LAUSR

INTRODUCTION There have been a number of studies that have led to the development of safety risk assessment models to quantify the probability of crash frequencies on roadway facilities (both at micro- and macro-levels), over a specified time period. However, past research has rarely focused on heterogeneous traffic conditions in developing countries. METHOD This paper puts forward several models related to the traditional count approach to estimate crash frequency at a micro-level in a non-lane based bi-directional heterogeneous traffic environment. The paper shows the results of dispersion, zero-inflation, and random heterogeneity effects of different exogenous factors by comparing Poisson (P); Negative Binomial (NB); random and fixed parameter Zero-Inflated Poisson (ZIP); and Latent Class Models (LCM). The empirical analysis is based on data from a section of a major national highway in Bangladesh. The performance of the models was validated using different statistical goodness-of-fit measures that compared the estimated and observed average crash frequencies at individual locations. With the identification of the most significant influencing factors, the paper discusses the practical policy implications using partial effects analysis and spatial distribution. RESULTS It was found that the Zero-Inflated Random Parameter model gives a slightly better statistical fit when compared to alternative approaches. Practical applications: This micro-level modeling approach would be useful to identify significant crash risk factors; to prioritize road sections according to their safety level; to select site-specific appropriate counter-measures; and devise proactive target oriented safety management strategies. Thus, the results shown here could be a point of reference in the planning, designing, maintaining, and managing two-lane highway sections in developing countries.