|
Abstract:
|
Case-control methods are used to examine safe driving from naturalistic studies, where exposure to a secondary task are compared between driving episodes with and without a crash. Matched case-control approaches are favored over independent approaches because they control for confounding factors. By matching cases and controls using identified confounding factors (namely matching variables), it provides a more robust estimate of the outcome of interest. We proposed a data-based approach of matching variable selection using the Lasso (least absolute shrinkage and selection operator) regression method. We considered variables related to the driver (demographics, driving history, and perceptions), vehicle (type, age, power) and environmental conditions. A driver-level and an epoch-level model was developed. At the driver level, the number of violations and the study site were ranked highest. At the epoch level, intersection, pre-incident maneuver and traffic control were the top ranked based on the AUC (Area Under Curve) of a ROC curve. These are used as the matching variables for developing the conditional logistic regression model to assess crash risk.
|
