Exploring single vehicle crash severity on rural, two-lane highways with crash-level and occupant-level multinomial logit models

This thesis is conducted to compare a crash-level severity model with an occupant-level severity model for single-vehicle crashes on rural, two-lane roads. A multinomial logit model is used to identify and quantify the main contributing factors to the severity of rural, two-lane highway, single-vehicle crashes including human, roadway, and environmental factors. A comprehensive analysis of 5 years of crashes on rural, two-lane highways in Illinois with roadway characteristics, vehicle information, and human factors will be provided. The modeling results show that lower crash severities are associated with wider lane widths, shoulder widths, and edge line widths, and larger traffic volumes, alcohol-impaired driving, no restraint use will increase crash severity significantly. It is also shown that the impacts of light condition and weather condition are counterintuitive but the results are consistent with some previous research. Goodness of fit test and IIA (independence of irrelevant alternatives) test are applied to examine the appropriateness of the multinomial logit model and to compare the fit of the crash-level model with the occupant-level model. It is found that there are consistent modeling results between the two models and the prediction of each severity level by crash-level model is more accurate than that of the occupant-level model.