Pedestrian crash analysis with latent class clustering method. (March 2019)

• Record Type: Journal Article
• Authors: Sun, Ming; Sun, Xiaoduan; Shan, Donghui
• Is Part Of: Accident analysis and prevention. Volume 124(2019)
• Page Start: 50
• Page End: 57
• Publication Date: 2019-03
• Subjects: Pedestrian safety -- Latent class -- Cluster analysis -- Severity -- Contributing factors; Accidents -- Prevention -- Periodicals; Accident Prevention -- Periodicals; Accidents -- Prévention -- Périodiques; 363.106
• DOI: 10.1016/j.aap.2018.12.016
• Languages: English
• ISSNs: 0001-4575

Abstract

Highlights: Fatal and severe crashes are closely linked to pedestrian alcohol or drugs involvement, pedestrian age over 65, and adverse weather conditions. Crossing, entering road roadway away from intersections significantly increase the likelihood of fatal pedestrian crash. High speeds in rural areas and dark lighting condition are more likely to cause fatally/seriously injured pedestrian crashes. The combined MNL and LCC method helps identify contributing factors that are not evident when using the data as a whole. Abstract: Pedestrians are the most vulnerable users of the highway transportation system. While encouraging "Green Transportation", a concerning fact emerges in the United States: pedestrian deaths are climbing faster than motorist fatalities, reaching nearly 6000 in 2016 - the highest in over two decades. In 2015, pedestrian fatalities reached 110, 14.6% of total traffic fatalities in Louisiana for that year. Consequently, the Louisiana pedestrian fatality rate (fatalities per 100, 000 population) was 2.18, exceeding the U.S. average of 1.67. In an effort to effectively reduce the pedestrian crashes, this paper investigates this problem for Louisiana. However, with the heterogeneity of provided crash data, it is difficult to identify major causation that contribute to these crashes. This study will reveal the findings of the Latent Class Cluster (LCC) model, utilizing it as a preliminary tool for the segmentation of 14, 236 pedestrian crashes in Louisiana, between the years of 2006–2015. Next, Multinomial Logit (MNL) models are used to identify the main factors in pedestrian crash severity, shown in the original dataset, by further analyzing the clusters previously obtained by the LCC model. The results shed lights on the crash characteristics that are not apparent without these combined data analysis methods. Certain variables that have not been identified as significant in whole data analysis are identified as significant for a specific cluster, such as pedestrian crossing and entering roads, crash hours between midnight to 6 pm, dark-unlighted conditions, dark-lighted conditions, and non-intersection locations. The study suggests that the LCC regression approach can reveal important, formerly hidden relationships in traffic safety analyses.