Abstract
This paper attempts to provide additional insights of using a bivariate ordered probit model to simultaneously examine influential factors that affect injury severity (IS) and collision type (CT) of traffic accidents occurring on intersections of national highways in Japan. The effectiveness of the bivariate model is re-confirmed using a large-scaled traffic accident data. The corrections between error terms of IS and CT are statistically significant, implying that there are some unobserved factors that simultaneously affect IS and CT, but are not well captured in traditional accident surveys. Six types of collisions are dealt with and their influential factors are examined with respect to both primary and secondary parties. A complicated structure of influential factors is clarified, and it is expected that such complexity cannot be properly captured based on the widely applied cross-aggregation method in practice.
