Presentation on September 29, 2016, at the 4th Annual BayesiaLab Conference:
Using Bayesian Networks to identify and quantify factors affecting injury severity of young drivers involved in single-vehicle crashes occurring within curves on rural two-lane roads in Louisiana
Dr. Cory Hutchinson
Director, Highway Safety Research Group, Louisiana State University
Cory Hutchinson currently serves as the Director for the Highway Safety Research Group (HSRG) at Louisiana State University. He earned a MS in Quantitative Business Analysis, a MBA, and a PhD in Human Resource Education and Workforce Development from LSU. Within the HSRG, Cory oversees all IT related projects including business analytics, web site design, data quality analysis, electronic crash data collection, data reporting, disaster recovery, graphical information systems, business intelligence, and crash data integration. Cory also teaches undergraduate and graduate Business Intelligence courses within the College of Business at LSU.
This study investigates factors affecting young driver injury levels for single vehicle crashes occurring within curves on rural two-lane roads in Louisiana. Although the number of fatal and serious injury crashes involving young drivers is declining, young drivers are still overrepresented in crashes and crashes are still the leading cause of death for young drivers.
Driver injury prediction models are formulated using binary logistic regression and Bayesian Network (BN) modeling. Binary logistic regression models have commonly been used in safety studies to analyze injury levels of occupants involved in crashes over the past few decades. More recently, a few safety studies have begun to use BN models to evaluate injury levels.
This study identifies eight significant factors affecting youth driver injury levels: air bag, distracted, ejected, gender, protection system, substance suspected, violation, and most harmful event. Of these factors distracted, protection system, substance suspected, and violation are human factors which can be modified through educational programs.
While both models are able to identify statistical significant variables, more insight is gained from the BN model. For instance, both models found gender to be statistically significant. While the logistical regression model finds males are 0.751 times less likely to be injured than female, the BN finds gender only has a 0.02% direct effect on injury. The BN shows that it is not gender itself that affects driver injury level, but the different behavior characteristics of males versus females which affect injury levels. Males are less likely to wear seatbelts and more likely to be suspected of alcohol in crashes. It is these driver behaviors, not the gender of the driver, which affects injuries.
This study also has a number of theoretical and practical implications. As the first study to utilize BN modeling in evaluating driver injury levels in Louisiana, it expands the literature of BN models being used for analyzing injury levels in car crashes. The findings are also important to driver educational and safety professionals. By identifying factors affecting young driver injury levels, educational and training programs can be enhanced to target specific human behaviors to save more lives.