Safety effects of automated section speed control on the Italian motorway network

Introduction: Automated Section Speed Control (ASSC) has been identified as an effective countermeasure to reduce speeds and improve speed limit compliance. Method: An Empirical Bayes (EB) before-and-after study was performed in this research in order to evaluate the impact of the ASSC system on the expected crash frequency. The study was carried out on a sample of 125 ASSC sites of the Italian motorway network covering 1252 km, where a total of 21,721 crashes were recorded during a 10-year analysis period from 2004 to 2013. Results: Overall, the EB analysis estimated a significant 22% reduction in the expected crash frequency due to the implementation of the ASSC system. The analysis indicated that the effect is slightly larger on property damage only (PDO) crashes (− 23%) than on fatal injury (FI) crashes (− 18%) and that the highest reductions in crash frequency are expected for multi-vehicle FI crashes (− 25%) and multi-vehicle PDO crashes (− 31%). Furthermore, the results indicated that the ASSC system is more effective in reducing crash rates when traffic volume increases and it is therefore strongly recommended as a countermeasure to improve safety on high-traffic-volume motorway sections.     

Examining crash injury severity and barrier-hit outcomes from cable barriers and strong-post guardrails on Alabama’s interstate highways

Introduction: This study investigates the impact of several risk factors (i.e., roadway, driver, vehicle, environmental, and barrier-specific characteristics) on the injury severity resulting from barrier-related crashes and also on barrier-hit outcomes (i.e., vehicle containment, vehicle redirection, and barrier penetration). A total of 1,685 barrier-related crashes, which occurred on three major interstate highways (I-65, I-85, and I-20) in the state of Alabama, were collected for a seven-year period (2010–2016), and all relevant information from the police reports was reviewed. Features that were rarely explored before (e.g., median width, barrier length, barrier offset or lateral position, left shoulder width, blockout type, and number of cables) were also collected and examined. Two types of longitudinal barriers were analyzed: high-tension cable barriers installed on medians and strong-post guardrails installed on medians and/or roadsides. Method: Two separate mixed logit (MXL) models were used to analyze crash injury severity in median and roadside barrier-related crashes. Two additional MXL models were separately adopted for median and roadside barrier-related crashes to estimate the probability of three barrier-hit outcomes (vehicle containment, vehicle redirection, and barrier penetration). Results: The results of crash injury severity MXL models showed that, for both median and roadside barrier crashes, barrier penetration, female drivers, and driver fatigue were associated with a higher probability of injury or fatal crashes. The results of barrier-hit MXL models showed that longer barrier length, Brifen cable barrier system, and barrier lateral position were significant predictors of median barrier-hit outcomes, whereas dark lighting condition, driving under the influence (DUI), presence of curved freeway sections, and right shoulder width significantly contributed to roadside barrier-hit outcomes. Conclusions: The MXL model succeeded in identifying several contributing factors of crash severity and barrier-hit outcomes along Alabama’s interstate highways. Practical applications: One study application is to design longer barrier run length (greater than 1230 feet or 0.2 miles) to reduce the barrier penetration likelihood.     

Investigation of injury severities in single-vehicle crashes in North Carolina using mixed logit models

IntroductionRoadway departure (RwD) crashes, comprising run-off-road (ROR) and cross-median/centerline head-on collisions, are one of the most lethal crash types. According to the FHWA, between 2015 and 2017, an average of 52 percent of motor vehicle traffic fatalities occurred each year due to roadway departure crashes. An avoidance maneuver, inattention or fatigue, or traveling too fast with respect to weather or geometric road conditions are among the most common reasons a driver leaves the travel lane. Roadway and roadside geometric design features such as clear zones play a significant role in whether human error results in a crash. Method: In this paper, we used mixed-logit models to investigate the contributing factors on injury severity of single-vehicle ROR crashes. To that end, we obtained five years' (2010–2014) of crash data related to roadway departures (i.e., overturn and fixed-object crashes) from the Federal Highway Administration's Highway Safety Information System Database. Results: The results indicate that factors such as driver conditions (e.g., age), environmental conditions (e.g., weather conditions), roadway geometric design features (e.g., shoulder width), and vehicle conditions significantly contributed to the severity of ROR crashes. Conclusions: Our results provide valuable information for traffic design and management agencies to improve roadside design policies and implementing appropriately forgiving roadsides for errant vehicles. Practical applications: Our results show that increasing shoulder width and keeping fences at the road can reduce ROR crash severity significantly. Also, increasing road friction by innovative materials and raising awareness campaigns for careful driving at daylight can decrease the ROR crash severity.     

Inclusion of phone use while driving data in predicting distraction-affected crashes

Introduction: Given the tremendous number of lives lost or injured, distracted driving is an important safety area to study. With the widespread use of cellphones, phone use while driving has become the most common distracted driving behavior. Although researchers have developed safety performance functions (SPFs) for various crash types, SPFs for distraction-affected crashes are rarely studied in the literature. One possible reason is the lack of critical distracted behavior information in the commonly used safety data (i.e., roadway inventory, traffic, and crash counts). Recently, the frequency of phone use while driving (referred to as phone use data) is recorded by mobile application companies and has become available to safety researchers. The primary objective of this study is to examine if phone use data can potentially predict distracted-affected crashes. Method: The authors first integrated phone use data with roadway inventory, traffic, and crash data in Texas. Then, the Random Forest (RF) algorithm was applied to assess the significance of the feature - phone use while driving - for predicting the number of distraction-affected crashes on a road segment. Further, this study developed two SPFs for distraction-affected crashes with and without the phone use data, separately. Both SPFs were assessed in terms of model fitting and prediction performances. Results: RF results rank the frequency of phone use as an important factor contributing to the number of distraction-affected crashes. Performance evaluations indicated that the inclusion of phone use data in the SPFs consistently improved both fitting and prediction abilities to predict distracted-affected crashes. Practical Applications: The phone use data provide new insights into the safety analyses of distraction-affected crashes, which cannot be achieved by only using the conventional roadway inventory and crash data. Therefore, safety researchers and practitioners are encouraged to incorporate the emerging data sources in reducing distraction-affected crashes.     

Application of machine learning technique for optimizing roadside design to decrease barrier crash costs,a quantile regression model approach

Introduction: In-transport vehicles often leave the travel lane and encroach onto natural objects on the roadsides. These types of crashes are called run-off the road crashes (ROR). Such crashes accounts for a significant proportion of fatalities and severe crashes. Roadside barrier installation would be warranted if they could reduce the severity of these types of crashes. However, roadside barriers still account for a significant proportion of severe crashes in Wyoming. The impact of the crash severity would be higher if barriers are poorly designed, which could result in override or underride barrier crashes. Several studies have been conducted to identify optimum values of barrier height. However, limited studies have investigated the monetary benefit associated with adjusting the barrier heights to the optimal values. In addition, few studies have been conducted to model barrier crash cost. This is because the crash cost is a heavily skewed distribution, and well-known distributions such as linear or poison models are incapable of capturing the distribution. A semi-parametric distribution such as asymmetric Laplace distribution can be used to account for this type of sparse distribution. Method: Interaction between different predictors were considered in the analysis. Also, to account for exposure effects across various barriers, barrier lengths and traffic volumes were incorporated in the models. This study is conducted by using a novel machine-learning-based cost-benefit optimization to provide an efficient guideline for decision makers. This method was used for predicting barrier crash costs without barrier enhancement. Subsequently the benefit was obtained by optimizing traffic barrier height and recalculating the benefit and cost. The trained model was used for crash cost prediction on barriers with and without crashes. Results: The results of optimization clearly demonstrated the benefit of optimizing the heights of road barriers around the state. Practical Applications: The findings can be utilized by the Wyoming Department of Transportation (WYDOT) to determine the heights of which barriers should be optimized first. Other states can follow the procedure described in this paper to upgrade their roadside barriers.     

The effect of extending graduated driver licensing to older novice drivers in Indiana

Introduction: Graduated driver licensing (GDL) systems have been shown to reduce rates of crashes, injuries, and deaths of young novice drivers. However, approximately one in three new drivers in the United States obtain their first driver’s license at age 18 or older, and thus are exempt from most or all provisions of GDL in most states. Method: In July 2015, the state of Indiana updated its GDL program, extending its restrictions on driving at night and on carrying passengers during the first 6 months of independent driving, previously only applicable to new drivers younger than 18, to all newly-licensed drivers younger than 21 years of age. The current study examined monthly rates of crashes per licensed driver under the affected conditions (driving at night and driving with passengers) among Indiana drivers first licensed at ages 18, 19, and 20 under the updated GDL system compared with drivers licensed at the same ages under the previous GDL system. We used Poisson regression to estimate the association between the GDL system and crash rates, while attempting to control for other factors that might have also influenced crash rates. We used linear regression to estimate the association between the GDL system and the proportion of all crashes that occurred under conditions restricted by the GDL program. Results: Results showed, contrary to expectations, that rates of crashes during restricted nighttime hours and with passengers were higher among drivers licensed under the updated GDL system. This mirrored a statewide increase in crash rates among drivers of all ages over the study period and likely reflected increased overall driving exposure. The proportions of all crashes that were at night or with passengers did not change. Practical Applications: More research is needed to understand how older novice drivers respond when GDL systems originally designed for younger novice drivers are applied to them.     

Ability to monitor driving under the influence of marijuana among non-fatal motor-vehicle crashes: An evaluation of the Colorado electronic accident reporting system

Introduction: As more states legalize medical/recreational marijuana use, it is important to determine if state motor-vehicle surveillance systems can effectively monitor and track driving under the influence (DUI) of marijuana. This study assessed Colorado's Department of Revenue motor-vehicle crash data system, Electronic Accident Reporting System (EARS), to monitor non-fatal crashes involving driving under the influence (DUI) of marijuana. Methods: Centers for Disease Control and Prevention guidelines on surveillance system evaluation were used to assess EARS' usefulness, flexibility, timeliness, simplicity, acceptability, and data quality. We assessed system components, interviewed key stakeholders, and analyzed completeness of Colorado statewide 2014 motor-vehicle crash records. Results: EARS contains timely and complete data, but does not effectively monitor non-fatal motor-vehicle crashes related to DUI of marijuana. Information on biological sample type collected from drivers and toxicology results were not recorded into EARS; however, EARS is a flexible system that can incorporate new data without increasing surveillance system burden. Conclusions: States, including Colorado, could consider standardization of drug testing and mandatory reporting policies for drivers involved in motor-vehicle crashes and proactively address the narrow window of time for sample collection to improve DUI of marijuana surveillance. Practical applications: The evaluation of state motor-vehicle crash systems' ability to capture crashes involving drug impaired driving (DUID) is a critical first step for identifying frequency and risk factors for crashes related to DUID.     

Why is the rate of annual road fatalities increasing? A unit record analysis of New Zealand data (2010–2017)

IntroductionRecent increases in road crashes have reversed New Zealand’s formerly declining crash rates to produce annual fatal and serious injury counts that are 49% higher than the lowest rates achieved in 2013. Method: We model twenty-one factors in fatal and serious injury crashes, four years before and after 2013 using logistic regression. Three major factors are significantly different in the period after 2013, when crash rates increased: (1) alcohol as a cause, (2) learner licence holders, and (3) a regional effect for Auckland. Newly defined speed zones are a more common setting for crashes in the period of upturn but there is no coinciding elevated likelihood of ‘speed as a causal factor’. Three factors related to road safety were less common: aged under 25-years old, fatigue, and not wearing a seatbelt. Results: Results are compared to rates of prosecutions for alcohol-related driving offences over this period. It is possible that New Zealand’s successful road safety initiatives of the past have been undermined by reduced levels of enforcement and an unexpected outcome from the graduated driving licence system.     

Ordered logistic models of influencing factors on crash injury severity of single and multiple-vehicle downgrade crashes: A case study in Wyoming

Introduction: The state of Wyoming, like other western United States, is characterized by mountainous terrain. Such terrain is well noted for its severe downgrades and difficult geometry. Given the specific challenges of driving in such difficult terrain, crashes with severe injuries are bound to occur. The literature is replete with research about factors that influence crash injury severity under different conditions. Differences in geometric characteristics of downgrades and mechanics of vehicle operations on such sections mean different factors may be at play in impacting crash severity in contrast to straight, level roadway sections. However, the impact of downgrades on injury severity has not been fully explored in the literature. This study is thus an attempt to fill this research gap. In this paper, an investigation was carried out to determine the influencing factors of crash injury severities of downgrade crashes. Method: Due to the ordered nature of the response variable, the ordered logit model was chosen to investigate the influencing factors of crash injury severities of downgrade crashes. The model was calibrated separately for single and multiple-vehicle crashes to ensure the different factors influencing both types of crashes were captured. Results: The parameter estimates were as expected and mostly had signs consistent with engineering intuition. The results of the ordered model for single-vehicle crashes indicated that alcohol, gender, road condition, vehicle type, point of impact, vehicle maneuver, safety equipment use, driver action, and annual average daily traffic (AADT) per lane all impacted the injury severity of downgrade crashes. Safety equipment use, lighting conditions, posted speed limit, and lane width were also found to be significant factors influencing multiple-vehicle downgrade crashes. Injury severity probability plots were included as part of the study to provide a pictorial representation of how some of the variables change in response to each level of crash injury severity. Conclusion: Overall, this study provides insights into contributory factors of downgrade crashes. The literature review indicated that there are substantial differences between single- and multiple vehicle crashes. This was confirmed by the analysis which showed that mostly, separate factors impacted the crash injury severity of the two crash types. Practical applications: The results of this study could be used by policy makers, in other locations, to reduce downgrade crashes in mountainous areas.     

Multi-agent traffic simulations to estimate the impact of automated technologies on safety

Abstract

Objective: The objective of this article was to develop a multi-agent traffic simulation methodology to estimate the potential road safety improvements of automated vehicle technologies.

Methods: We developed a computer program that merges road infrastructure data with a large number of vehicles, drivers, and pedestrians. Human errors are induced by modeling inattention, aimless driving, insufficient safety confirmation, misjudgment, and inadequate operation. The program was applied to simulate traffic in a prescribed area in Tsukuba city. First, a 100% manual driving scenario was set to simulate traffic for a total preset vehicle travel distance. The crashes from this simulation were compared with real-world crash data from the prescribed area from 2012 to 2017. Thereafter, 4 additional scenarios of increasing levels of automation penetration (including combinations of automated emergency braking [AEB], lane departure warning [LDW], and SAE Level 4 functions) were implemented to estimate their impact on safety.

Results: Under manual driving, the system simulated a total of 859 crashes including single-car lane departure, car-to-car, and car-to-pedestrian crashes. These crashes tended to occur in locations similar to real-world crashes. The number of crashes predicted decreased to 156 cases with increasing level of automation. All of the technologies considered contributed to the decrease in crashes. Crash reductions attributable to AEB and LDW in the simulations were comparable to those reported in recent field studies. For the highest levels of automation, no assessment data were available and hence the results should be carefully treated. Further, in modeling automated functions, potentially negative aspects such as sensing failure or human overreliance were not incorporated.

Conclusions: We developed a multi-agent traffic simulation methodology to estimate the effect of different automated vehicle technologies on safety. The crash locations resulting from simulations of manual driving within a limited area in Japan were preliminary assessed by comparison with real-world crash data collected in the same area. Increasing penetration levels of AEB and LDW led to a large reduction in both the frequency and severity of rear-end crashes, followed by car-to-car head-on crashes and single-vehicle lane departure crashes. Preliminary estimations of the potential safety improvements that may be achieved with highly automated driving technologies were also obtained.     

Validating the underlying assumption of quasi-induced exposure technique disaggregated by crash injury severity

Introduction: Quasi-induced exposure (QIE) technique has been popularly applied in the field of traffic safety research for decades. One of the basic assumptions of QIE theory is that the not-at-fault driving parties (D2s) involved in the crashes are the random selection of overall driving population at the event of crash occurrence. Very few literatures, however, can be identified to validate the assumption for crashes with specific injury severities that may not be satisfied in reality. Method: The study aims to check the validity of the assumption categorized by crash injury severity with the use of Michigan crash data. Latent class analysis is employed to generate several latent classes for the crashes with specific injury outcomes. Chi-square test is adopted to identify the significance of the similarity of D2 distributions among the latent classes. Results: The results indicate that: (a) for fatal crashes the statistical tests do not identify the significant discrepancies for D2 distributions of driver gender, age, and vehicle type between latent classes; (b) for injury crashes, both D2 driver gender and age have the similar distributions between/among various classes, while the D2 vehicle types show the inconsistent distributions; and (c) with respect to property damage only crashes, the distributions of three vehicle-driver characteristics are significantly different among the latent classes. It implies that the underlying assumption may not entirely hold true for all the injury severities and driver-vehicle characteristics. Practical Applications: The findings pinpoint the applicability of the QIE technique under specific scenarios and highlight the importance of validating the underlying assumption of QIE prior to its application.     

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

IntroductionAlthough 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.MethodTwo 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).ResultsThe 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 applicationsBased 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.     

Validation of not-at-fault driver representativeness assumption for quasi-induced exposure using U.S. national traffic databases

Introduction:The quasi-induced exposure (QIE) method has been widely implemented into traffic safety research. One of the key assumptions of QIE method is that not-at-fault drivers represent the driving population at the time of a crash. Recent studies have validated the QIE representative assumption using not-at-fault drivers from three-or-more vehicle crashes (excluding the first not-at-fault drivers; D3_other) as the reference group in single state crash databases. However, it is unclear if the QIE representativeness assumption is valid on a national scale and is a representative sample of driving population in the United States. The aims of this study were to assess the QIE representativeness assumption on a national scale and to evaluate if D3_other could serve as a representative sample of the U.S. driving population. Method: Using the Fatality Analysis Reporting System (FARS) and the National Occupant Protection Use Survey (NOPUS), distributions of driver gender, age, vehicle type, time, and roadway type among the not-at-fault drivers in clean two-vehicle crashes, the first not-at-fault drivers in three-or-more-vehicle crashes, and the remaining not-at-fault drivers in three-or-more vehicle crashes were compared to the driver population observed in NOPUS. Results: The results showed that with respect to driver gender, vehicle type, time, and roadway type, drivers among D3_other did not show statistical significant difference from NOPUS observations. The age distribution of D3_other driver was not practically different to NOPUS observations. Conclusions: Overall, we conclude that D3_other drivers in FARS represents the driving population at the time of the crash. Practical applications: Our study provides a solid foundation for future studies to utilize D3_other as the reference group to validate the QIE representativeness assumption and has potential to increase the generalizability of future FARS studies.     

Understanding crash potential associated with teen driving: Survey analysis using multivariate graphical method

Introduction: Teen crash involvement is usually higher than other age groups, and they are typically overrepresented in car crashes. To infer teen drivers' understanding of crash potentials (factors that are associated with crash occurrence), two sources of data are generally used: retrospective data and prospective data. Retrospective data sources contain historical crash data, which have limitations in determining teen drivers' knowledge of crash potentials. Prospective data sources, like surveys, have more potential to minimize the research gap. Prior studies have shown that teen drivers are more likely to be involved in crashes during their early driving years. Thus, there is a benefit in examining how teen drivers' understanding of crash potentials change during their transition through licensing stages (i.e., no licensure to unrestricted licensure). Method: This study used a large set of teen driver survey data (a dataset from approximately 88,000 respondents) of Texas teens to answer the research question. Researchers provided rankings of the crash potentials by gender and licensure stages using a multivariate graphical method named taxicab correspondence analysis (TCA). Results: The findings show that driving behavior and understanding of crash potentials differ among teens based upon various licensing stages. Practical applications: Findings from this study can help government authorities to refine policies of teen driver licensing and implement potential countermeasures for safety improvement.     

Characteristics of crashes and injuries among 14 and 15 year old drivers,by rurality

Purpose: Motor-vehicle crashes continue to be the leading cause of death for teenagers in the United States. The United States has some of the youngest legal driving ages worldwide. The objective of this study was to determine rates and factors associated with injury crashes among 14- and 15-year-old drivers and how these varied by rurality. Methods: Data for this cross-sectional study of 14- and 15-year-old drivers were obtained from the Iowa Department of Transportation from 2001 to 2013. Crash and injury crash rates were calculated by rurality. The relationship between crash and driver factors and injury was assessed using logistic regression. Findings: Teen drivers, aged 14 and 15 years, had a statewide crash rate of 8 per 1,000 drivers from 2001 to 2013. The majority of crashes occurred in urban areas (51%), followed by in town (29%), remote rural areas (13%), and suburban areas (7%). Crash and injury crash rates increased as level of rurality increased. The odds of an injury crash increased more than 10-fold with the presence of multiple other teens as passengers, compared to no passengers (OR = 10.7, 95% CI: 7.1–16.2). Conclusions: Although 14- and 15-year-old drivers in Iowa have either limited unsupervised (school permits) or supervised only driving restrictions, they are overrepresented in terms of crashes and injury crashes. Rural roads and multiple teen passengers are particularly problematic in terms of injury outcomes. Practical applications: Results from this study support passenger restrictions and teen driving interventions designed with a rural focus.     

Crash risk by driver age,gender, and time of day using a new exposure methodology

Introduction: Concerns have been raised that the nonlinear relation between crashes and travel exposure invalidates the conventional use of crash rates to control for exposure. A new metric of exposure that bears a linear association to crashes was used as basis for calculating unbiased crash risks. This study compared the two methods – conventional crash rates and new adjusted crash risk – for assessing the effect of driver age, gender, and time of day on the risk of crash involvement and crash fatality. Method: We used police reports of single-car and multi-car crashes with fatal and nonfatal driver injuries that occurred during 2002–2012 in Great Britain. Results: Conventional crash rates were highest in the youngest age group and declined steeply until age 60–69 years. The adjusted crash risk instead peaked at age 21–29 years and reduced gradually with age. The risk of nighttime driving, especially among teenage drivers, was much smaller when based on adjusted crash risks. Finally, the adjusted fatality risk incurred by elderly drivers remained constant across time of day, suggesting that their risk of sustaining a fatal injury due to a crash is more attributable to excess fragility than to crash seriousness. Conclusions: Our findings demonstrate a biasing effect of low travel exposure on conventional crash rates. This implies that conventional methods do not yield meaningful comparisons of crash risk between driver groups and driving conditions of varying exposure to risk. The excess crash rates typically associated with teenage and elderly drivers as well as nighttime driving are attributed in part to overestimation of risk at low travel exposure. Practical Applications: Greater attention should be directed toward crash involvement among drivers in their 20s and 30s as well as younger drivers. Countermeasures should focus on the role of physical vulnerability in fatality risk of elderly drivers.     

Long-term crash trends at single- and double-lane roundabouts in Washington State

Introduction: Despite the proven safety benefits, crashes still occur at roundabouts. This study examined long-term trends in total crash counts, crash severity, and crashes involving common driver errors (failing to yield the right-of-way and speeding) in the period following the completion of single- and double-lane roundabouts in Washington state. Method: Crashes occurring during 2010–2016 at single- and double-lane roundabouts completed between 2009 and 2015 in Washington state were included. Poisson regression examined changes in annual total crash counts over time. Logistic regression estimated average annual changes in the odds that a crash involved an evident/incapacitating/fatal injury and that a crash involved a driver error. Regression models were estimated for single- and double-lane roundabouts separately. Results: Annual total crash counts declined significantly by 8.8% over time at double-lane roundabouts and increased nonsignificantly over time at single-lane roundabouts. The study estimated a significant 32.1% annual reduction in the odds that a crash involved an evident or incapacitating injury at double-lane roundabouts and a nonsignificant 18.9% reduction at single-lane roundabouts. There was a significant 10.6% annual decline in the odds that a crash was right-of-way related at double-lane roundabouts and a significant 19.1% annual decline in the odds that a crash was speeding-related at single-lane roundabouts. Conclusions: The current study demonstrates that safety can improve over time at double-lane roundabouts as drivers gain experience navigating them. At the same time, it is important that roundabouts include design elements that will prevent right-of-way mistakes and reduce speeds. Practical applications: Communities installing double-lane roundabouts may find that their benefits will increase the longer they are in place, even if initial changes in crashes and injuries are underwhelming.