Development of crash modification factors for changing lane width on roadway segments using generalized nonlinear models

Effectiveness of road safety interventions: An evidence and gap map

Background

Road Traffic injuries (RTI) are among the top ten leading causes of death in the world resulting in 1.35 million deaths every year, about 93% of which occur in low- and middle-income countries (LMICs). Despite several global resolutions to reduce traffic injuries, they have continued to grow in many countries. Many high-income countries have successfully reduced RTI by using a public health approach and implementing evidence-based interventions. As many LMICs develop their highway infrastructure, adopting a similar scientific approach towards road safety is crucial. The evidence also needs to be evaluated to assess external validity because measures that have worked in high-income countries may not translate equally well to other contexts. An evidence gap map for RTI is the first step towards understanding what evidence is available, from where, and the key gaps in knowledge.

Objectives

The objective of this evidence gap map (EGM) is to identify existing evidence from all effectiveness studies and systematic reviews related to road safety interventions. In addition, the EGM identifies gaps in evidence where new primary studies and systematic reviews could add value. This will help direct future research and discussions based on systematic evidence towards the approaches and interventions which are most effective in the road safety sector. This could enable the generation of evidence for informing policy at global, regional or national levels.

Search Methods

The EGM includes systematic reviews and impact evaluations assessing the effect of interventions for RTI reported in academic databases, organization websites, and grey literature sources. The studies were searched up to December 2019.

Selection Criteria

The interventions were divided into five broad categories: (a) human factors (e.g., enforcement or road user education), (b) road design, infrastructure and traffic control, (c) legal and institutional framework, (d) post-crash pre-hospital care, and (e) vehicle factors (except car design for occupant protection) and protective devices. Included studies reported two primary outcomes: fatal crashes and non-fatal injury crashes; and four intermediate outcomes: change in use of seat belts, change in use of helmets, change in speed, and change in alcohol/drug use. Studies were excluded if they did not report injury or fatality as one of the outcomes.

Data Collection and Analysis

The EGM is presented in the form of a matrix with two primary dimensions: interventions (rows) and outcomes (columns). Additional dimensions are country income groups, region, quality level for systematic reviews, type of study design used (e.g., case-control), type of road user studied (e.g., pedestrian, cyclists), age groups, and road type. The EGM is available online where the matrix of interventions and outcomes can be filtered by one or more dimensions. The webpage includes a bibliography of the selected studies and titles and abstracts available for preview. Quality appraisal for systematic reviews was conducted using a critical appraisal tool for systematic reviews, AMSTAR 2.

Main Results

The EGM identified 1859 studies of which 322 were systematic reviews, 7 were protocol studies and 1530 were impact evaluations. Some studies included more than one intervention, outcome, study method, or study region. The studies were distributed among intervention categories as: human factors (n = 771), road design, infrastructure and traffic control (n = 661), legal and institutional framework (n = 424), post-crash pre-hospital care (n = 118) and vehicle factors and protective devices (n = 111). Fatal crashes as outcomes were reported in 1414 records and non-fatal injury crashes in 1252 records. Among the four intermediate outcomes, speed was most commonly reported (n = 298) followed by alcohol (n = 206), use of seatbelts (n = 167), and use of helmets (n = 66). Ninety-six percent of the studies were reported from high-income countries (HIC), 4.5% from upper-middle-income countries, and only 1.4% from lower-middle and low-income countries. There were 25 systematic reviews of high quality, 4 of moderate quality, and 293 of low quality.

Authors' Conclusions

The EGM shows that the distribution of available road safety evidence is skewed across the world. A vast majority of the literature is from HICs. In contrast, only a small fraction of the literature reports on the many LMICs that are fast expanding their road infrastructure, experiencing rapid changes in traffic patterns, and witnessing growth in road injuries. This bias in literature explains why many interventions that are of high importance in the context of LMICs remain poorly studied. Besides, many interventions that have been tested only in HICs may not work equally effectively in LMICs. Another important finding was that a large majority of systematic reviews are of low quality. The scarcity of evidence on many important interventions and lack of good quality evidence-synthesis have significant implications for future road safety research and practice in LMICs. The EGM presented here will help identify priority areas for researchers, while directing practitioners and policy makers towards proven interventions.

Lessons learned from characteristics of extraordinarily severe traffic crashes in China, 2004-2019

China has experienced remarkable achievements in terms of reducing the number of extraordinarily severe traffic crashes (ESTCs) that cause more than 10 deaths each crash. However, ESTCs still occur occasionally and result in extremely adverse social impacts. This study aims at investigating the common characteristics, characteristic patterns, and changes of characteristics of ESTCs in China with the expectation to learn from the past and act for the future. A total of 373 ESTCs occurred in 2004-2019 were collected, and characteristics of driver factors, road factors, vehicle factors, environment factors, and other factors were analyzed through the multiple correspondence analysis (MCA). The results show that run off road crashes, not qualified drivers, improper driving, large bus, overload, class II highway, and straight road sections are the most common categories of characteristics. In addition, four underlying characteristic patterns are identified through the MCA. Significant changes in characteristics and characteristic patterns are also found, and these changes are the results of various law enforcement, safety policies, educational interventions, and engineering interventions. It is also inferred that the specific law enforcement targeting to certain category of characteristics is more effective than the corresponding safety campaigns or policies in terms of ESTC prevention.

Impact of an exclusive narrow automated vehicle lane on adjacent lane driver behavior

This study contributes to understanding the behavioral impacts of infrastructure adaptation to Automated Vehicles (AVs) on non-AV drivers. It attempts to answer the question of how a narrow (9 ft) lane dedicated to AVs would affect the behavior of drivers in terms of safety measures who are driving in the adjacent lane to the right. To this end, a custom designed driving simulator world was designed mimicking the Interstate 15 smart corridor in San Diego. A group of participants were assigned to drive next to the simulated 9 ft narrow lane while a control group were assigned to drive next to a regular 12 ft AV lane. Behavior of drivers was analyzed by measuring the mean lane position, mean speed, and the mental effort. In addition to AV lane width, AV headway, gender, and right lane traffic were taken into consideration in the experimental design to investigate interaction effects. The results showed no significant differences in speed and mental effort of drivers while indicating significant differences in lane positioning. Although the overall effect of AV lane width was not significant, there were some significant interaction effects between lane width and other factors (i.e., driver gender and presence of traffic on the next regular lane to the right). In all the significant interactions, there was no case in which those factors stayed constant while AV lane width changed between the groups indicating that the significant difference might be stemmed from the other factors rather than the lane width. However, the trend observed was that drivers driving next to the 12 ft lane had better lane centering compared to the 9 ft lane. The analysis also showed that while in general female drivers tended to drive further away from the 9 ft lane and performed worse in terms of lane centering, they performed better than male drivers when right lane traffic was present.

Temporal stability of associations between crash characteristics: A multiple correspondence analysis

Understanding the associations between crash characteristics facilitates the development of traffic safety policies for improving traffic safety. This study investigates the temporal stability of associations between crash characteristics at different temporal levels using multiple correspondence analysis (MCA). For each date in 2020, crash data from the previous week, month, season, half year, one year, two years, three years, and four years are collected respectively as eight temporal levels. MCA plots and chi-square distance analysis are used to assess the temporal stability of associations between crash characteristics across dates in 2020 with data from various temporal levels. The key findings of this study demonstrate that associations between crash characteristics at lower temporal levels show notable and potential cyclical variations across dates, while more stable and long-term trend of associations between crash characteristics may be identified as the temporal level increases, especially at the two-year level and higher temporal levels at which temporal stability may be expected. The study contributes to the literature by presenting a challenge for traffic analysts in that both temporally stable and unstable associations between crash characteristics may be observed at any point in time when different temporal levels are considered as study periods. Therefore, it may serve as a foundation for future research and practical works to identify traffic safety issues and optimal policies as well as facilitate the interpretation of statistical modeling in the presence of temporally unstable data.

A Spatial Autoregressive Quantile Regression to Examine Quantile Effects of Regional Factors on Crash Rates

Spatial autocorrelation and skewed distribution are the most frequent issues in crash rate modelling analysis. Previous studies commonly focus on the spatial autocorrelation between adjacent regions or the relationships between crash rate and potentially risky factors across different quantiles of crash rate distribution, but rarely both. To overcome the research gap, this study utilizes the spatial autoregressive quantile (SARQ) model to estimate how contributing factors influence the total and fatal-plus-injury crash rates and how modelling relationships change across the distribution of crash rates considering the effects of spatial autocorrelation. Three types of explanatory variables, i.e., demographic, traffic networks and volumes, and land-use patterns, were considered. Using data collected in New York City from 2017 to 2019, the results show that: (1) the SARQ model outperforms the traditional quantile regression model in prediction and fitting performance; (2) the effects of variables vary with the quantiles, mainly classifying three types: increasing, unchanged, and U-shaped; (3) at the high tail of crash rate distribution, the effects commonly have sudden increases/decrease. The findings are expected to provide strategies for reducing the crash rate and improving road traffic safety.

Traffic volume and crashes and how crash and road characteristics affect their relationship - A meta-analysis

The present study has investigated the relationship between traffic volume and crash numbers by means of meta-analysis, based on 521 crash prediction models from 118 studies. The weighted pooled volume coefficient for all crashes and all levels of crash severity (excluding fatal crashes) is 0.875. The most important moderator variable is crash type. Pooled volume coefficients are systematically greater for multi vehicle crashes (1.210) than for single vehicle crashes (0.552). Regarding crash severity, the results indicate that volume coefficients are smaller for more fatal crashes (0.777 for all fatal crashes) than for injury crashes but no systematic differences were found between volume coefficients for injury and property-damage-only crashes. At higher levels of volume and on divided roads, volume coefficients tend to be greater than at lower levels of volume and on undivided roads. This is consistent with the finding that freeways on average have greater volume coefficients than other types of road and that two-lane roads are the road type with the smallest average volume coefficients. The results indicate that results from crash prediction models are likely to be more precise when crashes are disaggregated by crash type, crash severity, and road type. Disaggregating models by volume level and distinguishing between divided and undivided roads may also improve the precision of the results. The results indicate further that crash prediction models may be misleading if they are used to predict crash numbers on roads that differ from those that were used for model development with respect to composition of crash types, share of fatal or serious injury crashes, road types, and volume levels.

Safety effects of traffic lane and shoulder widths on two-lane undivided rural roads: A matched case-control study from Norway

This study estimates the effects of lane and shoulder widths on occurrence of head-on and single-vehicle accidents on rural two-lane undivided roads in Norway while considering the differences between winter and non-winter accidents and their severity levels. A matched case-control method was applied to calculate the odds ratios for lane and shoulder width categories, while controlling for the effects of AADT and adjusting for the effects of region, speed limit, segment length, share of long vehicles in AADT and horizontal alignment. The study used a sample of 71,999 roadway segments identified in GIS and 1886 related accidents recorded by the police in five-year period. The results suggest that it is relevant to consider winter and non-winter accidents as well as severe and slight accidents separately when studying the effects of lane and shoulder widths on the occurrence of head-on and single-vehicle accidents. When examining lane and shoulder widths for all related accidents, the lane widths 1.50-2.50 m and shoulder widths 0.50-0.75 m were relatively safer than other categories on Norwegian two-lane rural undivided roads.

Developing crash prediction models using parametric and nonparametric approaches for rural mountainous freeways: A case study on Wyoming Interstate 80

Interstate 80 in Wyoming is one of the busiest freight corridors that is characterized with harsh winter conditions and challenging mountainous roadway sections. The fatality rates in Wyoming are always typically higher than the national level. The 402-mile I-80 corridor in Wyoming was selected by the USDOT FHWA for piloting connected vehicle technology to improve the safety and mobility of heavy trucks. To accurately quantify the effectiveness of the pilot, evaluation of the pre-deployment safety performance is essential. Unlike other studies, the full 402-mile of I-80 corridor passing through Wyoming was investigated as a requirement of the USDOT. Homogeneous segmentation was used to divide the corridor based on horizontal and vertical roadway characteristics. A transferability analysis was conducted to investigate whether a short portion of the corridor would be representative of the whole 402-miles of I-80. Results showed that the whole 402 miles should be considered in the analysis due to the radical changes throughout the corridor. Several SPFs were developed using three models; negative binomial (NB) model, spatial autoregressive (SAR) model, and non-parametric multivariate adaptive regression splines (MARS). Comparisons were performed for the developed models. Crash prediction models for total crashes and Fatal and Injury (F + I) crashes in addition to truck crashes were calibrated utilizing five years of crash data from 2012 to 2016. The results obtained from the three statistical approaches showed that MARS model provided a better model fit compared to NB and SAR models, given the lower AIC values for the developed models. Yet, SAR models showed the significant spatial dependency between the neighbor roadway segments. Additionally, the NB model showed its superiority on SAR when the spatial correlation was not significant. Parametric and non-parametric techniques should be interchangeably used in developing SPFs according to the modeling needs.

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.

Examination of the reliability of the crash modification factors using empirical Bayes method with resampling technique

There have been plenty of studies intended to use different methods, for example, empirical Bayes before-after methods, to get accurate estimation of CMFs. All of them have different assumptions toward crash count if there was no treatment. Additionally, another major assumption is that multiple sites share the same true CMF. Under this assumption, the CMF at an individual intersection is randomly drawn from a normally distributed population of CMFs at all intersections. Since CMFs are non-zero values, the population of all CMFs might not follow normal distributions, and even if it does, the true mean of CMFs at some intersections may be different from that at others. Therefore, a bootstrap method based on before-after empirical Bayes theory was proposed to estimate CMFs, but it did not make distributional assumptions. This bootstrap procedure has the added benefit of producing a measure of CMF stability. Furthermore, based on the bootstrapped CMF, a new CMF precision rating method was proposed to evaluate the reliability of CMFs. This study chose 29 urban four-legged intersections as treated sites, and their controls were changed from stop-controlled to signal-controlled. Meanwhile, 124 urban four-legged stop-controlled intersections were selected as reference sites. At first, different safety performance functions (SPFs) were applied to five crash categories, and it was found that each crash category had different optimal SPF form. Then, the CMFs of these five crash categories were estimated using the bootstrap empirical Bayes method. The results of the bootstrapped method showed that signalization significantly decreased Angle+Left-Turn crashes, and its CMF had the highest precision. While, the CMF for Rear-End crashes was unreliable. For KABCO, KABC, and KAB crashes, their CMFs were proved to be reliable for the majority of intersections, but the estimated effect of signalization may be not accurate at some sites.

Examining the influence of link function misspecification in conventional regression models for developing crash modification factors

This study further examined the use of regression models for developing crash modification factors (CMFs), specifically focusing on the misspecification in the link function. The primary objectives were to validate the accuracy of CMFs derived from the commonly used regression models (i.e., generalized linear models or GLMs with additive linear link functions) when some of the variables have nonlinear relationships and quantify the amount of bias as a function of the nonlinearity. Using the concept of artificial realistic data, various linear and nonlinear crash modification functions (CM-Functions) were assumed for three variables. Crash counts were randomly generated based on these CM-Functions. CMFs were then derived from regression models for three different scenarios. The results were compared with the assumed true values. The main findings are summarized as follows: (1) when some variables have nonlinear relationships with crash risk, the CMFs for these variables derived from the commonly used GLMs are all biased, especially around areas away from the baseline conditions (e.g., boundary areas); (2) with the increase in nonlinearity (i.e., nonlinear relationship becomes stronger), the bias becomes more significant; (3) the quality of CMFs for other variables having linear relationships can be influenced when mixed with those having nonlinear relationships, but the accuracy may still be acceptable; and (4) the misuse of the link function for one or more variables can also lead to biased estimates for other parameters. This study raised the importance of the link function when using regression models for developing CMFs.

A generalized nonlinear model-based mixed multinomial logit approach for crash data analysis

The mixed multinomial logit (MNL) approach, which can account for unobserved heterogeneity, is a promising unordered model that has been employed in analyzing the effect of factors contributing to crash severity. However, its basic assumption of using a linear function to explore the relationship between the probability of crash severity and its contributing factors can be violated in reality. This paper develops a generalized nonlinear model-based mixed MNL approach which is capable of capturing non-monotonic relationships by developing nonlinear predictors for the contributing factors in the context of unobserved heterogeneity. The crash data on seven Interstate freeways in Washington between January 2011 and December 2014 are collected to develop the nonlinear predictors in the model. Thirteen contributing factors in terms of traffic characteristics, roadway geometric characteristics, and weather conditions are identified to have significant mixed (fixed or random) effects on the crash density in three crash severity levels: fatal, injury, and property damage only. The proposed model is compared with the standard mixed MNL model. The comparison results suggest a slight superiority of the new approach in terms of model fit measured by the Akaike Information Criterion (12.06 percent decrease) and Bayesian Information Criterion (9.11 percent decrease). The predicted crash densities for all three levels of crash severities of the new approach are also closer (on average) to the observations than the ones predicted by the standard mixed MNL model. Finally, the significance and impacts of the contributing factors are analyzed.

Use of empirical and full Bayes before-after approaches to estimate the safety effects of roadside barriers with different crash conditions

INTRODUCTION

Although many researchers have estimated the crash modification factors (CMFs) for specific treatments (or countermeasures), there is a lack of prior studies that have explored the variation of CMFs. Thus, the main objectives of this study are: (a) to estimate CMFs for the installation of different types of roadside barriers, and (b) to determine the changes of safety effects for different crash types, severities, and conditions.

METHOD

Two observational before-after analyses (i.e. empirical Bayes (EB) and full Bayes (FB) approaches) were utilized in this study to estimate CMFs. To consider the variation of safety effects based on different vehicle, driver, weather, and time of day information, the crashes were categorized based on vehicle size (passenger and heavy), driver age (young, middle, and old), weather condition (normal and rain), and time difference (day time and night time).

RESULTS

The results show that the addition of roadside barriers is safety effective in reducing severe crashes for all types and run-off roadway (ROR) crashes. On the other hand, it was found that roadside barriers tend to increase all types of crashes for all severities. The results indicate that the treatment might increase the total number of crashes but it might be helpful in reducing injury and severe crashes. In this study, the variation of CMFs was determined for ROR crashes based on the different vehicle, driver, weather, and time information.

PRACTICAL APPLICATIONS

Based on the findings from this study, the variation of CMFs can enhance the reliability of CMFs for different roadway conditions in decision making process. Also, it can be recommended to identify the safety effects of specific treatments for different crash types and severity levels with consideration of the different vehicle, driver, weather, and time of day information.

Evaluation of safety effectiveness of multiple cross sectional features on urban arterials

This research evaluates the safety effectiveness of multiple roadway cross-section elements on urban arterials for different crash types and severity levels. In order to consider the nonlinearity of predictors and obtain more reliable estimates, the generalized nonlinear models (GNMs) were developed using 5-years of crash records and roadway characteristics data for urban roadways in Florida. The generalized linear models (GLMs) were also developed to compare model performance. The cross-sectional method was used to develop crash modification factors (CMFs) for various safety treatments. The results from this paper indicated that increasing lane, bike lane, median, and shoulder widths were safety effective to reduce crash frequency. In particular, the CMFs for changes in median and shoulder widths consistently decreased as their widths increased. On the other hand, the safety effects of increasing lane and bike lane widths showed nonlinear variations. It was found that crash rates decrease as the lane width increases until 12ft width and it increases as the lane width exceeds 12ft. The crash rates start to decrease again after 13ft. It was also found that crash rates decreases as the bike lane width increases until 6ft width and it increases as the bike lane width exceeds 6ft. This paper demonstrated that the GNMs clearly captured the nonlinear relationship between crashes and multiple roadway cross-sectional features, which cannot be reflected by the estimated CMFs from the GLMs. Moreover, the GNMs showed better model fitness than GLMs in general. Therefore, in order to estimate more accurate CMFs, the proposed methodology of utilizing the GNMs in the cross-sectional method is recommended over using conventional GLMs when there are nonlinear relationships between the crash rate and roadway characteristics.

Assessing the safety effects of multiple roadside treatments using parametric and nonparametric approaches

This study evaluates the safety effectiveness of multiple roadside elements on roadway segments by estimating crash modification factors (CMFs) using the cross-sectional method. To consider the nonlinearity in crash predictors, the study develops generalized nonlinear models (GNMs) and multivariate adaptive regression splines (MARS) models. The MARS is one of the promising data mining techniques due to its ability to consider the interaction impact of more than one variables and nonlinearity of predictors simultaneously. The CMFs were developed for four roadside elements (driveway density, poles density, distance to poles, and distance to trees) and combined safety effects of multiple treatments were interpreted by the interaction terms from the MARS models. Five years of crash data from 2008 to 2012 were collected for rural undivided four-lane roadways in Florida for different crash types and severity levels. The results show that the safety effects decrease as density of driveways and roadside poles increase. The estimated CMFs also indicate that increasing distance to roadside poles and trees reduces crashes. The study demonstrates that the GNMs show slightly better model fitness than negative binomial (NB) models. Moreover, the MARS models outperformed NB and GNM models due to its strength to reflect the nonlinearity of crash predictors and interaction impacts among variables under different ranges. Therefore, it can be recommended that the CMFs are estimated using MARS when there are nonlinear relationships between crash rate and roadway characteristics, and interaction impacts among multiple treatments.