Author ORCID Identifier

https://orcid.org/0000-0002-9123-0332

Year of Publication

2020

Degree Name

Master of Science in Civil Engineering (MSCE)

Document Type

Master's Thesis

College

Engineering

Department/School/Program

Civil Engineering

First Advisor

Dr. Reginald Souleyrette

Second Advisor

Dr. H M Abdul Aziz

Abstract

This study investigates the factors associated with single-vehicle crash injury severity using five years (2014 – 2018) of crash data from Kentucky, USA, using a mixed (random-parameter) logit model. We also explore the temporal heterogeneity of the correlated factors across different times of the day. Most crash-severity models assume that the estimated parameters remain temporally stable. For instance, the effect of light conditions on crash severity may differ based on the time of the crash occurrence—noon vs. dusk. The temporal instability of the factors due to the time-of-day variation can lead to (under) overestimating the parameters that influence the development and implementation of safety countermeasures—crash modification factors and safety performance functions.

To account for the temporal variations and associated instability, we estimated crash severity models for five periods of the day: 12 am – 5 am, 5 am – 9 am, 9 am – 2 pm, 2 pm – 7 pm, and 7 pm – 12 am. Each model considers five crash injury-severity outcomes: (a) fatal, (b) suspected serious injury, (c) suspected minor injury, (d) possible injury, and (e) property-damage only (as defined by the Kentucky State Police). Log-likelihood tests confirm the statistical validity of the time-of-day grouping of the crash severity models. The Chi-Square test-statistic indicates the significance of using five different models instead of a single aggregate model for the dataset. The used dataset is a collection of police crash investigation reports, and these reports were prepared after the crashes have occurred. So, data on traffic volume/ADT/AADT were not used for this study.

Further, the pseudo direct elasticity values are estimated to find the sensitivity of the explanatory variables—how much change in the probability of different injury outcomes. Explanatory variables such as age, gender, and lighting condition are incorporated into the models to examine the associated effects. Results show that being a female driver increases the probability of fatal injury by 76.85% for crashes occurring in the 5 am to 9 am window. Also, being a driver within the age-group of 50 years or more increases fatality probability by 49.07% for crashes occurring from 2 pm to 7 pm. Alcohol-involvement significantly increases the probability of fatal and severe injury in all the models (five-time periods). Further, our estimated results indicate that icy road surface, losing control of vehicles, and oversteering have a temporally stable effect (do not change across different time-of-the-day models) and are found to have a positive correlation with fatality and severe injury severity outcomes. On the other hand, variables such as drivers younger than 25 years, male drivers, streetlights turned on exhibit varying influence on the injury-severity outcome at different times of a day.

The findings of this research can be used to develop (and calibrate) Safety Performance Functions (SPF) and Crash Modification Factors (CMF) for the State of Kentucky. The time-of-day analyses will make the SPFs and CMFs more robust and flexible by accommodating temporal heterogeneity in the factors correlated with single-vehicle crash severity.

Digital Object Identifier (DOI)

https://doi.org/10.13023/etd.2020.464

Share

COinS