Accident modelling: An overview

The goal of this chapter is to provide an overview of key concepts related to modelling accident frequency, considering the random nature of traffic accidents. This includes topics such as the expected number of accidents (the 'true' yet unobserved measure of safety), regression-to-the-mean (RTM) and the safety-in-numbers phenomenon, which were briefly introduced in the chapter on accident data. The main objective here is to explain the foundational principles for designing and conducting safety evaluation studies, with regression modelling being the state-of-the-art tool. However, the RTM effect and the complex relationship between crashes and exposure factors pose challenges in accurately estimating expected accident frequency and, consequently, assessing the impact of safety interventions. Accurately calculating the long-term mean of accident frequency is essential for identifying the effects of safety countermeasures. This remains one of the most critical aspects of traffic safety analysis. Reliable accident frequency prediction models are fundamental for identifying hazardous locations and evaluating possible solutions to address safety issues. Therefore, this chapter provides an extensive discussion of the main research efforts since the introduction of modern road safety principles, largely credited to Ezra Hauer in 1997 [1]. It addresses the challenges posed by RTM and the non-linear relationship between accident frequency and exposure. The final section introduces the methodological approach for accident modeling, with a particular focus on tackling the RTM bias. The chapter has been inspired by the Ph.D. thesis 'A stochastic approach to safety management of roadway segment' of Carmelo D'Agostino [2].