SWUTC Research Project Description
A Novel Approach to Modeling and Predicting Crash Frequency at Rural Intersections by Crash Type and Injury Severity Level
University: University of Texas at Austin
Principal Investigator:
Chandra Bhat
Department of Civil and Environmental Engineering
(512) 471-4535
Project Monitor:
Johanna Zmud
Senior Transportation Policy Researcher
RAND Corporation
200 South Hayes Street
Arlington, VA 22202-5050
Funding Source: USDOT and State of Texas General Revenue Funds
Total Project Cost: $75,900
Project Number: 600451-00077
Date Started: 1/1/13
Estimated Completion Date: 12/21/13
Project Summary
Project Abstract:
This study formulates and applies a novel approach for the joint modeling of crash frequency and crash type/injury severity. The crash data used in the analysis will be drawn from the Texas Department of Transportation crash incident files corresponding to a set of rural intersections in Texas, and will include yearly crash frequency information by collision type and by injury severity level for year 2010.
Project Objectives:
The objective of this study is to estimate number of crashes by collision type and by injury severity level per year at intersections, based on a flexible joint count-discrete choice model that accommodates heterogeneity effects (which correspond to varying effects of exogenous variables on crash frequency across different intersections due to unobserved effects). The crash data used in the analysis will be drawn from the Texas Department of Transportation crash incident files corresponding to a set of rural intersections, Texas, and will include yearly crash frequency information by collision type and by injury severity level for year 2010.
Using the results of the estimation process, we will compare and contrast the data fit and the predicted effects of variables from the proposed model and extant methods, and will recommend roadway design and other countermeasures to reduce the number and severity of traffic crashes.
Task Descriptions:
TASK 1: Preparation of data and sample for analysis
In this task, we will process the crash data base from the Texas Department of Transportation (TxDOT) Crash Records Information System (CRIS) for the year 2010, obtained from the records of the crashes reported by police and drivers. The CRIS contains the characteristics of crashes occurring at intersection as well as non-intersection locations in Texas, along with supplementary information on road design and geometric variables. For the current study, crashes at rural intersection locations will be extracted out from the CRIS data base. This task will be a critical first step in our process of developing the novel approach for the joint modeling of crash frequency and crash type/injury severity. In particular, the descriptive statistics will help inform the specification of the dependent variables as well as the independent variables.
TASK 2: Develop the model formulation
The problem with the existing studies that have examined crash counts by injury severity/crash type is that they ignore the joint nature of crash count and injury severity/crash type. In this task we will develop a modeling framework to explicitly link a count data model (for crash counts) with an event type multinomial choice model (for injury severity/crash type). Further, the proposed framework will accommodate varying effects of exogenous variables on crash frequency across different intersections due to unobserved effects.
TASK 3: Model estimation and specification analysis
The purpose of this task is to estimate a model to predict number of crashes by collision type and by injury severity level per year at intersections, explicitly considering the joint nature of the phenomenon under study. This will include estimation code development and testing a variety of different model specifications. The final model specification will be based on a systematic process of eliminating variables found to be statistically insignificant, intuitive considerations, parsimony in specification, and results from earlier studies. Different variable specifications, functional forms of variables as well as interaction variables will be examined.
TASK 4: Model application for policy analysis
The joint model estimated in this study can be used to examine the effects of policy actions that involve a change in intersection design and traffic characteristics over time. In this task, we will compare and contrast the data fit and the predicted effects of variables from the proposed model and extant methods
TASK 5: Prepare and submit final report
This task will prepare and deliver a final research report documenting all aspects of the research effort, including descriptive statistical analysis, complete details of the conceptual approach as well as the econometric methods used, results, and recommendations.
Implementation of Research Outcomes:
In this study, researchers formulate and apply a novel approach for the joint modeling of crash frequency and crash type/injury severity at rural intersections in Central Texas that explicitly models the effects of variables on each of these dimensions, while also accommodating the joint nature of these two dimensions. The empirical results clearly reveal the benefits, both in terms of capturing flexibility in variable effects and data fit, to adopting the proposed structure. From a substantive standpoint, the results underscore the important effects of intersection design and major road characteristics in determining the number of crashes in each category.
Products developed by this research:
New Methodology: A new methodology has been developed for jointly modeling crash counts and injury severity.
Thesis Developed: A Novel Approach to Modeling and Predicting Crash Frequency at Rural Intersections by Crash Type and Injury Severity Level, Jun Deng, MS Thesis, Department of Civil, Architectural and Environmental Engineering, The University of Texas at Austin, December 2013.
Impacts/Benefits of Implementation:
The joint modeling developed by this research can help many other fields where the dependent variable takes the form of a multivariate count.
The research results will impact decision making by allowing transportation analysts to prioritize countermeasures to reduce crashes and their severity.
Web Links:
Project Final Report