Doctor of Philosophy
Date of Defense
Donald C. Sweeney II, PhD.
L. Douglas Smith
Motor vehicle crashes are a leading cause of death in the United States, cost Americans $277 billion annually, and generate serious psychological burdens. As a result, extensive vehicle safety research focusing on the explanatory factors of crash severity is undertaken using a wide array of methodological techniques including traditional statistical models and contemporary data mining approaches. This study advances the methodological frontier of crash severity research by completing an empirical investigation that compares the performance of popular, longstanding techniques of multinomial logit and ordinal probit models with more recent methods of decision tree and artificial neural network models. To further the investigation of the benefits of data analytics, individual models are combined into model ensembles using three popular combinatory techniques. The models are estimated using 2002 to 2012 crash data from the Missouri State Highway Patrol Traffic Division - Statewide Traffic Accident Records System database, and variables examined include various driver characteristics, temporal factors, weather conditions, road characteristics, crash type, crash location, and injury severity levels. The accuracy and discriminatory power of explaining crash severity outcomes among all methods are compared using classification tables, lift charts, ROC curves, and AUC values. The CHAID decision tree model is found to have the greatest accuracy and discriminatory power relative to all evaluated modeling approaches. The modeling reveals that the presence of alcohol, driving at speeds that exceed the limit, failing to yield, driving on the wrong side of the road, violating a stop sign or signal, and driving while physically impaired lead to a large number of fatalities each year. Yet, the effect of these factors on the probability of a severe outcome is dependent upon other variables, including number of occupants involved in the crash, speed limit, lighting condition, and age of the driver. The CHAID decision tree is used in conjunction with prior literature and the current Missouri rules of the road to provide better formulated driving policies. This study concludes that policy makers should consider the interaction of conditions and driver related contributing factors when crafting future legislation or proposing modifications in driving statues.
Bernard, Jill Marie, "An Application of Data Analytics to Outcomes of Missouri Motor Vehicle Crashes" (2015). Dissertations. 168.