Analysis of Crash Frequency and Crash Severity in Thailand: Hierarchical Structure Models Approach

Currently, research on the development of crash models in terms of crash frequency on
road segments and crash severity applies the principles of spatial analysis and heterogeneity due
to the methods’ suitability compared with traditional models. This study focuses on crash severity
and frequency in Thailand. Moreover, this study aims to understand crash frequency and fatality.
The result of the intra-class correlation coefficient found that the spatial approach should analyze the
data. The crash frequency model’s best fit is a spatial zero-inflated negative binomial model (SZINB).
The results of the random parameters of SZINB are insignificant, except for the intercept. The crash
frequency model’s significant variables include the length of the segment and average annual traffic
volume for the fixed parameters. Conversely, the study finds that the best fit model of crash severity
is a logistic regression with spatial correlations. The variances of random effect are significant such as
the intersection, sideswipe crash, and head-on crash. Meanwhile, the fixed-effect variables significant
to fatality risk include motorcycles, gender, non-use of safety equipment, and nighttime collision.
The paper proposes a policy applicable to agencies responsible for driver training, law enforcement,
and those involved in crash-reduction campaigns.