Exploring the overall and specific crash severity levels at signalized intersections

https://doi.org/10.1016/j.aap.2004.11.002Get rights and content

Abstract

Many studies have shown that intersections are among the most dangerous locations of a roadway network. Therefore, there is a need to understand the factors that contribute to injuries at such locations. This paper addresses the different factors that affect crash injury severity at signalized intersections. It also looks into the quality and completeness of the crash data and the effect that incomplete data has on the final results. Data from multiple sources have been cross-checked to ensure the completeness of all crashes including minor crashes that are usually unreported or not coded into crash databases. The ordered probit modeling technique has been adopted in this study to account for the fact that injury levels are naturally ordered variables. The tree-based regression methodology has also been adopted in this study to explore the factors that affect each severity level. The probit model results showed that a combination of crash-specific information and intersection characteristics result in the highest prediction rate of injury level. More specifically, having a divided minor roadway or a higher speed limit on the minor roadway decreased the level of injury while crashes involving a pedestrian/bicyclist and left turn crashes had the highest probability of a more severe crash. Several regression tree models showed a difference in the significant factors that affect the different severity types. Completing the data with minor non injury crashes improved the modeling results and depicted differences when modeling the no injury crashes.

Introduction

Intersections are a common place for crashes, which may be due to the fact that there are several conflicting movements as well as a myriad of different intersection design characteristics. Intersections also tend to experience severe crashes due to the fact that some of the injurious crashes such as angle and left turn collisions commonly occur at intersection. During 1999, there were 243,409 crashes recorded in the Florida Crash Database. Of these, 98,756 crashes (about 40%) occurred at or were influenced by a signalized intersection. About 9.6 crashes occur at signalized intersections per year compared to 2 per year where stop or yield signs control traffic. The factors affecting injury levels of crashes occurring at signalized intersections are not well understood. Therefore, there is a need to identify the effects that certain geometric and crash-specific aspects have on the injury level of crashes occurring at signalized intersection.

Furthermore, when a crash occurs and the local police department is notified, the responding officer will determine whether to fill out a long- or short-form crash report. For instance, if a crash involves an injury or a felony (e.g., hit and run), the crash must be filed on a long-form. If a crash involved only property damage (a minor crash with no injuries), usually it is up to the officer to report it on a long- or a short-form. Crash forms are then forwarded to the respective counties. From here, only the crashes reported on long-forms are forwarded onto the Florida Department of Transportation (FDOT) and the Department of Highway Safety and Motor Vehicles (DHSMV), which maintain electronic records based on only crashes reported on long-forms. By focusing on state-maintained databases, most non injury crashes are neglected. Ignoring these non injury crashes will bias the distribution of crash injury severity levels. Therefore, for this study, two databases were considered. The first database consisted of crashes from the state agencies (reported only on long-forms) and was considered the restricted dataset. The second database consisted of all crashes (reported on both long- and short-forms) and was completed by obtaining all crashes reported on short-forms (we here refer to it as the complete dataset). Furthermore, multiple databases were cross-checked to ensure that the crashes reported on long-forms are as complete as possible (it is worth noting that even the complete data set does not include non reported crashes).

This study explores the hypothesis that crash injury levels are affected by both crash-specific and intersection-specific variables. Furthermore, the authors investigate the significant differences in the important crash-related factors between models based solely on crashes reported on long-forms and models based on crashes reported on both long- and short-forms (i.e., models based on restricted and complete datasets). Additionally, several databases were cross-checked to ensure the completeness of our data. The authors anticipate that these results will provide a significant contribution to the area of safety at signalized intersections as well as consider the possible consequences of the common practice of analyzing restricted datasets. Separate tree regression models for crashes of every severity level were also estimated to identify the significant factors that affect each. Injury levels are categorized as follows: no injury (property damage only), possible injury (no visible signs of injury), non-incapacitating injury (any visible injuries, e.g., bruises or limping), incapacitating injury (any visible signs of injury and the person is carried from the scene) and fatal injury (an injury sustained in a motor vehicle crash that results in death within 90 days).

Section snippets

Relevant studies

Many previous studies have used the ordered probit modeling methodology to study injury severity at different roadway locations. For example, Abdel-Aty (2003) applied the ordered probit models to predict crash injury severity on roadway sections, signalized intersections and toll plazas. Jianming and Kockelman (2004) used the ordered probit technique to predict injury severity based on factors including traffic, roadway and occupant characteristics and weather conditions at the time of a crash

Data collection

Data collection began in early 2003 when several counties across the midsection of the State of Florida were contacted for cooperation. Four agencies that comprise a majority of Central Florida were identified and contacted: Brevard County, Seminole County, City of Orlando and Hillsborough County. Each jurisdiction provided drawings for several hundred intersections and each drawing was then individually examined and identified by the authors. Information obtained from each drawing was the

Model definition

Due to the fact that some variables are naturally ordered, such as the severity level in a motor vehicle crash, various types of models can be specified for these types of data. The data for this research included crash-specific information such as the injury type, which was categorized into one of five groups: no injury, possible injury, non-incapacitating injury, incapacitating injury and fatal injury. These groups were then ranked from 0 to 4 with no-injury corresponding to the lowest level.

Approach

Hierarchical tree-based regression (HTBR) was used to estimate the expected number of crashes for each injury severity level. While the ordered probit models illustrated the significant factors that affect the overall severity levels whether accounting for collision types, intersection characteristics, or both, the HTBR models will focus on identifying the factors that affect each severity level, including no injury, one at a time. This method involves splitting the data into branches on a tree

Conclusion

This paper explored the severity levels of crashes at signalized intersections. The ordered probit modeling methodology has been adopted to analyze the overall severity levels, while regression tree models looked into each severity type and the factors that affect each of them.

Ordered probit models were created in this study for three different types of variables; one based on collision types, another for intersection characteristics and the last for a combination of significant variables. Both

Acknowledgement

The authors wish to thank the Florida Department of Transportation for funding this research. All opinions and results are those of the authors.

References (11)

There are more references available in the full text version of this article.

Cited by (0)

View full text