Evaluating alternative variations of Negative Binomial–Lindley distribution for modelling crash data

Abstract

Several studies have reported the superior performance of the Negative Binomial–Lindley (NB-L) compared to the commonly used Negative Binomial distribution. Consequently, different parameterisations of the NB-L distribution have been introduced to further improve crash data modelling. However, little is known on how these models perform for different data domains. This study is documenting a comparative analysis among previously developed and two newly proposed parameterisations of the NB-L distribution, the negative binomial weighted Lindley (NB-WLindley) and the negative binomial quasi-Lindley (NB-QL). The results show that the NB-WLindley distribution performed better for the majority of data domains. Also, its generalised linear model (NB-WLindley GLM) showed superior statistical performance relative to the NB GLM and NB-L GLM. The results of this study contribute to the advancement of current predictive models used in transportation safety and provide insights for safety analysts and researchers when these models should be used.

Keywords:

• Crash prediction models
• negative binomial Lindley
• weighted Lindley
• Lindley parameterisations
• generalised linear model
• excess zero

Disclosure statement

The authors declare that they have no known competing financial interests or personal relationships that could have influenced the work reported in this paper.

Additional information

Funding

The study was funded by the A.P. and Florence Wiley Faculty Fellow [512697].