Citation

Khodadadi A, Shirazi M, Geedipally S, Lord D. Transportmetrica A: Transp. Sci. 2023; 19(3): e2062480.

Copyright

(Copyright © 2023, Informa - Taylor and Francis Group)

DOI

10.1080/23249935.2022.2062480

PMID

unavailable

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.

Language: en

Keywords

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