Home Explore Help
Sign In
ardo
/
Obsidian
1
0
Fork 0
Files Issues 0 Pull Requests 0 Wiki
Tree: 49b8810ad7
Branches Tags
Obsidian
/
Zotero
/
001_artiklid
/
Road Safety Indicators
/
moeinaddinianalyzingrelationshipsbetweennumber2015.md

moeinaddinianalyzingrelationshipsbetweennumber2015.md 65 KB
History Raw

category: literaturenote citekey: moeinaddinianalyzingrelationshipsbetweennumber2015 title: Analyzing the relationships between the number of deaths in road accidents and the work travel mode choice at the city level authors: "Moeinaddini, Mehdi; Asadi-Shekari, Zohreh; Sultan, Zahid; Zaly Shah, Muhammad" year: 2015 date: 2015-00-00 2015 doi: 10.1016/j.ssci.2014.09.015 publication: Safety Science url: "https://doi.org/10.1016/j.ssci.2014.09.015" zotero_key: VQINFARA zotero_storage: C2FADKD3 collections: transpordi_planeerimine folder: 001_artiklid/Road Safety Indicators firstAuthor: "Moeinaddini, Mehdi"

status: converted

Contents lists available at ScienceDirect

Safety Science

journal homepage: www.elsevier.com/locate/ssci

Analyzing the relationships between the number of deaths in road accidents and the work travel mode choice at the city level

Mehdi Moeinaddini a , Zohreh Asadi-Shekari b,⇑ , Zahid Sultan a , Muhammad Zaly Shah a

  • aDepartment of Urban and Regional Planning, Faculty of Built Environment, Universiti Teknologi Malaysia, Skudai, Johor, Malaysia
  • b Centre for Innovative Planning and Development (CiPD), Faculty of Built Environment, Universiti Teknologi Malaysia, Skudai, Johor, Malaysia

article info

Article history: Received 7 May 2014 Received in revised form 22 September 2014 Accepted 22 September 2014 Available online 10 October 2014

Keywords: Work travel mode choice Urban transport safety Crash fatality City level European cities

abstract

Currently, several efforts and strategies are used to reduce deaths in road accidents. However, only a few studies have considered the effectiveness of work travel mode choice on the number of accident-related fatalities at the city level. This study introduces a city-level (across cities) model to estimate the relationships between the number of deaths in road accidents (as the dependent variable) and several work travel mode choice indicators, including walking, cycling and public transport (as independent variables). Generalized linear modeling (GLM), which is a common technique for modeling crash data, was used to estimate this relationship. Data sets from various European cities were used to develop this city level model. Overall, the percentages of the journeys to work by public transport, motorcycle, bicycle and foot were effective variables in the model.

-2014 Elsevier Ltd. All rights reserved.

1. Introduction

According to World Health Organization (2013) data, road traffic crashes cause nearly 3400 deaths each day and injure or disable more than ten million people every year worldwide. Although approximately 70% of traffic-related deaths occur in developing countries (Augustus, 2012), nearly 28,000 road fatalities were reported in European countries in 2012 (European Commission, 2013). These statistics indicate that concerns about road traffic crashes are very serious in developing countries and still exist in European countries (Moeinaddini et al., 2014a). Among all of the road fatalities in the EU, approximately 50% are car occupants, 15% are motorcyclists, 10% are cyclists or moped riders, and 20% are pedestrians (European Commission, 2013). In the past two decades, bicycle use by adults has increased in some European cities. In some countries where cycling is common, such as the Netherlands, cyclists account for more fatal accidents than pedestrians do (Mohan, 2002).

In many countries, especially industrial countries, road traffic accidents are a major type of fatal work accident (Charbotel et al., 2001 and Charbotel et al., 2010; Murray, 2007). The proportion of work-related traffic fatalities among all accidental fatalities in Finland has increased by 63% (Statistics Finland, 1997). In 2012, the number of casualties in Britain was higher during working hours (RRCGB, 2013). The number of daily work trips is the highest in Mashhad City, Iran (Naderan and Shahi, 2010). Christoforou et al. (2010) found that approximately 93% of the individuals that are injured on working days in accidents at the A4-A86 junction in the Paris region are seriously injured. The annual crash number and traffic flow are higher during working hours according to observations that were made of 2000 km of French interurban motorways in 1997 and 1998 (Martin, 2002).

In the past several decades, a large number of empirical mode choice studies have considered a large range of factors for the transportation mode used to reach different destinations (university or school, work, favorite leisure activities and shopping). The majority of these studies considered several socio-economic and demographic factors, including age, gender, income, education level, employment status, family size, marital status, vehicle ownership, the number of workers in the household, the presence of children in the household, and the number of household motor vehicles (e.g., Whalen et al., 2013; Vega and Reynolds-Feighan, 2009; Can, 2013; Carse et al., 2013; Muhs, 2013). In addition, several studies considered the environmental factors that impact travel mode choice, such as the availability of parking, sidewalk density, bike density, street density, availability of park and ride facilities, shelter, weather and daylight (e.g., Whalen et al., 2013; Klöckner and Friedrichsmeier, 2011; Commins and Nolan, 2011). Furthermore, some trip characteristics, such as travel cost, travel time, travel distance, comfort, safety, car availability, public

⇑ Corresponding author. Tel.: +60 129410543. E-mail addresses: asadi41360@gmail.com, aszohreh2@live.utm.my (Z. Asadi-Shekari).

transport availability and waiting time, influence travel behavior (e.g., Habib and Sasic, 2014; Whalen et al., 2013; Vega and Reynolds-Feighan, 2009; Can, 2013; Bhatta and Larsen, 2011).

Although safety is one of the most important factors to consider when investigating mode choice (Lovegrove et al., 2010; Lovegrove and Sayed, 2006a,b, and 2007; Can, 2013; Deka, 2013), few studies have focused on this important factor in Europe (Asadi-Shekari et al., in press). Several efforts have been made to explain the different factors that affect work travel mode choice (e.g., Vega and Reynolds-Feighan, 2009; Commins and Nolan, 2011; Liu, 2007; Habib, 2012; Pooley and Turnbull, 2000; Palma and Rochat, 2000; Xie et al., 2003). However, limited studies have focused on travel safety (e.g., Amoh-Gyimah and Aidoo, 2013). The need to go to school or work is the main reason why many people travel. Nearly all previous efforts have focused on safe routes to schools (e.g., Nevelsteena et al., 2012; National Center for Safe Routes to School, 2012) and the mode choice when traveling to a school or university (e.g., Whalen et al., 2013; Mitra et al., 2010; Muller et al., 2008; Ewing et al., 2004). In contrast, only a few studies have evaluated the impacts of work travel mode on traffic safety. This study attempts to compensate for this shortcoming by analyzing the effects of work travel mode choice on traffic fatalities.

Work travel mode impacts are important because urban traffic congestion is worse during the periods of travel to and from work (Quarmby, 1967). Traffic congestion appears to significantly impact road safety. However, Quddus et al. (2010) found that the effects of traffic congestion on accident severity are limited. Earlier studies (e.g., Shefer, 1994; Shefer and Rietveld, 1997; Baruya, 1998) suggested that because speed would be lower in congested areas, there should be a negative relationship between traffic congestion and road accidents. However, recent studies (e.g., Noland and Quddus, 2005) have revealed that a negative relationship does not exist between traffic congestion and road accidents. Kononov et al. (2008) found that the total accident rate and fatality and injury rates increased in congested areas. These authors used data from Texas, California and Colorado in the US. In addition, Wang et al. (2013) found a positive relationship between traffic congestion and fatalities and serious injuries from accidents in London. This positive relationship may be due to poor driving behaviors and higher speed variances within and between the lanes.

Nearly all people who work make two trips per day. Thus, the trip to work (one of the most commonly experienced types of travel) is a special scenario (Quarmby, 1967). To our knowledge, little is known in the transportation literature regarding the impacts of key work travel modes (e.g., walking, cycling, public transport and car) on traffic fatalities. This research represents an attempt to understand this phenomenon better in a quantitative sense. In addition, limited studies are available regarding macro-level collision prediction models at the city or neighborhood level (e.g., Hadayeghi et al., 2003, 2010; Lovegrove and Sayed, 2006a; Lovegrove, 2007; Hadayeghi, 2009). Lovegrove et al. (2010) used data from 400 Greater Vancouver neighborhoods in British Columbia, Canada to evaluate a regional transportation plan regarding road safety. These authors utilized previous community-based macro-level models. In addition, Lovegrove and Sayed (2006a) proposed macro-level collision models. These authors used data from 577 neighborhoods in the Greater Vancouver Regional District and found that traffic collisions were related to several neighborhood characteristics (e.g., network shape). Moeinaddini et al. (2014a) observed a relationship between macro-level indicators, including traffic fatalities and the street network, at the city level. Although these studies proposed important factors that affect travel safety, they cannot represent travel safety and the choice of travel mode at the city level.

The city level can result in better planning decisions and strategies. Thus, as part of continuing efforts in this area, this study examines how work travel modes, including the percentages of journeys to work by public transport, motorcycle, bicycle and foot, contribute to traffic fatalities at the city level. This research uses data from different European cities to estimate the relationships between work travel modes and the numbers of transport fatalities in different European cities (city level). Finally, this research attempts to explain the use of macro-level models in the city-level planning process.

2. Data and methodology

This study estimated the strengths of relationships between work travel mode indicators and transport fatalities in various European cities at a macro-level scale (across cities). The numbers of deaths that occurred due to road accidents per one-hundred thousand people was used as the dependent variable, and the work travel mode choice indicators, including walking, cycling and public transport, were used as the independent variables to estimate this relationship (refer to Table 1). Zero traffic was tested by adding the numbers of registered private cars as an exposure variable. The Eurostat transport data from 2011 were used in this research (from the Eurostat database website). The Eurostat database has different mobility indicators for more than 1000 cities from different parts of Europe. The cities were chosen based on data availability for all of the variables after removing the outliers. Overall, four indicators for work travel mode choices were used (i.e., the percentages of journeys to work by public transport, motorcycle, bicycle and foot).

A collision is a random, rare and multi-factor event. Various methods of analysis have been used previously in different approaches for crash data. To model crash data, the Poisson (PO) distribution and the NB models have been used widely (e.g., Wei and Lovegrove, 2013; Mitra and Washington, 2007; Hadayeghi et al., 2007, 2003; Memon, 2006; Lovegrove and Sayed, 2006b; Noland and Quddus, 2004; Miaou and Lord, 2003; Abdel-Aty and Radwan, 2000; Kim et al., 2002). However, the Poisson model is not suitable for handling over- or underdispersed data. The NB model is more suitable for overdispersion but is not suitable for handling underdispersion (Lord and Mannering, 2010).

Both the NB and Poisson distributions have been used to predict accidents related to counting data (Kibria, 2006). Counting data are distributed as non-negative integers. The NB distribution is a more flexible extension of the Poisson distribution (Kibria, 2006). The Poisson distribution is suitable for modeling crash data for a single site (Hadayeghi, 2009). Crash data from a series of sites often exhibit large variances and small means that result in overdispersion when the variance-to-mean ratio is greater than 1 (Hadayeghi, 2009). For modeling crash data at a series of sites, the NB distribution, which is also known as the Poisson–gamma distribution, is the most commonly used method (Hadayeghi, 2009).

Several studies have used other analysis methods, such as linear regression (e.g., Clark and Cushing, 2004), geographically weighted regressions (e.g., Hadayeghi et al., 2003), and log-linear models using ordinary least squares (OLS) estimations (e.g., Wier et al., 2009; Washington et al., 2006). Linear regression models that have normal distributed error structures cannot handle road collisions data because the data are non-negative and discrete (Moeinaddini et al., 2014a). The Poisson and Negative Binomial (NB) regressions in the generalized linear model (GLM) framework have been more successfully adopted for these count data (Hadayeghi, 2009). Previously, GLMs were commonly used in collision models because they are useful for interpreting random, rare, sporadic and non-negative data (Lovegrove et al., 2010; Wei and Lovegrove, 2013).

The analysis methods are selected based on the measurement scale, the number of groups, the natures of the relationships

Table 1 Research data.

Cities Percentage of journeys to work by
public transport (rail, metro, bus,
tram)		 Percentage of
journeys to work by
motor cycle		 Percentage of
journeys to work
by bicycle		 Percentage of
journeys to work
by foot		 Number of
private cars
registered		 Number of deaths in road
accidents per 100,000
population
Bruxelles/
Brussel		 40.3 1.2 1.9 4.1 513,050 2.2
Antwerpen 20.7 1.48 14.81 2.01 192,115 3.41
Gent 14.27 2 14.09 1.43 103,265 4.43
Liège 8.43 0.92 1.13 5.13 154,495 5.04
Brugge 9.75 2.21 24.71 1.99 54,307 2.56
Namur 15.95 0.85 1.08 3.2 48,704 11.8
Leuven 17.98 1.13 15.26 1.85 73,667 3.08
Mons 11.42 1.19 1.57 2.63 41,605 15.1
Kortrijk 6.04 1.36 15.74 1.23 38,302 6.65
Oostende 8.08 2.89 20.1 3.46 26,857 4.29
Plovdiv 28 10 8 30 182,602 4.14
Varna 32 11 9 21 185,600 4.48
Burgas 26 14 13 26 114,967 13.98
Pleven 25 11 15 27 69,510 10.27
Ruse 21 9 12 36 88,820 8.68
Vidin 18 9 14 33 33,526 6.24
Stara 25 15 17 17 84,857 13.01
Zagora
Sliven 25 14 15 20 65,139 31.64
Dobrich 28 13 16 22 48,245 2.2
Shumen 19 13 16 30 48,697 8.65
Pernik 29 4 12 24 72,106 3.74
Yambol 24 13 18 23 40,031 4.04
Haskovo 23 15 18 25 50,844 7.85
Pazardzhik 27 14 18 18 62,003 19.44
Blagoevgrad 22 13 17 22 41,043 5.64
Veliko
Tarnovo		 28 14 9 29 46,996 17.44
Dublin 24.4 0.9 8.2 22.7 217,805 1.16
Dublin 21 1 5.3 13.5 411,836 0.87
(greater
cit)
Cork 9.2 0.7 2.9 23.3 58,357 2.53
Limerick 5.3 0.3 2.8 22.4 21,369 1.76
Galway 6.5 0.3 5 17.3 30,908 2.7
Waterford 3 0.5 2 15.5 20,057 2.15
Helsinki/ 37.8 0.5 9.5 12.5 240,175 1.36
Helsingf
Helsinki/ 30.8 0.3 7.7 9.8 468,314 1.82
Helsingf
Oulu/ 7 0 24 9 67,644 4.94
Uleåborg
Espoo/Esbo 20 2 7 21 120,957 2.42
Lahti/Lahtis 8 1 14 13 49,861 5.91

between the groups, the number of variables and the assumptions of the statistical tests (Moeinaddini et al., 2014b). In this study, the dependent variable is the number of deaths in road accidents per one hundred thousand people. This variable is similar to nonnegative count data, in which the numbers of deaths only take positive and discrete values. In contrast, the dependent variables are scaled factors because they are divided by the population to normalize the variables between cities. Gamma with log link and lognormal models can be used for scale data in the GLM framework. GLMs with a gamma distribution can handle collision data slightly better than the log-transformed model (Moeinaddini et al., 2014a). In addition, lognormal is not related to the exponential family (e.g., Gamma, Poisson, Negative Binominal, etc.). The Gamma model assumes a log link, such as the following NB model (refer to Eq. (1)):

$$Y = EXP\left(\beta0 + \sum{i=1}^{i} \beta_i \times X_i\right), \tag{1}$$

where Y is the dependent variable, i is a subscript that shows the number of independent variables, X is the independent variable, b0 is a constant, and bi is a coefficient of the independent variable. The constants and coefficients will be calculated during the calibration process.

3. Results

In this section, the results of the mode choice fatality GLM analysis are discussed. Descriptive statistics for the variables included in the model are indicated in Tables 2 and 3. No strong correlation occurred between the independent variables that were included in the model (refer to Table 4). The variables that were included in the model, their parameter estimates and the significance of these parameters (5% level) are indicated in Table 5. The modeled goodness of fit for the Omnibus test, the likelihood ratio, the Chi-Square test statistics, the scaled deviance (SD) and the Pearson Chi-Square statistic are indicated in Tables 6 and 7. Therefore, the final model can be defined as follows:

$$D = EXP (-7.789 - 0.102P - 0.031F + 0.238M - 0.044B),$$ (2)

where D is the number of road accident deaths per one-hundred thousand people, P is the percentage of journeys to work by public transport, F is the percentage of journeys to work by foot, M is the percentage of journeys to work by motorcycle, and B is the

Table 2 Continuous variable information.

N Minimum Maximum Mean Std. Deviation
Dependent Variable D 37 .87 31.64 6.6918 6.35116
Covariate P 37 3.00 40.30 19.6195 9.56392
F 37 1.23 36.00 16.4873 10.32461
M 37 .00 15.00 5.8035 5.81870
B 37 1.08 24.71 11.5078 6.50596
Offset LNCR 37 9.91 13.15 11.2568 .82816

Where D is the number of deaths in road accidents per hundred thousand population, P the percentage of journeys to work by public transport, F the percentage of journeys to work by foot, M the percentage of journeys to work by motorcycle, B the percentage of journeys to work by bicycle and CR is the number of private cars registered.

Table 3 Case processing summary.

N Percent (%)
Included
Excluded		 37
0		 100.0
0.0
Total 37 100.0

Table 4 Correlation matrix.

| Coefficientsa | | | | | | | | |---------------|---|-------------------------|-------|--|--|--|--| | Model | | Collinearity statistics | | | | | | | | | Tolerance | VIF | | | | | | 1 | P | .748 | 1.337 | | | | | | | F | .522 | 1.915 | | | | | | | M | .354 | 2.824 | | | | | | | B | .680 | 1.471 | | | | |

Where D is the number of deaths in road accidents per hundred thousand population, P the percentage of journeys to work by public transport, F the percentage of journeys to work by foot, M the percentage of journeys to work by motorcycle and B is the percentage of journeys to work by bicycle.

percentage of journeys to work by bicycle. This model shows that the numbers of deaths in road accidents were significantly affected by the work travel mode indicators in this model. Among these indicators, the percentage of journeys to work by motorcycle had a positive parameter, which indicated a greater effect on the number of road accident deaths in this model. The second most effective indicator of the percentage of journeys to work by public transport had a negative relationship. The percentage of journeys to work by bicycle and the percentage of journeys to work by foot also had negative parameters and were inversely associated with transport fatalities in this model.

Table 5 Parameter estimates.

Parameter B Std. error 95% Wald confidence interval Hypothesis test
Lower Upper Wald Chi-Square df Sig.
(Intercept) 7.789 .3763 8.526 7.051 428.454 1 .000
P .102 .0151 .132 .072 45.460 1 .000
F .031 .0148 .060 .002 4.431 1 .035
M .238 .0320 .175 .300 55.091 1 .000
B .044 .0202 .084 .004 4.725 1 .030
(Scale) .513a .1105 .336 .782

Dependent variable: D.

Model: (Intercept), P, F, M, B, offset = LNCR.

Where D is the number of deaths in road accidents per hundred thousand population, P the percentage of journeys to work by public transport, F the percentage of journeys to work by foot, M the percentage of journeys to work by motorcycle, B the percentage of journeys to work by bicycle and CR is the number of private cars registered.

Table 6 Omnibus test.a

Likelihood ratio Chi-Square df Sig.
34.598 4 .000

Dependent variable: D.

Model: (Intercept), P, F, M, B, offset = LNCR.

Where D is the number of deaths in road accidents per hundred thousand population, P the percentage of journeys to work by public transport, F the percentage of journeys to work by foot, M the percentage of journeys to work by motorcycle, B the percentage of journeys to work by bicycle and CR is the number of private cars registered.

Table 7 SD and Pearson Chi-Square goodness of fit.

Value df Value/df
Deviance 20.554 32 .642
Scaled deviance 40.087 32
Pearson Chi-Square 20.794 32 .650
Scaled Pearson Chi-Square 40.555 32

Dependent Variable: D.

Model: (Intercept), P, F, M, B, offset = LNCR.

Where D is the number of deaths in road accidents per hundred thousand population, P the percentage of journeys to work by public transport, F the percentage of journeys to work by foot, M the percentage of journeys to work by motorcycle, B the percentage of journeys to work by bicycle and CR is the number of private cars registered.

Overall, fewer journeys to work by motorcycle and more journeys to work by public transport, bicycle and foot are effective work travel mode indicators that are associated with fewer transport fatalities. This finding suggests that macro-scale urban planning that reduces the number of journeys to work by motorcycle and increases the numbers of journeys to work by public transport, bicycle and foot could contribute to fewer transport fatalities in different cities.

a Dependent Variable: D.

a Maximum likelihood estimate.

a Compares the fitted model against the intercept-only model.

4. Discussion and conclusions

Macro-scale work travel mode indicators (as the independent variables) and road accident fatalities (as the dependent variable) were analyzed in this study using GLM analysis. Overall, the percentages of journeys to work by public transport, motorcycle, bicycle and foot were the work travel mode factors that were associated with road accident fatalities.

The model shows that a greater number of journeys to work by motorcycle is correlated with more road accident deaths. Thus, distributing more motorcycles in urban areas could produce more conflicts (Nguyen et al., 2014). Motorcycles are not as safe as other travel modes, and their smaller size allows more mobility, especially in congested areas (Nguyen et al., 2014). Therefore, considering more connected routes for exclusive use by motorcycles may result in fewer conflicts and transport fatalities involving vehicle transport. However, this suggestion is not supported by the proposed model.

In addition, the model shows that the percentage of journeys to work by public transport is negatively associated with road accident fatalities. Consequently, the cities that have more journeys to work by public transport mode have fewer fatalities. Public transportation has an overall low crash risk (Lovegrove et al., 2010; Litman, 2014). The total number of traffic fatalities declines significantly as transit ridership increases (Kenworthy and Laube, 2000).

This model also indicates that more journeys to work by bicycle and foot are correlated with fewer transport fatalities. Safe walking and cycling can be more attractive options for urban travelers than crowded, dangerous and costly streets that are used by private vehicles (Asadi-Shekari et al., 2013a; Asadi-Shekari et al., 2013b and Asadi-Shekari et al., 2014; Asadi-Shekari and Zaly Shah, 2011). Increasing the number of journeys to work by bicycle and foot may decrease the use of other travel modes. More private motorized facilities, such as freeways, do not contribute to improved street conditions for pedestrians and cyclists because freeways are used exclusively by motorized vehicles. Thus, pedestrians and cyclists do not benefit from freeways. Improving the street conditions for pedestrians and cyclists and improving the distribution of destinations that are within walking distance in mixed-land-use areas may increase walking and cycling (Moeinaddini et al., 2013).

This study evaluated the relationships of work travel mode indicators and deaths in road accidents in different cities in Europe at the city level. Therefore, the results of this study can illustrate this relationship regardless of the micro-level effects. In addition, this study attempted to evaluate the relationships between work travel mode indicators and safety, which have not been sufficiently studied in the literature. Moreover, this study addresses work travel mode indicators at the city level instead of the neighborhood level.

A lower number of traffic-related fatalities is an effective indicator of a sustainable and safe city. To promote these types of cities, it is important to conduct more research at the city level. Therefore, this research attempted to indicate the relationships between work travel mode indicators and road accident fatalities at the city level. These relationships can be used in fatality reduction strategies in different cities. In addition, considering the city level can result in better planning decisions for new cities.

Furthermore, this research is limited to European cities in the Eurostat database and work travel mode indicators. Universal scale relationship models may be replicated in future studies. Other urban structure indicators can be included in future studies to identify effective factors for sustainable travel patterns, regardless of the context effects. This research used Eurostat data sources, which are free, due to financial limitations; further studies may enhance the model by using more data. Moreover, travel mode indicators can change rapidly. Future studies may evaluate these changes by updating their data sources and examining the relationships between the travel mode indicators and fatalities in various parts of the world to address changing results.

Acknowledgments

The authors wish to thank all of those who have supported this research for their useful comments during its completion. In particular, we would like to acknowledge Centre for Innovative Planning and Development (CIPD) and Prof. Dr. Amran Hamzah who helped us during the research process. The research underlying this paper was supported by Universiti Teknologi Malaysia Research Management Centre (Grant Number: Q.J130000.2721.00K96).

References