A Comparative Study of Count Models: Application to Pedestrian-Vehicle Crashes Along Malaysia Federal Roads

• Published in: Traffic injury prevention Vol. 14; no. 6; pp. 630 - 638
• Main Authors: Hosseinpour, Mehdi, Prasetijo, Joewono, Yahaya, Ahmad Shukri, Ghadiri, Seyed Mohammad Reza
• Format: Journal Article
• Language: English
• Published: Philadelphia, PA Taylor & Francis Group 2013; Taylor & Francis; Taylor & Francis Ltd
• Subjects: Accidents, Traffic - statistics & numerical data; Binomial Distribution; Biological and medical sciences; Comparative analysis; count models; Environment Design - statistics & numerical data; General populations; Humans; hurdle model; Malaysia; Medical sciences; Miscellaneous; model selection criteria; Models, Statistical; overdisperion; Pedestrians; Poisson Distribution; Prevention and actions; Public health. Hygiene; Public health. Hygiene-occupational medicine; Reproducibility of Results; Traffic accidents & safety; vehicle-pedestrian crashes; Walking - injuries; Asia; Malaysia; hurdle model; overdisperion; Vehicle driving; Government; Political aspect; Selection; Highway; Pedestrian; Traffic accident; count models; model selection criteria; Health policy; Criterion; Models; vehicle-pedestrian crashes; Application; Comparative study; Public health
• ISSN: 1538-9588; 1538-957X
• DOI: 10.1080/15389588.2012.736649

Objectives: The objective of this study was to examine the effects of various roadway characteristics on the incidence of pedestrian-vehicle crashes by developing a set of crash prediction models on 543 km of Malaysia federal roads over a 4-year time span between 2007 and 2010. Methods: Four count models including the Poisson, negative binomial (NB), hurdle Poisson (HP), and hurdle negative binomial (HNB) models were developed and compared to model the number of pedestrian crashes. Results and Conclusions: The results indicated the presence of overdispersion in the pedestrian crashes (PCs) and showed that it is due to excess zero rather than variability in the crash data. To handle the issue, the hurdle Poisson model was found to be the best model among the considered models in terms of comparative measures. Moreover, the variables average daily traffic, heavy vehicle traffic, speed limit, land use, and area type were significantly associated with PCs.