Defining new exposure measures for crash prediction models by type of collision
Date of Completion
January 2008
Keywords
Transportation
Degree
Ph.D.
Abstract
Through accident prediction models, researchers have identified correlation between crash risk and many different explanatory factors, such as traffic volume, roadway geometries, temporal effects, driver characteristics, and and use, but cannot give definite safety effects of countermeasures. This dissertation proposes a methodology that accounts for crash causality by defining collision type categories, with the crashes in each sharing common contributing factors to support estimation of prediction models that offer a more accurate understanding of crash risk correlation for each collision type as well as suggesting appropriate countermeasures for reducing the predicted collisions. Furthermore, this dissertation intends to advance the state of crash prediction modeling by redefining crash exposure to consider the practical (or causal) relationship between traffic volume and crashes. Therefore, each crash category could conceivably use a different measure of exposure measurement for its distinct occurrence mechanism. Combining this with the idea of linking prediction models to crash causalities through the established collision type categories, the proposed exposure measurement can potentially explain with more accuracy the variation observed in crash risk due to the traffic flow state. ^ Studies using statistical methodologies of generalized linear models are carried out to evaluate the new exposure definitions. The new crash exposure for opposite-direction collisions including head-on and sideswipe is defined as the number of times vehicles traveling in opposite directions meet. Vehicle time spent following is then proposed as a new exposure definition for same-direction collisions. This new exposure is found to have a linear relationship with the same-direction crashes while the collision categories defined based on contributing factors are also considered in the models. This linearity implies a well defined traffic flow intensity and state by the new exposure, and provides for the possibility of a constant crash rate by dividing the number of rear-end crashes by this exposure measurement. Moreover, finding VTSF being linearly related to the rear-end crash count further suggests categorizing crashes by crash causality rather than the nature of the collision and defining the exposure measurement based on the contributing factor can lead to more accurate and explainable prediction models. ^
Recommended Citation
Zhang, Chen, "Defining new exposure measures for crash prediction models by type of collision" (2008). Doctoral Dissertations. AAI3293728.
https://digitalcommons.lib.uconn.edu/dissertations/AAI3293728