Date of Award

August 2020

Degree Type


Degree Name

Doctor of Philosophy



First Advisor

Xiao Qin

Committee Members

Robert Schneider, Jie Yu, Yin Wang, Wilkistar Otieno


Crash Data, Crash Injury Severity Analysis, Data Mining, Data Quality, Decision Trees, Responsibility Analysis


This dissertation explores, identifies, and evaluates a multitude of factors significantly affecting motor vehicle crashes involving pedestrians and bicyclists, commonly defined as vulnerable road users (VRUs). The methodologies are guided by the concept of safe behavior of different parties that are primary responsible for a crash, either a pedestrian, a bicyclist or a driver, pertaining to roadway design, traffic conditions, land use and built environment variables; and the findings are beneficial for recommending targeted and effective safety interventions.

The topic is motivated by the fact that human factors contribute to over ninety percent of the crashes, especially the ones involving VRUs. Studying the effect of road users’ behavior, their responses to the dynamics of traveling environment, and compliance rate to traffic rules is instrumental to precisely measure and evaluate how each of the investigated variables changes the crash risk. To achieve this goal, an extensive database is established based on data collected from sources such as the linework from topologically integrated geographic encoding and referencing, Google maps, motor vehicle accident reports, Wisconsin Information System for Local Roads, and Smart Location Dataset from Environmental Protection Agency. The crosscutting datasets represent various aspects of motorist and non-motorists travel decisions and behaviors, as well as their safety status. With this comprehensive database, intrinsic relationships between pedestrian-vehicle crashes and a broad range of socioeconomic and demographic factors, land use and built environment, crime rate and traffic violations, road design, traffic control, and pedestrian-oriented design features are identified, analyzed, and evaluated.

The comprehensive safety analysis begins with the structural equation model (SEM) that is employed to discover possible underlying factor structure connecting exogenous variables and crashes involving pedestrians. Informed by the SEM output, the analysis continues with the development of crash count models and responsible party choice models to respectively address factors relating to roles in a crash by pedestrians and drivers. As a result, factors contributing to crashes where a pedestrian is responsible, a driver is responsible, or both parties are responsible can be specified, categorized, and quantified. Moreover, targeted and appropriate safety countermeasures can be designed, recommended, and prioritized by engineers, planners, or enforcement agencies to jointly create a pedestrian-friendly environment.

The second aspect of the analysis is to specify the crash party at-fault, which provides evidence about whether pedestrians, bicyclists or drivers are more likely to be involved in severe crashes and to identify the contributing factors that affect the fault of a specific road user group. An extensive investigation of the available information regarding the crash (i.e., issued citations, actions/circumstances that may have played a role in the crash occurrence, and crash scenario completed by the police officer) are considered. The goal is to recognize and measure the factors affecting a specific party at-fault. This provides information that is vital for proactive crisis management: to decrease and to prevent future crashes. As a part of the result, a guideline is proposed to assign the party at-fault through crash data fields and narratives. Statistical methods such as the extreme gradient boosting (XGboost) decision tree and the multinomial logit (MNL) model are used. Appealing conclusions have been found and suggestions are made for law enforcement, education, and roadway management to enhance the safety countermeasures.

The third aspect is to evaluate the enhancements of crash report form for its effectiveness of reporting VRU involved motor vehicle crashes. One of the State of Wisconsin projects aiming to develop crash report forms was to redesign the old MV4000 crash report form into the new DT4000 crash report form. The modification was applied from January 1, 2017, statewide. The reason behind this switch is to resolve some matters with the old MV4000 crash report form, including insufficient reporting in roadway-related data fields, lack of data fields describing driver distraction, intersection type, no specification of the exact traffic barrier, insufficient information regarding safety equipment usage by motorists and non-motorists, unclear information about the crash location, and inadequate evidence concerning non-motorists actions, circumstances and condition prior to the crash. Hence, the new DT4000 crash form modified some existing data fields incorporated new crash elements and more detailed attributes. The modified and new data fields, their associated attribute values have been thoroughly studied and the effectiveness of improved data collection in terms of a better understanding of factors associated with and contributing to VRU crashes has been comprehensively evaluated. The evaluation has confirmed that the DT4000 crash form provided more specific, details, and useful about the crash circumstances.

Available for download on Saturday, September 03, 2022