| Summary: | 碩士 === 國立交通大學 === 交通運輸研究所 === 99 === For two-vehicle crash accidents, one of the parties involved in an accident with more serious violation is termed as the first party. Most of the previous studies only consider the severity levels of the first party or combine the two parties in a whole. However, the severity levels of two parties involved in the same accident might be
rather different, contributed by different driving behaviors, vehicle characteristics, traffic environment and other risk factors of both parties. Undoubtedly, to consider the severity levels of two parties along with corresponding factors is imperative for obtaining more insights from crash data and proposing more effective safety
improvement strategies accordingly.
The severity levels of two parties along with corresponding contributory factors can not be separately modeled, since these factors are usually closely correlated. In view of this, this study employs bivariate generalized ordered probit (BGOP) model to modeling the severity levels of both parties simultaneously without losing
important relevant crash information of both parties. In addition, the threshold function of BGOP can be calibrated during the model estimation process to depict model heterogeneity and to provide more insights for severity classification.
To validate the applicability of the BGOP to severity modeling of two parties and to investigate their contributory factors, a total of 2,661 two-vehicle accidents at signalized intersections in Taipei City during 2008 and 2009 were collected for model estimation, in which the first party was identified as the major traffic regulation
violator in comparing to another party. The estimated results show that the BGOP model, which relaxes the assumption with fixed threshold values of severity levels, can not only perform better in terms of log-likelihood values and prediction errors, but also consider the effects of the risk factors to the severity levels of two parties.
Moreover, significant and positive correlation coefficients for both parties are also found, suggesting the necessity of simultaneous modeling of the severity levels of both parties. It is also interesting to note that the illegal driving behaviors (e.g.
drunk driving) of the first party significantly contribute to the severity level of the second party. Some corresponding safety improvement strategies are then proposed accordingly.
Keywords: Signalized intersections, Severity level, Bivariate generalized ordered
probit
|
|---|
