Graduation Year

2022

Document Type

Dissertation

Degree

Ph.D.

Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Civil and Environmental Engineering

Major Professor

Fred Mannering, Ph.D.

Committee Member

Qing Lu, Ph.D.

Committee Member

Pei-Sung Lin, Ph.D.

Committee Member

Mingyang Li, Ph.D.

Committee Member

Lu Lu, Ph.D.

Keywords

Accident analysis, heterogeneity modeling, prediction, unobserved heterogeneity

Abstract

Roadway accidents have long been a major cause of casualties among all roadway users including drivers and pedestrians. Researchers have used various advanced methodologies to precisely identify affecting factors to help policy makers implement safety measures that effectively mitigate such losses. For a variety of reasons such factors, especially those related to roadway users, are subject to constant changes over time. This work investigates, in depth, the effect of multiple variables on the injury severity levels of both drivers and pedestrians using the most recent and advanced methodological approaches.

First, the dissertation starts by investigating factors that significantly contribute to the injury severity of different drivers of different nationality backgrounds. Using the data from Riyadh, Saudi Arabia, a random parameters multinomial logit model of driver-injury severity was estimated to explore the effects of a wide range of variables on driver injury-severity outcomes. With three possible outcomes (no injury, injury, fatality), single-vehicle crashes involving domestic and international drivers are considered and were modeled separately. Model estimation results show that a wide range of factors significantly affect the injury severity outcomes in single-vehicle crashes; and that the influence these variables have on injury-severity probabilities vary considerably between domestic and international drivers. While Saudi Arabia is rather unique because of the large numbers of non-national drivers, the results suggest that different nationalities, with their different cultural, educational and behavioral backgrounds, may affect risk-taking behavior and resulting crash-injury severities. This leads to the conclusion that studying drivers as one group may be incorrect, and that points to the need to consider sub-populations. Second, the dissertation moves on to explore the differences between day and night pedestrian-injury severities in vehicle-pedestrian crashes in Kansas over a five-year period. Separate statistical models (random parameters logit models with possible heterogeneity in the means and variances of the random parameters) were estimated for day and night crashes to examine different pedestrian injury severity outcomes (no visible injury, moderate injury, and severe injury). Likelihood ratio tests were conducted to explore the temporal stability of the model estimations over different times of day and years. Many variables affecting injury severities were considered in model estimation and the findings indicate that the factors affecting pedestrian injury severities did change over time but that there is a clear day-night difference in the resulting injury severities of pedestrians, with nighttime crashes consistently resulting in more severe injuries overtime. This suggests policies and technologies that seek to essentially replicate daytime conditions (improved illumination, infrared pedestrian detection in vehicles, etc.) in nighttime conditions could have considerable safety benefits.

Predictive analysis was extensively performed in both parts of the dissertation to compare and fully understand the variations between different models. For the first part, model estimation results of each type of driver (nationality background) were used to predict what the injury severities of other type of drivers would have been and vice versa. Results show that domestic drivers were found to be more involved in severe crashes even when controlling for other crash characteristics compared to international drivers. For the second part, out-of-sample prediction simulations are used to provide estimates of the potential benefits of such nighttime mitigation policies and technologies, as well as how daytime/nighttime pedestrian injury severity probabilities have been changing over time. The findings of the predictive comparison reveal some temporal stability between day and night forecasts, with using daytime parameter estimates to predict nighttime probabilities overestimates no visible injury and underestimates severe injury in all years analyzed. However, there was a variability in the prediction simulation results when estimating the probabilities of each time of day in multiple subsequent years, indicating clear evidence of model parameter shifts over time.

Download

Share

COinS