Predicting Crash Rate Using Logistic Quantile Regression With Bounded Outcomes

Abstract

Various approaches and perspectives have been presented in safety analysis during the last decade, but when some continuous outcome variables take on values within a bounded interval, the conventional statistical methods may be inadequate, and frequency distributions of bounded outcomes cannot be used to handle it appropriately. Therefore, in this paper, a logistic quantile regression (QR) model is provided to fill this gap and deal with continuous bounded outcomes with crash rate prediction. The crash data set from 2003 to 2005 maintained by the Nevada Department of Transportation is employed to illustrate the performance of the proposed model. The results show that average travel speed, signal spacing, driveway density, and annual average daily traffic on each lane are significantly influencing factors on crash rate, and logistic QR is verified as an alternative method in predicting crash rate.