| dc.description.abstract | This dissertation had three main objectives related to improving road safety by investigating factors that contribute to injury severity in different types of single-vehicle crashes. The first objective was to develop a generalized ordered logit model to examine factors affecting injury severity of occupants in single-vehicle rollover crashes using 5 years of U.S. crash data from 2012-2016. Results showed likelihood of serious/fatal injuries increased in rollovers with occupant ejection, speeding, higher speed limits, roadside/median rollovers, undulating terrain, blacktop surfaces, rural roads, evenings, weekdays, older drivers, lack of occupant protection, previous driver crashes, distracted/aggressive driving, and passenger cars. Airbag deployment reduced serious/fatal injury risk. Regional variations also impacted injury severity.
The second objective identified high-risk areas for lane departure crashes on rural North Dakota roads using techniques like Global/Local Moran's I, network kernel density estimation (NetKDE), and emerging hotspot analysis. While Global Moran's I indicated clustering, Local Moran's I revealed specific hot/cold spots. NetKDE quantified and prioritized crash clusters by density along roadways. Emerging hotspot analysis evaluated temporal patterns of hot/cold spots. This approach can guide deployments of education, enforcement, and infrastructure countermeasures.
The third objective used a mixed logit model to analyze factors contributing to injury severity in single-vehicle run-off-road (ROR) crashes for passenger cars, SUVs, and pickups. Common factors increasing injury risk were older driver age, impaired driving, no seatbelt, no airbag, high speeds, and older vehicles. However, driver age impacts were most pronounced for pickups. Seatbelts substantially mitigated injury severity across all vehicle classes. Passenger cars had a higher injury risk than SUVs/pickups, especially over 75 mph. Future research should examine additional factors stratified by vehicle class using larger datasets. | en_US |
