Using conditional inference forests to identify the factors affecting crash severity on arterial corridors

Introduction

The study aims at identifying traffic/highway design/driver-vehicle information significantly related with fatal/severe crashes on urban arterials for different crash types. Since the data used in this study are observational (i.e., collected outside the purview of a designed experiment), an information discovery approach is adopted for this study.

Method

Random Forests, which are ensembles of individual trees grown by CART (Classification and Regression Tree) algorithm, are applied in numerous applications for this purpose. Specifically, conditional inference forests have been implemented. In each tree of the conditional inference forest, splits are based on how good the association is. Chi-square test statistics are used to measure the association. Apart from identifying the variables that improve classification accuracy, the methodology also clearly identifies the variables that are neutral to accuracy, and also those that decrease it.

Results

The methodology is quite insightful in identifying the variables of interest in the database (e.g., alcohol/ drug use and higher posted speed limits contribute to severe crashes). Failure to use safety equipment by all passengers and presence of driver/passenger in the vulnerable age group (more than 55 years or less than 3 years) increased the severity of injuries given a crash had occurred. A new variable, ‘element’ has been used in this study, which assigns crashes to segments, intersections, or access points based on the information from site location, traffic control, and presence of signals.

Impact

The authors were able to identify roadway locations where severe crashes tend to occur. For example, segments and access points were found to be riskier for single vehicle crashes. Higher skid resistance and k-factor also contributed toward increased severity of injuries in crashes.     

Implied-consent laws: A review of the literature and examination of current problems and related statutes

Problem

A substantial proportion of drivers arrested for DUI refuse the BAC test, thereby reducing the likelihood that they will be convicted and potentially increasing the number of high-risk multiple offenders contributing to alcohol-related crashes.

Method

This paper reviews the information on the current status of implied-consent laws (which impose a sanction on offenders who refuse the BAC test) in the 50 states and the other relevant traffic safety laws and policies that may influence state refusal rates.

Results

Although there appears to be only a weak relationship between state refusal rates and crash rates, there is strong evidence that BAC test refusals significantly compromise the arrest, prosecution, and sentencing of DUI suspects and the overall enforcement of DUI laws in the United States.

Discussion

Laws and policies that may reduce the number of refusals are discussed.

Impact on industry

Alcohol-related crash injuries are an important cost problem for U.S. industry because of property damage from crashes, crash injuries to employees that raise health costs, or the reduction of time on the job resulting from a highway injury.     

Injury outcomes and costs for cross-median and median barrier crashes

Introduction

The objective of this research was to quantify the injury outcomes and develop reliable and comprehensive injury costs for cross-median crashes (CMC) and median barrier crashes (MBC).

Method

A three-step methodology was developed to quantify the crash costs for each crash severity and type. All CMC and MBC between 2001 and 2007 in Wisconsin were identified and used in this analysis. The Wisconsin CODES database provided comprehensive injury costs based on the injury types and severities suffered by participants in study crashes.

Results

As expected, multi-vehicle CMC result in more total injuries and more severe injuries than single-vehicle CMC. Injury costs for the same injury level on KABCO scale are different for different crash types. Injury costs for concrete MBC are 33% to 50% less than those of multi-vehicle CMC, while the injury costs of concrete MBC for lower severities (B and C) are similar to those of single-vehicle CMC for the same severities; but for incapacitating injuries the costs are 30% less. As expected, concrete MBC result in lower severities than CMC. The costs, by crash severity, vary significantly between different crash types. Concrete median barrier injury crashes are roughly 20% of multi-vehicle CMC costs and 50% of single-vehicle CMC costs.

Conclusions

Results indicate that using one set of crash costs for all crash types biases any evaluation. Therefore, it is recommended that crash-type-specific costs be used in applications such as development of median barrier warrant where specific types of crashes are considered (CMC and MBC).

Impact on industry

Using crash specific costs can lead to a more realistic benefit-cost analysis and enable better decision-making.     

A national evaluation of the nighttime and passenger restriction components of graduated driver licensing

Introduction

The high crash rate of youthful novice drivers has been recognized for half a century. Over the last decade, graduated driver licensing (GDL) systems, which extend the period of supervised driving and limit the novice's exposure to higher-risk conditions (such as nighttime driving), have effectively reduced crash involvements of novice drivers.

Method

This study used data from the Fatality Analysis Reporting System (FARS) and the implementation dates of GDL laws in a state-by-year panel study to evaluate the effectiveness of two key elements of GDL laws: nighttime restrictions and passenger limitations.

Results

Nighttime restrictions were found to reduce 16- and 17-year-old driver involvements in nighttime fatal crashes by an estimated 10% and 16- and 17-year-old drinking drivers in nighttime fatal crashes by 13%. Passenger restrictions were found to reduce 16- and 17-year-old driver involvements in fatal crashes with teen passengers by an estimated 9%.

Conclusions

These results confirm the effectiveness of these provisions in GDL systems. Impact on Public Health. States without the nighttime or passenger restrictions in their GDL law should strongly consider adopting them.

Impact on Industry

The results of this study indicate that nighttime restrictions and passenger limitations are very important components of any GDL law.     

Risks older drivers pose to themselves and to other road users

Problem

Studies have shown that older drivers have high death rates and lower rates of involvement in crashes that kill others; but most studies have not considered drivers' responsibility for their crashes, and many have considered only one particular measure of risk.

Method

This study examines risks that drivers of various ages pose to themselves and to others on per-driver, per-trip, and per-mile bases, taking responsibility for crashes into account, using United States fatal crash data from 1999 through 2003 and travel estimates from 2001.

Results

Relative to other age groups, drivers aged 85 and older face the highest risk of their own death, whereas teens pose the greatest risk to passengers, occupants of other vehicles, and non-motorists.

Discussion

The oldest drivers pose more risk to other road users than middle-aged drivers do; the degree of their excess risk depends strongly upon how risk is measured.

Impact on industry

These results demonstrate the importance of keeping clear the meaning and implications of various risk measures.     

Crash involvement during the different phases of the New Zealand Graduated Driver Licensing System (GDLS)

Introduction

The New Zealand Graduated Driver Licensing System (GDLS) is designed to allow novice drivers to gain driving experience under conditions of reduced risk.

Method

To examine the effectiveness of the GDLS, an analysis of how the crash involvement of novice drivers changes as drivers move through the GDLS was undertaken. Crash profiles were created by data matching the New Zealand license and crash databases, covering a time period from 1999-2006.

Results

The crash profiles show that the initial learner period of the GDLS is relatively safe and the time at which novice drivers have the highest rate of crash involvement is during the first few months of solo driving. Analysis using logistic regression also showed an effect of age and gender, with higher crash involvement associated with younger drivers and males. In addition, individuals who gained a full license within 12-18 months of holding a restricted license, due to completion of a time-discount associated educational program, had a higher level of involvement in crashes than individuals who gained a full license after 18 months.

Conclusions

The crash profiles provide an insight into the crash risk associated with different phases of the New Zealand GDLS.

Impact on Industry

Increasing the age at which drivers first begin to solo drive and the removal of the time-discount associated with completion of an educational program should be considered.     

The Long Beach/Fort Lauderdale relative risk study

Problem

The role of alcohol as a major factor in traffic crash causation has been firmly established. However, controversy remains as to the precise shape of the relative risk function and the BAC at which crash risk begins to increase.

Methods

This study used a case-control design in two locations: Long Beach, California, and Fort Lauderdale, Florida. Data were collected on 2,871 crashes of all severities and a matched control group of drivers selected from the same time, location, and direction of travel as the crash drivers. Of the 14,985 sample drivers, 81.3% of the crash drivers and 97.9% of the controls provided a valid BAC specimen.

Results

When adjusted for covariates and nonparticipation bias, increases in relative risk were observed at BACs of .04-.05, and the elevations in risk became very pronounced when BACs exceeded .10.

Discussion

The results provide strong support for .08 per se laws and for state policies that increase sanctions for BACs in excess of .15.

Impact on Industry

This study provides further precision on the deleterious effects of alcohol on driving and, by implication, on other complex tasks.     

Severity models of cross-median and rollover crashes on rural divided highways in Pennsylvania

Introduction

Crossover and rollover crashes in earth-divided, traversable medians on rural divided highways can lead to severe injury outcomes. This study estimated severity models of these two crash types. Vehicle, driver, roadway, and median cross-section design data were factors considered in the models. A unique aspect of the data used to estimate the models were the availability of median cross-slope data, which are not commonly included in roadway inventory data files.

Methods

A binary logit model of cross-median crash severity and a multinomial logit model of rollover crash severity were estimated using five years of data from rural divided highways in Pennsylvania.

Results

The highest probability of a fatal or major injury in cross-median and rollover crashes was found to occur in cases when a driver was not wearing a seatbelt. While flatter cross-slopes and narrower medians were associated with more severe cross-median crash outcomes, steeper cross-slopes and narrower medians significantly increased rollover crash severity outcomes. The presence of horizontal curves was associated with increased probabilities of high-severity outcomes in a median rollover crash.

Impact on Industry

Modeling results in this study confirmed that cross-median and median rollover crash severity outcomes are associated with median cross-section design characteristics. Based on the estimated models, it appears that flatter and narrower medians lead to more severe injury outcomes in cross-median crashes. Steeper median cross-slopes and narrower medians were associated with higher probabilities of more severe outcomes in median rollover crashes. The results presented in this study suggest that there is a trade-off between median cross-section design and cross-median and rollover crashes in earth-divided, traversable medians on rural divided highways. While the severity models can be included in a framework to develop design guidance in relation to this trade-off, models of crash frequency should also be considered.     

A traffic safety evaluation of California's traffic violator school citation dismissal policy

Introduction

This study evaluated California's traffic violator school (TVS) citation dismissal policy.

Method

This study identified and compared two large samples of drivers either completing a TVS (N = 210,015) or convicted of a traffic citation (N = 168,563).

Results

Prior to adjudication, the TVS group had characteristics (e.g., lower prior conviction rate and smaller proportion of males) that were predictive of a lower subsequent crash risk. However, the TVS group exhibited significantly more crashes than did the convicted group in the subsequent one-year period. The difference (4.83%) increased to 10% after adjusting for the more favorable characteristics of the TVS group. The TVS group also had a higher adjusted subsequent crash rate at each prior driver record entry level, reflecting a loss in the general and specific deterrence of the non-conviction masked status of TVS dismissed citations. It was reported that the TVS dismissal policy results in approximately 12,300 additional crashes annually with economic costs of approximately $398,000,000.

Conclusions

The avoidance of licensing actions resulting from the dismissal policy assists in explaining why the driving public is exposed to an increased crash risk. A number of recommendations are offered to reduce the negative traffic safety impact of the TVS citation dismissal policy.     

The accident-exposure association: self-reported versus recorded collisions

Problem

It has been claimed that exposure to risk of road traffic accidents (usually conceptualized as mileage) is curvilinearly associated with crashes (i.e., the increase in number of crashes decreases with increased mileage). However, this effect has been criticized as mainly an artifact of self-reported data.

Method

To test the proposition that self-reported accidents create part of the curvilinearity in data by under-reporting by high-accident drivers, self-reported and recorded collisions were plotted against hours of driving for bus drivers.

Results

It was found that the recorded data differed from self-reported information at the high end of exposure, and had a more linear association with the exposure measure as compared to the self-reported data, thus supporting the hypothesis.

Discussion

Part of the previously reported curvilinearity between accidents and exposure is apparently due to biased methods. Also, the interpretation of curvilinearity as an effect of exposure upon accidents was criticized as unfounded, as the causality may just as well go the other way. Impact on industry: The question of how exposure associates with crash involvement is far from resolved, and everyone who uses an exposure metric (mileage, time, induced) should be careful to investigate the exact properties of their variable before using it.     

Television news' coverage of motor-vehicle crashes

Problem

Traffic crashes are a major cause of injury and death. Although it has been argued that a skewed estimation of personal risks may be partly attributable to news representations of mortality causes, the manners in which traffic crashes are covered in the news have not received much attention in the literature.

Method

The current study used content analysis to examine the framing of traffic crashes in television news.

Results

The data indicated that Flemish television news does not report traffic crashes from a public health perspective and consequently it misses opportunities to inform the public on the causes and consequences of risk-taking in traffic.

Impact on industry

It is important that professionals are aware of these findings and stress contextual factors surrounding traffic crashes in their communication with journalists. This may lead to more accurate reports and a more accurate risk perception of viewers.     

Effects of red light camera enforcement on fatal crashes in large U.S. cities

Objective

To estimate the effects of red light camera enforcement on per capita fatal crash rates at intersections with signal lights.

Methods

From the 99 large U.S. cities with more than 200,000 residents in 2008, 14 cities were identified with red light camera enforcement programs for all of 2004-2008 but not at any time during 1992-1996, and 48 cities were identified without camera programs during either period. Analyses compared the citywide per capita rate of fatal red light running crashes and the citywide per capita rate of all fatal crashes at signalized intersections during the two study periods, and rate changes then were compared for cities with and without cameras programs. Poisson regression was used to model crash rates as a function of red light camera enforcement, land area, and population density.

Results

The average annual rate of fatal red light running crashes declined for both study groups, but the decline was larger for cities with red light camera enforcement programs than for cities without camera programs (35% vs. 14%). The average annual rate of all fatal crashes at signalized intersections decreased by 14% for cities with camera programs and increased slightly (2%) for cities without cameras. After controlling for population density and land area, the rate of fatal red light running crashes during 2004-2008 for cities with camera programs was an estimated 24% lower than what would have been expected without cameras. The rate of all fatal crashes at signalized intersections during 2004-2008 for cities with camera programs was an estimated 17% lower than what would have been expected without cameras.

Conclusions

Red light camera enforcement programs were associated with a statistically significant reduction in the citywide rate of fatal red light running crashes and a smaller but still significant reduction in the rate of all fatal crashes at signalized intersections.

Impact on Industry

The study adds to the large body of evidence that red light camera enforcement can prevent the most serious crashes. Communities seeking to reduce crashes at intersections should consider this evidence.     

Survival risk factors for fatal injured car and motorcycle drivers in single alcohol-related and alcohol-unrelated vehicle crashes

Introduction

A high percentage of drivers who die as a result of a single vehicle crash are under the influence of alcohol. We aimed to better understand the prevalence of these fatalities and the ratio of death to injuries based on various risk factors. We focused on alcohol-related and -unrelated single-vehicle crashes to investigate the influence of such risk factors on the time until death for car and motorcycle drivers.

Methods

We combined data from national police reports and a vital registration database in Taiwan. Survival analysis using Cox regression models was used to identify the risk factors of time until death.

Results

Overall, nearly 60% of car driver fatalities and 40% of motorcycle driver fatalities involved the consumption of alcohol. Survival analysis of single-vehicle crashes suggested that the traffic island separation between a car moving at a higher speed and motorcycle traffic resulted in a higher risk of death over time for motorcycle drivers who consumed alcohol. The factors attributed to a higher risk of death over time for motorcycle drivers were older age, crashing into trees, night-time driving, driving on curved roads, and driving on local roads. Driving without restraints and driving on roads with higher speed limits attributed to a higher risk of death over time for car drivers.

Conclusions

The factors that influence the risk of death over time in a motor-vehicle accident involving alcohol depended on different elements, which should each be considered when attempting to reduce this risk.

Impact on Industry

More efforts should be made to investigate the various risk factors in areas with large motorcycle populations.     

Age and gender differences in conviction and crash occurrence subsequent to being directed to Iowa's driver improvement program

Introduction

This paper investigates potential gender and age differences in conviction and crash occurrence subsequent to being directed to attend Iowa's Driver Improvement Program (DIP).

Methods

Binary logit models were developed to investigate the factors that influence conviction occurrence after DIP by gender and age. Because of the low crash occurrence subsequent to DIP, association rules were applied to investigate the factors that influence crash occurrence subsequent to DIP, in lieu of econometric models.

Results

There were statistical significant differences by driver gender, age, and conviction history in the likelihood of subsequent convictions. However, this paper found no association between DIP outcome, crash history, and crash occurrence.

Impact on industry

Evaluating the differences in conviction and crash occurrence subsequent to DIP between female and male drivers, and among different age groups can lead to improvements of the effectiveness of DIPs and help to identify low-cost intervention measures, customized based on drivers’ gender and age, for improving driving behaviors.     

Does the built environment affect when American teens become drivers? Evidence from the 2001 National Household Travel Survey

Problem

Motor vehicle crashes are the most common cause of death for American adolescents. However, the impact of where teens live on when they begin driving has not been studied.

Method

Data from the 2001 National Household Travel Survey were used to estimate the effect of residential density on the driver status of teens aged 16 to 19 years after matching on demographic characteristics.

Results

Controlling for demographic characteristics, 16 and 17 year old teens in high density neighborhoods had driver rates 15 percentage points below teens living in less dense areas (p < 0.001). The effect for 18 and 19 year olds was a 9 percentage point decrease (p < 0.001).

Summary

These results suggest teens living in less dense and more sprawling communities initiate driving at a younger age than comparable teens in compact areas, placing them at increased risk for crash related injuries. Impact on Industry: The role of environmental factors, such as neighborhood walkability and provision of transit, should be considered in young driver programs.     

A Bayesian assessment of the effect of highway bypasses in Iowa on crashes and crash rate

Introduction

A common contention is that the construction of highway bypasses negatively impacts the economy of local communities by reducing pass-by traffic for businesses. However, as access to specific business' account records is limited, this impact is difficult to quantify. Another common contention is that bypasses contribute to a reduction in overall crashes in the community and in the surrounding areas. Even though a large number of bypasses have been constructed in the State of Iowa over the past several years, their actual impact in terms of traffic safety has not been quantified.

Objectives

This study seeks answers to the following questions: (a) Are bypasses in Iowa associated with a reduction in crash frequencies and crash rates on the bypassed highway? (b) Do bypasses in Iowa introduce a reduction of overall crash frequencies and rates or do they merely shift crashes from the highways through the communities to the bypasses with no significant overall reduction?

Method

We obtained crash information from the Iowa DOT at 19 sites on which a bypass was constructed sometime during the past 23 years. We also obtained the same information at six sites used as comparison sites on which no bypasses were constructed at least until 2005. We them employed a Bayesian approach to estimating the association between the construction of the bypass and crash rates, while also accounting for other factors.

Results

The construction of bypasses in Iowa is associated with a significant increase in traffic safety both on the main road through town and on the combined main road and bypass roadway.     

Coverage of motor vehicle crashes with injuries in U.S. newspapers, 1999-2002

Problem

The aims of the study were to evaluate information on motor-vehicle crashes with injuries provided in newspaper reports and to assess the frequency of thematic and episodic reporting of motor-vehicle crashes.

Method

The study used Fatal Analysis Reporting System (FARS) derived variables to code a nationally representative sample of U.S. newspaper reports of motor-vehicle crashes from 1999-2002. A total of 473 newspaper reports of motor-vehicle crashes with injuries were included. Information on the crash event, people involved, and vehicles was extracted. The reports were coded for episodic and thematic news framing.

Results

A majority of newspaper reports used episodic framing. The majority of reports included information on the type of crash, but characteristics about people and vehicles were rarely reported.

Discussion

Lack of information in newspapers makes them an incomplete source from which to influence public perceptions and attitudes.

Impact on industry

This provides an opportunity for news print media to improve public health content.

Impact on industry

Newspapers represent an important source of public information; they are, however, an incomplete source [Voight, B., Lapidus, G., Zavoski, R., & Banco, L. (1998). Injury reporting in Connecticut newspapers. Injury Prevention, 4, 292-294.; Baullinger, J., Quan, L., Bennett, E., Cummings, P., & Williams, K. (2001). Use of Washington state newspaper for submersion injury surveillance. Injury Prevention, 7, 339-342]. To increase the accuracy of information provided to the public through media sources, there is a need for increased communication between public health professionals and reporters.The results of this study raise concerns about the contents of motor-vehicle crash information provided in newspapers and suggest that newspapers do not provide information to allow public perception to be in accord with the importance of motor-vehicle crash injuries and health promoting actions to reduce risk of injury. More balanced and detailed information in newspapers would provide an opportunity for news print media to improve public health programs and public perception about the impact of motor-vehicle crashes on safety for all.     

Injury crashes with bicyclists at roundabouts: influence of some location characteristics and the design of cycle facilities

Problem

Previous research indicated that conversions of intersections into roundabouts appear to increase the number of injury crashes with bicyclists. However, it was assumed that the effectiveness of roundabouts could vary according to some differences in design types of cycle, facilities and other geometrical factors.

Method

Regression analyses on effectiveness-indices resulting from a before-and-after study of injury crashes with bicyclists at 90 roundabouts in Flanders, Belgium.

Results

Regarding all injury crashes with bicyclists, roundabouts with cycle lanes appear to perform significantly worse compared to three other design types (mixed traffic, separate cycle paths, and grade-separated cycle paths). Nevertheless, an increase of the severest crashes was noticed, regardless of the design type of the cycle facilities. Roundabouts that are replacing signal-controlled intersections seem to have had a worse evolution compared to roundabouts on other types of intersections.

Impact on industry

The results might affect design guidelines for roundabouts, particularly for the accommodation of bicyclists.     

Trends and characteristics of animal-vehicle collisions in the United States

Introduction

Since 1990, fatal animal-vehicle collisions (AVCs) in the United States have more than doubled. This paper examines annual AVC trends in the United States over a 19-year period, seasonal and diurnal patterns of AVC risk, the geographic distribution of crash risk by state, and the association between posted speed limit and AVC crash risk in darkness.

Method

AVCs were compiled from the Fatality Analysis Reporting System (FARS) and the General Estimates System (GES) for the years 1990-2008 to examine annual crash trends for fatal and nonfatal crashes. Seasonal trends for fatal AVCs were examined with the aggregated FARS dataset; seasonal trends for fatal and nonfatal AVCs were also examined by aggregating four years of Michigan crash data. State-by-state distributions of fatal AVCs were also described with the aggregated FARS dataset. Finally, the relationship between posted speed limit and the odds that a fatal or nonfatal AVC occurred in darkness were examined with logistic regressions using the aggregated FARS and Michigan datasets.

Results

Between 1990 and 2008, fatal AVCs increased by 104% and by 1.3 crashes per trillion vehicle miles travelled per year. Although not all AVCs involve deer, daily and seasonal AVC crash trends follow the general activity pattern of deer populations, consistent with prior reports. The odds that a fatal AVC occurred in darkness were also found to increase by 2.3% for each mile-per-hour increase in speed; a similar, albeit smaller, effect was also observed in the aggregated Michigan dataset, among nonfatal crashes.

Conclusion

AVCs represent a small but increasing share of crashes in the United States. Seasonal and daily variation in the pattern of AVCs seem to follow variation in deer exposure and ambient light level. Finally, the relative risk that a fatal and nonfatal AVC occurred in darkness is influenced by posted speed limit, suggesting that a driver's limited forward vision at night plays a role in AVCs, as it does in pedestrian collisions.

Impact on Industry

The association between speed limit and crash risk in darkness suggests that AVC risk might be reduced with countermeasures that improve a driver's forward view of the road.     

Validity of police-reported alcohol involvement in fatal motor carrier crashes in the United States between 1982 and 2005

Objective

To examine the validity of police-reported alcohol data for drivers involved in fatal motor carrier crashes.

Material and Methods

We determined the availability of blood alcohol concentration (BAC) and police-reported alcohol data on 157,702 drivers involved in fatal motor carrier crashes between 1982 - 2005 using Fatality Analysis and Reporting System (FARS) data. Drivers were categorized as motor carrier drivers if they operated a vehicle with a gross vehicle weight rating of greater than 26,000 pounds. Otherwise, they were classified as non motor carrier drivers. The sensitivity and specificity of police-reported alcohol involvement were estimated for both driver types.

Results

Of the 157,702 drivers, 18% had no alcohol information, 15% had BAC results, 42% had police-reported alcohol data, and 25% had both. Alcohol information varied significantly by driver, crash, and vehicle characteristics. For example, motor carrier drivers were significantly more likely (51%) to have BAC testing results compared to non motor carrier drivers (31%) (p < 0.001). The sensitivity of police-reported alcohol involvement for a BAC level ≥ 0.08 was 83% (95% CI 79%, 86%) for motor carrier drivers and 90% (95% CI 89%, 90%) for non motor carrier drivers. The specificity rates were 96% (95% CI 95%, 96%) and 91% (95% CI 90%, 91%), respectively.

Conclusions

The sensitivity and specificity of police-reported alcohol involvement are reasonably high for drivers involved in fatal motor carrier crashes. Further research is needed to determine the extent to which the accuracy of police-reported alcohol involvement may be overestimated because of verification bias.

Impact on the Industry

Based on the results of this study, the federal government should continue to work with states to strengthen their strategies to increase chemical testing of all drivers involved in fatal crashes.