Application of a random effects negative binomial model to examine tram-involved crash frequency on route sections in Melbourne, Australia

https://doi.org/10.1016/j.aap.2016.03.012Get rights and content

Highlights

  • Key factors influencing tram-involved crash frequency was identified.

  • A random effects negative binomial model was adopted.

  • Major crash contributing factors were tram stop spacing and tram service frequency.

  • Other factors were tram route section length and general traffic volume.

  • The results showed positive road safety benefits of tram priority measures.

Abstract

Safety is a key concern in the design, operation and development of light rail systems including trams or streetcars as they impose crash risks on road users in terms of crash frequency and severity. The aim of this study is to identify key traffic, transit and route factors that influence tram-involved crash frequencies along tram route sections in Melbourne. A random effects negative binomial (RENB) regression model was developed to analyze crash frequency data obtained from Yarra Trams, the tram operator in Melbourne. The RENB modelling approach can account for spatial and temporal variations within observation groups in panel count data structures by assuming that group specific effects are randomly distributed across locations. The results identify many significant factors effecting tram-involved crash frequency including tram service frequency (2.71), tram stop spacing (−0.42), tram route section length (0.31), tram signal priority (−0.25), general traffic volume (0.18), tram lane priority (−0.15) and ratio of platform tram stops (−0.09). Findings provide useful insights on route section level tram-involved crashes in an urban tram or streetcar operating environment. The method described represents a useful planning tool for transit agencies hoping to improve safety performance.

Introduction

Public transport is becoming more important as mobility, accessibility, and environmental problems are increasing in cities around the world (Vuchic, 1981, Gakenheimer, 1999, Fouracre et al., 2003). In addition to bus and rail, tram and light rail transit systems are also available in different parts of the world; especially in Australia, North America, many European cities and parts of Asia and Africa (Topp, 1999, Fouracre et al., 2003). Trams/streetcars have a number of attractive features including their high passenger capacity, good comfort, and very low emission of pollutants compared to other transport systems (Anna and Bruce, 2001, Cliche and Reid, 2007).

However trams present a range of inherent safety issues regarding their design and operational characteristics, since they are large and heavy vehicles operating in confined, mixed and complex environments with pedestrians and cars. Even at low speed trams have been identified to have high crash risks compared to other vehicles (Grzebieta et al., 1999). Previous studies have also identified that trams impose more crash risk at intersections and along arterials than buses, and this is likely due to the difference in methods of operation between buses and trams (Cheung et al., 2008, Shahla et al., 2009). Also trams are given priority along lanes, at intersections and at tram stops to improve tram travel time, and to provide efficient and reliable service to passengers (Yarra Trams, 2010, Currie et al., 2012). However implementation of tram priority measures adjust the nature of road spaces and have road safety impacts on all road users (Naznin et al., 2015a, Naznin et al., 2015b).

Several macro level transit crash frequency models have been developed in previous studies; most focus on bus transit, and evaluate key factors associated with bus-involved collisions (Quintero et al., 2013, Goh et al., 2014). Only two previous studies have attempted to identify the factors associated with tram-involved collisions; but rather than considering tram-involved collisions solely, they combined both tram and bus-involved collisions to identify transit-involved crash causation factors (Cheung et al., 2008, Shahla et al., 2009). So the relative contribution of factors influencing tram-involved collisions are not clear. Cheung et al. (2008) developed zonal level transit-involved collision prediction models for urban transit in Toronto, Canada using a negative binomial regression structure. The results show that vehicle kilometers travelled, transit kilometers travelled, arterial road length, transit stop density, percent of near side stops and average posted speed are significant indicators for transit-involved collisions. Shahla et al. (2009) developed a negative binomial crash prediction model for intersections in Toronto, Canada using five years of transit collision data and pointed out that annual average daily traffic (AADT), public transit and pedestrian traffic volumes, turn movement treatments, public transit stop location, mode technology and availability of transit signal priority technology have significant associations with public transit related collisions at signalized intersections.

It is clear from the above discussion that the factors associated with tram-involved collisions at a macro level are still unclear. In addition, the previous studies only focused on North American transit network, so very little is known about the validity of this prediction models in other countries, where traffic and transit environment vary considerably. Also the above mentioned two relevant studies adopted the basic negative binomial structure to model transit-involved crash counts and ignored the spatial and temporal variations within the observation groups.

The most commonly used crash data modelling approaches are poisson and negative binomial (NB) regression models (Washington et al., 2010), as crash occurrences are discrete, non-negative and often infrequent and random. However the poisson model has several limitations and one important constraint is that the mean must be equal to variance. A number of previous studies have identified that the crash data are significantly over dispersed that means the variance exceeds the mean (Miaou, 1994, Abdel-Aty and Radwan, 2000). The Poisson model cannot deal with over dispersed crash data and resulted in an incorrect estimation of the likelihood of crash occurrence. This problem was overcome by introducing negative binomial model which can deal with over-dispersed crash data (Washington et al., 2010).

When crash data is collected from N (1, . ., N) locations for T (1, . ., T) time periods, the negative binomial model assumes that there are N*T independent observations. The time variant nature in crash data is being omitted by the NB model and standard error of estimated coefficients may be underestimated. One way to overcome this limitation by treating the crash data in a time series cross sectional panel data structure with N location groups and T time periods, and by considering individual effects in the NB model as suggested by Hausman et al. (1984). Hausman et al. examined both fixed and random definitions of the individual effects and developed random effects (RE) and fixed effects (FE) models. However fixed effects models do not allow location specific variations, but random effects models consider randomly distributed location specific variations. Shankar et al. (1998) has identified the random effects negative binomial (RENB) model as more appropriate for modelling median crossover crash frequencies in relation to geometric and traffic variables in Washington State. Another study by Chin and Quddus (2003) used the RENB model to investigate the relationship between crash occurrence and the geometric, traffic and control characteristics of signalized intersections in Singapore. Both of the studies have found the RENB model suitable for the variables (i.e. geometric and traffic) which are likely to have location specific effects. Also from analytical viewpoint, RENB models offer advantages in terms of model transferability and updating (Shankar et al., 1998, Lord and Mannering, 2010, Washington et al., 2010).

For the present study the random effects negative binomial (RENB) modelling approach was deployed to identify the key traffic, transit and route factors that influence tram-involved crash frequency at a route-section level,1 as the available crash data follows panel data structure and the variables are likely to have location specific effects.

This paper is structured as follows. The next section provides the details of sites and data source used in this study. Development of the statistical regression model is then provided. The paper closes with a discussion of the research findings and conclusions.

Section snippets

Site selection and data

Crash data used in this study was obtained from “Tram Incident Database” which is the ‘Yarra Trams’ crash reporting system (Yarra Trams, 2014). The database contains all reported tram-involved incidents on all tram routes in Melbourne. The incidents are reported in two categories as ‘A’ and ‘B’ type incidents. ‘A’ type incidents represent serious injury or death. The Police and Yarra Trams (the tram operator) always investigate category ‘A’ type crashes. Whereas the category ‘B’ type incidents

Model formulation

The aim of this research is to identify the key traffic, transit and route factors that influence tram-involved crash frequencies along tram route sections. A wide range of crash frequency models have been developed over past decades to gain a better understanding of the factors that affect the crash causality and frequency depending upon the characteristics of available crash data and associated variables (Washington et al., 2010).

The most commonly adopted crash data modelling approach, the

Results and discussions

Table 2 presents the matrix of the Pearson’s correlation coefficients among all variables considered for the model. The results show that none of the variables are highly correlated with other variables as the correlation coefficients are less than ±0.7 (Hinkle et al., 2003, Mukaka, 2012).

Table 3 presents the parameter estimates obtained from both the random effects negative binomial (RENB) and the negative binomial (NB) model using the maximum likelihood technique in STATA version 13 (

Conclusion

Previous research is limited in the field of investigating the factors influencing tram-involved crash frequency at a macro level. The study aims to explore the safety effects of key traffic, transit and route factors on tram-involved crash frequency for tram route sections in Melbourne, Australia. A random effects negative binomial (RENB) regression model was adopted for this study to model crash frequency in relation to variables, as the available crash data followed a panel data structure

Acknowledgements

This research is a part of wider Australian Research Council Industry Linkage Program project LP100100159, ‘Optimizing the Design and Implementation of Public Transport Priority Initiatives’ Institute of Transport Studies, Monash University in association with the Transport Research Group, University of Southampton, UK. The Principal Chief Investigator is Professor Graham Currie, the Chief Investigator is Professor Majid Sarvi and the Partner Investigator is Dr. Nick Hounsell.

The authors would

References (40)

  • C. Cheung et al.

    Models for safety analysis of road surface transit

    Transp. Res. Rec.: J. Transp. Res. Board

    (2008)
  • D. Cliche et al.

    Growing Patronage-Think Tram?

    International Conference Series on Competition and Ownership in Land Passenger Transport—Hamilton Island

    (2007)
  • G. Currie et al.

    Vehicle and pedestrian safety at light rail stops in mixed traffic

    Transp. Res. Rec.: J. Transp. Res. Board

    (2010)
  • G. Currie et al.

    Managing trams and traffic at intersections with hook turns

    Transp. Res. Rec.: J. Transp. Res. Board

    (2011)
  • G. Currie et al.

    Active transit signal priority for streetcars: experience in melbourne, Australia, and Toronto, Canada

    Transp. Res. Rec.: J. Transp. Res. Board

    (2008)
  • G. Currie et al.

    Analysis and mitigation of safety issues at curbside tram stops

    Transp. Res. Rec.: J. Transp. Res. Board

    (2011)
  • G. Currie et al.
  • P. Fouracre et al.

    Mass rapid transit systems for cities in the developing world

    Transport Rev.

    (2003)
  • R. Grzebieta et al.

    Crash Compatibility of Trams

    (1999)
  • J.A. Hausman et al.

    Econometric Models for Count Data with an Application to the Patents-R&D Relationship

    (1984)
  • Cited by (39)

    • A deep generative approach for crash frequency model with heterogeneous imbalanced data

      2022, Analytic Methods in Accident Research
      Citation Excerpt :

      As shown in the Table 8, road length is positively associated with fatal and severe injury crash frequency (marginal effect: 0.02–0.04) at the 1% level of significance. Such finding is consistent with that of previous studies (Venkataraman et al., 2013; Huang et al., 2016; Naznin et al., 2016; Guo et al., 2020a; Guo et al., 2019; Guo et al., 2018). Additionally, effect of road length on the fatal and severe injury crash frequency is randomly distributed (with standard deviation of 0.08–2.30).

    View all citing articles on Scopus
    View full text