeneity of the safety effects of dangerous driving events in different time periods. Time-dependent hotspots are identified using potential for safety improvement (PSI) metric. According to the results of Wilcoxon signed-rank tests, hotspots identified by times of day are found to be mostly different from each other. The findings of this study provide insights into temporal effects of risk factors and can support the development of police patrolling plan and other road safety interventions in different times of day. Besides, this study also shows the potential of leveraging anonymized and aggregated driving data to assess traffic safety issues.