- Email: [email protected]
The relationship between urban street networks and the number of transport fatalities at the city level
Safety Science 62 (2014) 114–120
Contents lists available at ScienceDirect
Safety Science journal homepage: www.elsevier.com/locate/ssci
The relationship between urban street networks and the number of transport fatalities at the city level Mehdi Moeinaddini ⇑, Zohreh Asadi-Shekari, Muhammad Zaly Shah Department of Urban and Regional Planning, Faculty of Built Environment, Universiti Teknologi Malaysia, 81310 Skudai, Johor, Malaysia
a r t i c l e
i n f o
Article history: Received 25 December 2012 Received in revised form 20 July 2013 Accepted 24 August 2013
Keywords: Urban transportation safety Street network Crash fatality Block density Nodes per selected areas City level
a b s t r a c t There are factors that impact traffic safety and the number of accident-related fatalities, such as street users, environment, road design and vehicle characteristics, but there have been limited studies that examine the relationship between street network factors and traffic-related crashes and fatalities at the city level. Therefore, this paper focused on this relationship by introducing urban street network variables, such as blocks per area, nodes per selected areas and length of roads and motorways, as independent variables and the number of fatalities as the dependent variable. This study used Open Street Maps (OSM) and International Association of Public Transport (UITP) data from 20 cities around the world. The number of blocks per area and nodes per selected areas resulted from modifying and analyzing OSM maps in ArcGIS software. The strength of the relationship in this study was found using generalized linear modeling (GLM). The findings of this research indicated that increases in fatalities are correlated with an increasing number of blocks per area, number of nodes per selected areas and length of the motorways. Ó 2013 Elsevier Ltd. All rights reserved.
1. Introduction Annually, road traffic crashes cause over 1.2 million deaths and more than 50 million severe injuries worldwide (WHO, 2009). The number of traffic crashes has increased in recent decades, especially in developing countries, and statistics show that approximately 70 percent of traffic-related deaths occur in these countries (Augustus, 2012). For example, between 1975 and 1998, the number of road traffic deaths per capita increased over 200% in Colombia and Botswana and by 44% in Malaysia (Kopits, 2004). This issue is different in high-income countries. Over the same period, traffic fatalities per person decreased by amounts ranging from 25% to 50% in most European countries and by 60% in Canada and Hong Kong (Kopits, 2004). The Fatality Risk (Deaths/100,000 Persons) is estimated to be 7.8 in 2020 in highincome countries, and this rate may be between 14.9 and 31 in other regions (Kopits, 2004). Therefore, concerns about traffic safety are very serious in developing countries and still exist in developed countries. Ample studies have been devoted to evaluating the effects of specific safety factors, such as speed limits (Eluru et al., 2008; Abdel-Aty et al., 2007; McCarthy, 2001; Snyder, 1989), helmet laws (Lapparent, 2005; Sass and Zimmerman, 2000), seat belt laws (Evans, 1986; Wagenaar et al., 1988) and alcohol control policies (Mann et al., 2001; Ruhm, 1996; Whetten-Goldstein et al., 2000; ⇑ Corresponding author. Tel.: +60 129410543. E-mail address: [email protected] (M. Moeinaddini). 0925-7535/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ssci.2013.08.015
Dee, 1999; Wilkinson, 1987), on traffic fatalities based on withincountry data. Alcohol consumption, location of crash, speed, street conditions (lighting), time of day and day of the week, hit-and-run crashes, pedestrians and driver age are also some important factors that influence pedestrian fatalities in the US (Shankar, 2003). Permpoonwiwat and Kotrajaras (2012) concluded that policy (government budget for road safety), personal and socio-economic (health and safety, economic growth, income and education), demographic (vehicle kilometers traveled, age and gender), and geographic (population density) factors influenced the fatality rates for motor vehicle accidents in Thailand in 2006–2009. In India, increased enforcement of road traffic rules can decrease road traffic crash fatalities (Grimm and Treibich, 2012). Generally, accidents involve damage to vehicles and potential loss of human life, in addition to causing travel costs as a result of delays in traffic (Hadayeghi, 2002). The causes of motor vehicle crashes are multi-faceted and involve the interaction of a number of pre-crash variables that include humans, vehicles and road conditions (Augustus, 2012). Because there is no single variable that will reduce road crashes and fatalities, employing engineering, enforcement and education with a focus on both motorized and non-motorized users is a comprehensive approach. The purpose of this paper was to: 1. Review the literature concerning macro-level collision prediction models. 2. Estimate the relationship between urban street networks and the number of transport fatalities across different cities.
M. Moeinaddini et al. / Safety Science 62 (2014) 114–120
3. Demonstrate the use of macro-level models in a city level planning process. Only a few efforts have been undertaken to quantify and assess traffic safety impacts in the planning process at the city level (De Leur and Sayed, 2001). Collision prediction models at the community-based macro-level would improve the road safety analyses that are used by engineers and planners (Hadayeghi et al., 2003; Guevara et al., 2004). Some studies have developed macro-level collision prediction models to provide empirical tools for engineers and planners to conduct proactive road safety planning (Hadayeghi et al., 2003; Guevara et al., 2004; Hadayeghi et al., 2010a). In one of the early studies on urban form factors and traffic safety, Augustus (2012) proposed a model that used the length of roads (in km), the presence of a road safety corps (special traffic police in Nigeria) and population as independent variables and the number of road traffic accidents as the dependent variable in Lagos from 1970 to 2001. Although this model suggests considering two urban design factors, these indicators are not strong enough to illustrate urban street patterns on a macro-level scale at the city level. Rifaat et al. (2011) proposed the effect of street patterns on the severity of crashes involving vulnerable road users. In this study, street patterns were classified into four categories: grid-iron, warped parallel, loops and lollipops, and mixed patterns (refer to Fig. 1). This classification divided the city into different precincts. The results of this research showed that, compared to other street patterns, the loops and lollipops design increased the probability of an injury but reduced the probability of fatalities and property damage-only incidents in the event of a crash. However, it should be considered that the areas of these precincts may affect the results regardless of street patterns.
115
Marshall and Garrick (2011) focused on how some aspects of overall community design might affect road safety at city level. The aim in this research was to identify how the characteristics of the street network, such as street network density, street connectivity and street network patterns, can have an impact on road safety outcomes. This study concluded that road safety outcomes are affected by street network characteristics. They proposed a spatial analysis of crash data in 24 medium-sized California cities. Noland (2003) also found that increases in the total lane miles, the average number of lanes on collector roads, and the percent of collectors with lanes 12 feet or wider was correlated with an increase in total fatalities in the US. Some researchers have developed empirical road safety planning tools and macro-level collision prediction models at neighborhood level (e.g., Hadayeghi et al., 2003; Lovegrove, 2007; Lovegrove and Sayed, 2006a). Hadayeghi et al. (2010b) and Hadayeghi (2009) considered different types of data, including street networks, land use, traffic demands, socioeconomic and demographic characteristics, dwelling units, and employment, in the city of Toronto to reduce the number of collisions and, consequently, improve safety. Hadayeghi (2002) proposed that the number of households, major road kilometers, vehicle kilometers traveled, intersection density, posted speed and volume/speed ratio can impact accident occurrence. He proposed models for predicting the number of accidents per zone in the city of Toronto. Wei and Lovegrove (in press) proposed an empirical tool based on four groups of data as independent variables including exposure, socio-demographics, transportation demand management and road network in evaluating the safety of cyclists. These variables are not usually considered at the micro level (Khondakar et al., 2010). Detailed descriptions of these four groups are given
Fig. 1. Examples of street patterns (Rifaat et al., 2011).
116
M. Moeinaddini et al. / Safety Science 62 (2014) 114–120
in different studies, such as Lovegrove (2007) and Lovegrove and Sayed (2006a). In addition, Guevara et al. (2004) found that the significant factors of injury and property-damage crash models in Tucson, Arizona, are the number of employees, population density, intersection density, percentage of miles of principal arterial, percentage of miles of minor arterials, and percentage of miles of urban collectors. Hadayeghi et al. (2006) examined the temporal transferability of the zonal accident prediction models for the city of Toronto in 1996 and 2001. Hadayeghi et al. (2010a) considered land use, traffic, demographic, dwelling units, employment types and street networks, including number of rail stations, total rail kilometer, number of schools, total arterial road, expressway, collector, laneway, local road and ramp kilometers, number of 4-legged and 3-legged signalized intersections, as independent variables in developing planning level transportation safety tools using a geographically weighted Poisson regression. Although these studies evaluated different effective factors, they cannot show street network and transport fatality relationships across different cities. There have also been other efforts to classify street patterns. Lovegrove (2007) proposed that, in residential areas, cul-de-sac road networks are popular for limiting through-traffic but accessibility for emergency vehicles and transit is significantly less than a grid network. Lovegrove and Sayed (2006b) suggested that 3-way offset road networks are safer than grid networks and cul-de-sac networks. Sun (2009) proposed that the 3-way offset and Fused Grid road networks were the safest road patterns. The 3-way offset pattern also features significantly better road safety and accessibility than other street patterns (Ogden, 1996). Wei and Lovegrove (2012) reviewed some studies of neighborhood road patterns that can save the lives of vulnerable road users. However, these efforts cannot illustrate urban street patterns across different cities. Many researchers have proposed traditional micro-level collision prediction models, but less attention has been given to the transferability of macro-level collision prediction models (Khondakar et al., 2010). Although the effects of different factors have been explored in previous research, limited research has been conducted on exploring the impacts of street network factors on transport fatalities in different contexts (at the city level) with a macro-level scale. As a part of continuing efforts in this area, this study examines how these factors, including length of roads, blocks per area, nodes per selected areas and length of motorways, contribute to crash fatalities. This study uses data from 20 cities around the world to estimate this relationship in different contexts.
2. Materials and methods This paper focuses on the relationship between street network factors and transport fatalities in different contexts with a macro-level scale by considering street network variables, such as blocks per area, nodes per selected areas and length of roads and motorways, as independent variables and the number of passenger transport fatalities per million inhabitants as the dependent variable (refer Table 1). The data that are used in this study are collected from the mobility in cities database (UITP, 2006) and the Open Street Map (OSM) web site. The UITP database has 120 urban mobility indicators in 52 cities from various parts of the world, but the number of urban street network indicators is limited to 20 cities in this database. The cities are chosen based on availability of data for all indicators after removing the outliers (refer Table 1). In addition, there are limited indicators for street networks in the UITP data source. There are only four indicators for street networks at the city level that indicate the length of roads and length of motorways in different units (per hectare and per inhabitant). Other UITP street indicators, such as road investments and speed,
cannot be illustrated in the street network at the city level. These indicators also cannot represent the shape of real urban street networks. Because the real shapes of street networks are illustrated in street maps, this research also uses the OSM data source to include indicators of urban street network shape. Because block size and intersection density are important factors in street networks, connectivity and shape indicators are illustrated in this study by blocks per area and nodes per selected areas (refer Fig. 2). The number of blocks per area is the number of polygons that are bounded by street lines in a selected area. UITP database defined Metropolitan and central business district for each city in the database. The selected area is chosen based on UITP Metropolitan definition. The number of nodes per selected areas is the number of nodes in a selected area per the number of polygons that are bounded by street lines in that area. The number of blocks per area can show the size of the street network blocks, and the number of nodes per selected areas can introduce the shape of them. Blocks per area and nodes per selected areas can be achieved for urban streets by modifying and analyzing OSM data. OSM data are digital maps of street networks with various layers. Because only the street layer is needed in this research, these maps should be edited in JOSM software (Java open street map). JOSM is a free program for modifying OSM maps that is provided by the OSM web site. This software is user friendly, and the streets layer can be selected easily. This selected layer can be saved as a new .gpx file, and then this file can be analyzed in ArcGIS software. There are three different layers in the .gpx file when it is opened in ArcGIS. These layers are displayed as streets that are lines, intersections and junctions that are nodes and the selected area that is a polygon. Nodes also include the points at which a street changes direction (mid-block nodes in Fig. 2 are used to show these nodes). The streets layer is converted to polygons using Arc Toolbox to create polygons that are bounded by street lines. The intersect tool in Arc Toolbox can provide street polygons in the selected area and the number of nodes in that area. Therefore, the block (polygon) density and nodes per selected areas are the result of open street map analysis in ArcGIS software after editing the maps in JOSM software. Fig. 2 illustrates the result of converting the OSM data to block density and nodes per selected areas. The number of blocks per area can illustrate the size of street blocks, and nodes per selected areas show the average road segment density for blocks. Although the type of connection (e.g., 4-way, 3-way, interchange, etc.) is not considered in this macrolevel study, more nodes can indicate more routes. Therefore, these indicators also represent connection density. The length of motorway per ten thousand inhabitants, the length of road per million inhabitants and the number of passenger transport fatalities per million inhabitants are collected from UITP. There are also data on the length of road and length of motorway per urban hectare in the UITP database. Because the dependent variable is transport fatalities per inhabitant, the length of road and length of motorway per inhabitant are also selected in order to illustrate the relationship between transport fatalities and street length in the context of population. This issue shows the impacts of population growth, length of road and length of motorway on road safety. In addition, blocks per area and nodes per selected areas can indicate land development. The roads in this research are different from motorways. The roads can be used by motorized vehicles, pedestrians and cyclists but the motorways can be used by motorized vehicles while walking and cycling users do not experience benefits from them. Table 1 presents the data that are used in this study. Different methods have been successfully used by previous researchers in different approaches to crash studies. Clark and Cushing (2004) used linear regression. Linear regression with a
117
M. Moeinaddini et al. / Safety Science 62 (2014) 114–120 Table 1 Research data. Cities
Blocks per area
Nodes per selected areas
Length of road per million inhabitants (m)
Length of motor way per ten thousand inhabitants (m)
Passenger transport fatalities per million inhabitants
Amsterdam Athens Barcelona Berlin Bern Bilbao Bologna Brussels Budapest Chicago Clermont Ferrand Copenhagen Dubai Dublin Geneva Ghent Glasgow Graz Hamburg Helsinki Hong Kong Krakow Lille Lisbon London Lyons Madrid Manchester Marseilles Melbourne Milan Moscow Munich Nantes Newcastle Oslo Paris Prague Rome Rotterdam Sao Paulo Seville Singapore Stockholm Stuttgart Tallinn Tunis Turin Valencia Vienna Warsaw Zurich
53.6 – 54.97 – – – 18.2 – – – – 21.96 – – 26.96 – 35.26 20.03 – – 18.08 – – 37.61 53.16 34.96 59.48 – 23.14 – – – 52.5 32.16 – – 46.12 – 58.03 62.11 – 52.55 – 38.08 29.06 – – 35.09 – 46.37 20.38 –
6.71 – 4.89 – – – 8.55 – – – – 6.48 – – 8.26 – 6.91 9.8 – – 9.51 – – 7.01 8.48 5.11 5.43 – 8.52 – – – 6.3 10.44 – – 6.12 – 6.89 5.88 – 5.8 – 7.83 11.77 – – 4.24 – 5.77 8.77 –
2.75 2.31 2.1 1.57 3.92 4.36 2.49 1.94 2.43 4.77 3.4 3.85 3.1 4.26 4.9 5.48 5.8 4.4 – 3.61 0.284 1.45 3.48 0.889 2.03 2.47 4.87 3.7 1.63 – – 0.406 1.83 5.41 4.12 5.86 1.98 2.91 2.8 4.07 1.96 2.02 0.94 – 1.19 2.17 – 2.71 2.87 1.81 1.68 4.7
– 3.9 8.97 2.02 15.7 12 15.2 3.63 1.36 9.53 4.17 12.5 39.5 – 10.3 14.6 11.1 7.96 – 9.08 1.65 2.24 14.4 8.68 0.99 6.81 9.83 7.09 4.13 – – 1.14 4.78 12.3 12.9 17 6.95 6.44 11.1 – 2.02 11.6 4.52 26.9 4.92 2.51 – 17.2 2.87 2.64 4.5 10.9
33 77 84.3 19.2 34.8 54.3 99.1 45.6 63.1 79.7 114 46.9 203 47.2 40.5 – 52.7 39.8 34.5 20.6 30 – 24.5 73.1 41.8 22.1 71 42.2 62.5 77.4 – 47.1 22.3 64.9 23.2 38.7 65.9 57.5 108 33.9 109 103 58.4 36.4 57.2 65.2 – 97.9 33.8 26.4 – 44.6
spatial lag parameter was used by Levine et al. (1995). Chatterjee et al. (2004) utilized Poisson models. Negative binomial (NB) models were also used by numerous studies (e.g., Lovegrove and Sayed, 2006b; Wei and Lovegrove, in press; Abdel-Aty and Radwan, 2000; Hadayeghi et al., 2003 and Hadayeghi et al., 2007; Maher and Summersgill, 1996; Miaou and Lord, 2003; Mitra and Washington, 2007; Noland and Quddus, 2004; Memon, 2006). Zero-inflated Poisson and NB models (ZIP/ZINB) were used by Qin et al. (2004) and Shankar et al. (1997). Guevara et al. (2004) utilized simultaneous estimation of NB models. Full Bayes hierarchical models with spatial effects were used by Aguero-Valverde and Jovanis (2006) and Quddus (2008). Hadayeghi et al. (2003) proposed a geographically weighted regression (GWR). Log-linear models using ordinary least squares (OLS) estimation were used by Washington et al. (2006) and Wier et al. (2009). Road collisions are random and rare events. Moreover, count data are non-negative and discrete. Therefore, conventional linear
regression models with a normally distributed error structure are not suitable for modeling road collisions. Poisson and NB regressions in the generalized linear model (GLM) framework have been more successfully adopted for this issue (Hadayeghi, 2009). Kim et al. (2002) found that Poisson Gamma and Poisson lognormal models can be more successful for the prediction of collisions. The Poisson distribution has been shown to be suitable for modeling crash data for a single site, but crash data for different sites often exhibit large variances and small means that lead to overdispersion when there is a variance-to-mean ratio greater than 1 (Hadayeghi, 2009). Therefore, the NB distribution, also known as the Poisson–gamma distribution, has become the most commonly used method for modeling crash data at a series of sites (Hadayeghi, 2009). In previous collision prediction models studies, GLMs were commonly used. They have been successful in interpreting random, rare, sporadic, and non-negative collision data (Lovegrove et al., 2010; Wei and Lovegrove, in press).
118
M. Moeinaddini et al. / Safety Science 62 (2014) 114–120 Table 5 Parameter estimates. Parameter
(Intercept) Blocks per area Nodes per selected areas Length of road per million inhabitants Length of motor way per ten thousand inhabitants (Scale)
Fig. 2. An illustration of blocks and nodes.
Table 2 Continuous variable information. Indicators
N
Minimum
Maximum
Mean
Std. deviation
Passenger transport fatalities per million inhabitants Blocks per area Nodes per selected areas Length of road per million inhabitants Length of motor way per ten thousand inhabitants
20
22
108
60.47
27.435
20 20 20
18.1 4.24 .28
59.5 11.77 5.80
37.785 7.3140 2.7732
14.1208 2.02433 1.56602
20
.99
17.20
8.4805
4.40316
B
Std. error
95% Wald confidence interval
Hypothesis test
Lower
Upper
Wald ChiSquare
df
Sig.
1.922 .018 .114
.5343 .0054 .0429
.874 .008 .030
2.969 .029 .198
12.933 11.329 7.058
1 1 1
.000 .001 .008
.140
.0453
.229
.051
9.547
1
.002
.114
.0181
.079
.150
39.563
1
.000
.068a
.0214
.037
.126
Dependent variable: passenger transport fatalities per million inhabitants. Model: (Intercept), blocks per area, nodes per selected areas, length of road per million inhabitants, length of motor way per ten thousand inhabitants. a Maximum likelihood estimate.
Table 6 Omnibus testa. Likelihood ratio Chi-Square
df
Sig.
23.742
4
.000
Dependent variable: passenger transport fatalities per million inhabitants. Model: (Intercept), blocks per area, nodes per selected areas, length of road per million inhabitants, length of motor way per ten thousand inhabitants. a Compares the fitted model against the intercept-only model. Table 3 Case processing summary.
Included Excluded Total
N
Percent (%)
20 32 52
38.5 61.5 100.0
Table 4 Correlation matrix.
Table 7 SD and Pearson Chi-Square goodness of fit.
Deviance Scaled deviance Pearson Chi-Square Scaled Pearson Chi-Square
Value
df
Value/df
1.382 20.228 1.356 19.851
15 15 15 15
.092 .090
Pearson correlation
R
M
P
N
Dependent variable: passenger transport fatalities per million inhabitants. Model: (Intercept), blocks per area, nodes per selected areas, length of road per million inhabitants, length of motor way per ten thousand inhabitants.
R M P N
1 .510 .052 .012
.510 1 .101 .276
.052 .101 1 .554
.012 .276 .554 1
the NB model, the Gamma model assumes a log link (refer to Eq. (1)):
R = length of road per million inhabitants. M = length of motorway per ten thousand inhabitants. P = blocks per area. N = nodes per selected areas.
In this study, the dependent variable is the number of passenger transport fatalities per million inhabitants. It is considered to be similar to count data, where the number of deaths only takes positive and discrete values. On the other hand, this variable is a scaled factor because it is divided by inhabitants. Lognormal and Gamma with log link models are used to scale data in the GLM framework. Lognormal is not in the exponential family (e.g., Poisson, Gamma, Binominal, Inverse, Negative Binominal, etc.). Therefore, GLMs with a gamma distribution are recommended because they performed slightly better than the log transformed model. Similar to
i X Y ¼ EXP b0 þ bi X i
! ð1Þ
i¼1
where Y is the dependent variable, i the subscript showing the number of independent variables, X the independent variable, b0 the constant, calculated in the calibration process and bi is the coefficient of the independent variable, calculated in the calibration process of the model. 3. Results In this section, the results of the street network GLM analysis are discussed. Tables 2 and 3 indicate descriptive statistics for the variables included in the model. Table 4 indicates that there is no strong correlation between the independent variables
M. Moeinaddini et al. / Safety Science 62 (2014) 114–120
included in the model. Table 5 indicates the variables included in the model, their parameter estimates, and the significance of the parameters (5% level). The Omnibus test, likelihood ratio ChiSquare test statistics, the scaled deviance (SD) and the Pearson Chi-Square statistic show the model goodness of fit (refer to Tables 6 and 7). Therefore, the final model can be defined as follows:
PF ¼ EXPð1:992 0:140R þ 0:018P þ 0:114N þ 0:114MÞ
ð2Þ
where PF is the passenger transport fatalities per million inhabitants, R the length of road per million inhabitants, P the blocks per area, N the nodes per selected areas and M is the length of motorway per ten thousand inhabitants This model shows that passenger transport fatalities in this model are significantly affected by street network indicators. Among these indicators, the number of nodes per selected areas and the length of motorway per inhabitant have higher positive parameters, so these indicators have greater effects on passenger transport fatalities in this model. The third effective indicator with a positive relationship is blocks per area. The length of road per inhabitant has negative parameter and thus it is inversely associated with transport fatalities in this model. Overall, fewer nodes per selected areas and blocks per area, shorter motorways per inhabitant and more roads per inhabitants are effective urban street network indicators that are associated with fewer transport fatalities. This suggests that macro-scale street network planning that reduces block density, nodes per selected areas and motorway density and increases road density could contribute to fewer transport fatalities across different cities. 4. Discussion and conclusions Macro-scale urban street network indicators as independent variables and passenger transport fatalities as the dependent variable are analyzed in this research using GLM analysis. From this study, it can be concluded that the length of road per million inhabitants, blocks per area, nodes per selected areas and length of motorway per ten thousand inhabitants are the urban street network factors that are associated with passenger transport fatalities. The model shows that more blocks per area are correlated with more passenger transport fatalities. Distributing more blocks in urban areas can produce more nodes. These nodes produce more conflicts. On the other hand, more and smaller blocks create more connectivity, and this arrangement produces more direct routes for alternative travel modes. However, the model shows that more blocks are also correlated with more passenger transport fatalities involving vehicle transport. This suggests that considering more connected routes to be used exclusively by non-motorized modes is more effective at allowing fewer conflicts and transport fatalities involving vehicle transport than creating more blocks and nodes to increase connectivity for all modes. In addition, the model shows that more nodes per selected areas are associated with more passenger transport fatalities. Consequently, more simple, large blocks with fewer nodes are correlated with fewer fatalities. The model also indicates that longer motorway per inhabitant is correlated with more passenger transport fatalities, while road length has a negative relationship with fatalities. More motorways can increase convenience for private motorized vehicle users. This type of road belongs to motorized vehicles users with high speed, while walking and cycling users do not experience benefits from them. Therefore, streets patterns with more facilities, such as motorways for high speed car users, do not contribute to fewer fatalities. This study evaluates the relationship of urban street indicators and passenger transport fatalities in different cities within various contexts at the city level. Therefore, the results of this study can
119
illustrate this relationship regardless of context effects. In addition, this study also attempts to use some factors, such as nodes and blocks, to describe the shapes of the street networks at a macro scale. Moreover, it attempts to address street network indicators at the city level (instead of the neighborhood level) that have not been sufficiently studied in the literature. Currently, more sustainable urban areas are needed. To promote these kinds of areas, having fewer traffic-related fatalities is an important goal. Therefore, this research tries to indicate the relationship between urban street indicators and transport fatalities. This relationship can be used in fatality reduction strategies in different cities. In addition, because the urban street network is the result of macro-scale planning decisions, considering the relationship of urban street indicators and fatalities can lead to better planning decisions for new cities. This research is limited to selected UITP and open street map web site data in selected cities. However, the universal scale relationship models may be replicated by future studies for other urban structure indicators in order to identify effective factors for sustainable travel patterns regardless of context effects. In addition, this research uses OSM data sources that are free editable maps because we had financial limitations. Further studies may enhance the model by using the original versions of urban maps. Moreover, urban structure indicators change rapidly; future studies may evaluate these changes by updating their data sources and examining the relationship between urban structure indicators and fatalities in various parts of the world to address the changing results.
References Abdel-Aty, M., Chundi, S., Lee, C., 2007. Geo-spatial and log-linear analysis of pedestrian and bicyclist crashes involving school-aged children. J. Saf. Res. 38, 571–579. Abdel-Aty, M.A., Radwan, A.E., 2000. Modeling traffic accident occurrence and involvement. Accident Anal. Prevent. 32, 633–642. Aguero-Valverde, J., Jovanis, P.P., 2006. Spatial analysis of fatal and injury crashes in Pennsylvania. Accident Anal. Prevent. 38, 618–625. Augustus, A., 2012. Determinants of road traffic accident occurrences in Lagos state: some lessons for Nigeria. Int. J. Humanities Soc. Sci. 2, 252–259. Chatterjee, A., Everett, J.D., Reiff, B., Schwetz, T.C., Seaver, W.L., Wegmann, F.J., 2004. Tools for assessing safety impact of long-range transportation plans in urban areas, In: 80th Annual Meeting of the Transportation Research Board, TRB, Washington, DC. Clark, D.E., Cushing, B.M., 2004. Rural and urban traffic fatalities, vehicle-miles and population density. Accident Anal. Prevent. 36, 967–972. Dee, T.S., 1999. State alcohol policies, teen drinking, and traffic fatalities. J. Publ. Econ. 72, 289–315. De Leur, P., Sayed, T., 2001. Using claims prediction model for road safety evaluation. Can. J. Civ. Eng. 28, 804–812. Eluru, N., Bhat, C., Hensher, D., 2008. A mixed generalized ordered response model for examining pedestrian and bicyclist injury severity level in traffic crashes. Accident Anal. Prevent. 40, 1033–1054. Evans, L., 1986. The effectiveness of safety belts in preventing fatalities. Accident Anal. Prevent. 18, 229–241. Grimm, M., Treibich, C., 2012. Determinants of Road Traffic Crash Fatalities across Indian States. Health Economics. Guevara, F.L.D., Washington, S.P., Oh, J., 2004. Forecasting crashes at the planning level: simultaneous negative binomial crash model applied in Tucson, Arizona. Transp. Res. Rec. 1897, 191–199. Hadayeghi, A., 2002. Accident Prediction Models for Safety Evaluation of Urban Transportation Network. Department of civil engineering. University of Toronto, Ontario, Canada. Hadayeghi, A., 2009. Use of Advanced Techniques to Estimate Zonal Level Safety Planning Models and Examine Their Temporal Transferability. Department of Civil Engineering. University of Toronto, Ontario, Canada. Hadayeghi, A., Shalaby, A., Persaud, B., 2003. Macro-level accident prediction models for evaluating the safety of urban transportation systems. Transp. Res. Rec. 1840, 87–95. Hadayeghi, A., Shalaby, A., Pesuad, B., 2007. Safety prediction models: a proactive tool for safety evaluation in urban transportation planning applications. Transp. Res. Rec. 2019, 225–236. Hadayeghi, A., Shalaby, A., Persaud, B., 2010a. Development of planning level transportation safety tools using geographically weighted poisson regression. Accident Anal. Prevent. 42, 676–688.
120
M. Moeinaddini et al. / Safety Science 62 (2014) 114–120
Hadayeghi, A., Shalaby, A., Persaud, B., 2010b. Development of planning-level transportation safety models using full bayesian semiparametric additive techniques. J. Transp. Saf. Security 2, 45–68. Hadayeghi, A., Shalaby, A., Persaud, B., Cheung, C., 2006. Temporal transferability and updating of zonal-level accident prediction models. Accident Anal. Prevent. 38, 579–589. Khondakar, B., Sayed, T., Lovegrove, G., 2010. Transferability of community-based collision prediction models for use in road safety planning applications. J. Transp. Eng. 136, 871–880. Kim, H., Sun, D., Tsutakawa, R.K., 2002. Lognormal vs. gamma: extra variations. Biometrical J. 44, 305–323. Kopits, E.-A., 2004. Traffic Fatalities and Economic Growth. Department of Economics. University of Maryland, College Park. Lapparent, M., 2005. Individual cyclists’ probability distributions of severe/fatal crashes in large French urban areas. Accident Anal. Prevent. 37, 1086–1092. Levine, N., Kim, K.E., Nitz, L.H., 1995. Spatial analysis of Honolulu motor vehicle crashes. II. Zonal generators. Accident Anal. Prevent. 27, 675–685. Lovegrove, G., 2007. Road Safety Planning, New Tools for Sustainable Road Safety and Community Development. Verlag Dr, Müller, Germany. Lovegrove, G., Lim, C., Sayed, T., 2010. Community-based, macro-level collision prediction model use with a regional transportation plan. J. Transp. Eng. 136, 120–128. Lovegrove, G., Sayed, T., 2006a. Macro-level collision prediction models for evaluating neighbourhood traffic safety. Can. J. Civ. Eng. 33, 609–621. Lovegrove, G., Sayed, T., 2006b. Using macro-level collision prediction models in road safety planning applications. Transp. Res. Rec.: J. Transp. Res. Board 1950, 73–82. Maher, M.J., Summersgill, I., 1996. A comprehensive methodology for the fitting of predictive accident models. Accident Anal. Prevent. 28, 281–296. Mann, R.E., MacDonald, S., Stoduto, G., Bondy, S., Jonah, B., Shaikh, A., 2001. The effects of introducing or lowering legal per se blood alcohol limits for driving: an international review. Accident Anal. Prevent. 33, 569–583. Marshall, W.E., Garrick, N.W., 2011. Does street network design affect traffic safety? Accident Anal. Prevent. 43, 769–781. McCarthy, P., 2001. Effect of speed limits on speed distributions and highway safety: a survey of recent literature. Transp. Rev. 21, 31–50. Memon, A.Q., 2006. Road Accident Prediction Models Developed from a National Database: Poisson and Negative Binomial Regressions, In: Transportation Research Board 85th Annual Meeting CD-ROM Washington DC, United States. Miaou, S.P., Lord, D., 2003. Modeling traffic crash-flow relationships for intersections: dispersion parameter, functional form, and Bayes versus empirical Bayes. Transp. Res. Rec. 1840, 31–40. Mitra, S., Washington, S., 2007. On the nature of over-dispersion in motor vehicle crash prediction models. Accident Anal. Prevent. 39, 459–468. Noland, R.B., 2003. Traffic fatalities and injuries: the effect of changes in infrastructure and other trends. Accident Anal. Prevent. 35, 599–611. Noland, R.B., Quddus, M.A., 2004. Improvements in medical care and technology and reductions in traffic-related fatalities in Great Britain. Accident Anal. Prevent. 36, 103–113. Ogden, K.W., 1996. Safer Roads: A Guide to Road Safety Engineering. Avebury Technical.
Permpoonwiwat, C.K., Kotrajaras, P., 2012. Pooled time-series analysis on traffic fatalities in Thailand. World Rev. Business Res. 2, 170–182. Qin, X., Ivan, J.N., Ravishanker, N., 2004. Selecting exposure measures in crash rate prediction for two-lane highway segments. Accident Anal. Prevent. 36, 183– 191. Quddus, M.A., 2008. Modelling area-wide count outcomes with spatial correlation and heterogeneity: an analysis of London crash data. Accident Anal. Prevent. 40, 1486–1497. Rifaat, S.M., Tay, R., De.Barros, A., 2011. Effect of street pattern on the severity of crashes involving vulnerable road users. Accident Anal. Prevent. 43, 276–283. Ruhm, C.J., 1996. Alcohol policies and highway vehicle fatalities. J. Health Econ. 15, 435–454. Sass, T.R., Zimmerman, P.R., 2000. Motorcycle helmet laws and motorcyclist fatalities. J. Regul. Econ. 18, 195–215. Shankar, U., 2003. Pedestrian Roadway Fatalities. National Center for Statistics and Analysis Advanced Research and Analysis. U.S. Department of Transportation National Highway Traffic Safety Administration. Shankar, V., Milton, J., Mannering, F., 1997. Modeling accident frequencies as zeroaltered probability processes: an empirical inquiry. Accident Anal. Prevent. 29, 829–837. Snyder, D., 1989. Speeding, coordination, and the 55 MPH speed limit: comment. Am. Econ. Rev. 79, 922–925. Sun, J., 2009. Sustainable Road Safety: Development, Transference and Application of Community-Based Macro-Level Collision Prediction Models, Civil Engineering. University of British Columbia, Okanagan. UITP, 2006. Mobility in cities Database. International Association of Public Transport (UITP). Wagenaar, A., Maybee, R., Sullivan, K., 1988. Mandatory seat belt laws in eight states: a time-series evaluation. J. Saf. Res. 19, 51–70. Washington, S., Schalwyk, I.V., Meyer, M., Dumbaugh, E., Zoll, M., 2006. Incorporating Safety into Long-Range Transportation Planning. NCHRP Report 546, TRB, Washington, DC. Wei, F., Lovegrove, G., (in press). An empirical tool to evaluate the safety of cyclists: community based, macro-level collision prediction models using negative binomial regression. Accident Analysis and Prevention. Wei, V.F., Lovegrove, G., 2012. Sustainable road safety: a new (?) neighbourhood road pattern that saves VRU lives. Accident Anal. Prevent. 44, 140–148. Whetten-Goldstein, K., Sloan, F., Stout, E., Liang, L., 2000. Civil liability, criminal law, and other policies and alcohol-related motor vehicle fatalities in the United States: 1984–1995. Accident Anal. Prevent. 32, 723–733. Wier, M., Weintraub, J., Humphreys, E.H., Seto, E., Bhatia, R., 2009. An area-level model of vehicle-pedestrian injury collisions with implications for land use and transportation planning. Accident Anal. Prevent. 41, 137–145. Wilkinson, J.T., 1987. Reducing drunken driving: which policies are most effective? Southern Econ. J. 54, 322–334. World Health Organization (WHO), 2009. Global Status Report on Road Safety. Time for Action. World Health Organization, Geneva.
Contents lists available at ScienceDirect
Safety Science journal homepage: www.elsevier.com/locate/ssci
The relationship between urban street networks and the number of transport fatalities at the city level Mehdi Moeinaddini ⇑, Zohreh Asadi-Shekari, Muhammad Zaly Shah Department of Urban and Regional Planning, Faculty of Built Environment, Universiti Teknologi Malaysia, 81310 Skudai, Johor, Malaysia
a r t i c l e
i n f o
Article history: Received 25 December 2012 Received in revised form 20 July 2013 Accepted 24 August 2013
Keywords: Urban transportation safety Street network Crash fatality Block density Nodes per selected areas City level
a b s t r a c t There are factors that impact traffic safety and the number of accident-related fatalities, such as street users, environment, road design and vehicle characteristics, but there have been limited studies that examine the relationship between street network factors and traffic-related crashes and fatalities at the city level. Therefore, this paper focused on this relationship by introducing urban street network variables, such as blocks per area, nodes per selected areas and length of roads and motorways, as independent variables and the number of fatalities as the dependent variable. This study used Open Street Maps (OSM) and International Association of Public Transport (UITP) data from 20 cities around the world. The number of blocks per area and nodes per selected areas resulted from modifying and analyzing OSM maps in ArcGIS software. The strength of the relationship in this study was found using generalized linear modeling (GLM). The findings of this research indicated that increases in fatalities are correlated with an increasing number of blocks per area, number of nodes per selected areas and length of the motorways. Ó 2013 Elsevier Ltd. All rights reserved.
1. Introduction Annually, road traffic crashes cause over 1.2 million deaths and more than 50 million severe injuries worldwide (WHO, 2009). The number of traffic crashes has increased in recent decades, especially in developing countries, and statistics show that approximately 70 percent of traffic-related deaths occur in these countries (Augustus, 2012). For example, between 1975 and 1998, the number of road traffic deaths per capita increased over 200% in Colombia and Botswana and by 44% in Malaysia (Kopits, 2004). This issue is different in high-income countries. Over the same period, traffic fatalities per person decreased by amounts ranging from 25% to 50% in most European countries and by 60% in Canada and Hong Kong (Kopits, 2004). The Fatality Risk (Deaths/100,000 Persons) is estimated to be 7.8 in 2020 in highincome countries, and this rate may be between 14.9 and 31 in other regions (Kopits, 2004). Therefore, concerns about traffic safety are very serious in developing countries and still exist in developed countries. Ample studies have been devoted to evaluating the effects of specific safety factors, such as speed limits (Eluru et al., 2008; Abdel-Aty et al., 2007; McCarthy, 2001; Snyder, 1989), helmet laws (Lapparent, 2005; Sass and Zimmerman, 2000), seat belt laws (Evans, 1986; Wagenaar et al., 1988) and alcohol control policies (Mann et al., 2001; Ruhm, 1996; Whetten-Goldstein et al., 2000; ⇑ Corresponding author. Tel.: +60 129410543. E-mail address: [email protected] (M. Moeinaddini). 0925-7535/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ssci.2013.08.015
Dee, 1999; Wilkinson, 1987), on traffic fatalities based on withincountry data. Alcohol consumption, location of crash, speed, street conditions (lighting), time of day and day of the week, hit-and-run crashes, pedestrians and driver age are also some important factors that influence pedestrian fatalities in the US (Shankar, 2003). Permpoonwiwat and Kotrajaras (2012) concluded that policy (government budget for road safety), personal and socio-economic (health and safety, economic growth, income and education), demographic (vehicle kilometers traveled, age and gender), and geographic (population density) factors influenced the fatality rates for motor vehicle accidents in Thailand in 2006–2009. In India, increased enforcement of road traffic rules can decrease road traffic crash fatalities (Grimm and Treibich, 2012). Generally, accidents involve damage to vehicles and potential loss of human life, in addition to causing travel costs as a result of delays in traffic (Hadayeghi, 2002). The causes of motor vehicle crashes are multi-faceted and involve the interaction of a number of pre-crash variables that include humans, vehicles and road conditions (Augustus, 2012). Because there is no single variable that will reduce road crashes and fatalities, employing engineering, enforcement and education with a focus on both motorized and non-motorized users is a comprehensive approach. The purpose of this paper was to: 1. Review the literature concerning macro-level collision prediction models. 2. Estimate the relationship between urban street networks and the number of transport fatalities across different cities.
M. Moeinaddini et al. / Safety Science 62 (2014) 114–120
3. Demonstrate the use of macro-level models in a city level planning process. Only a few efforts have been undertaken to quantify and assess traffic safety impacts in the planning process at the city level (De Leur and Sayed, 2001). Collision prediction models at the community-based macro-level would improve the road safety analyses that are used by engineers and planners (Hadayeghi et al., 2003; Guevara et al., 2004). Some studies have developed macro-level collision prediction models to provide empirical tools for engineers and planners to conduct proactive road safety planning (Hadayeghi et al., 2003; Guevara et al., 2004; Hadayeghi et al., 2010a). In one of the early studies on urban form factors and traffic safety, Augustus (2012) proposed a model that used the length of roads (in km), the presence of a road safety corps (special traffic police in Nigeria) and population as independent variables and the number of road traffic accidents as the dependent variable in Lagos from 1970 to 2001. Although this model suggests considering two urban design factors, these indicators are not strong enough to illustrate urban street patterns on a macro-level scale at the city level. Rifaat et al. (2011) proposed the effect of street patterns on the severity of crashes involving vulnerable road users. In this study, street patterns were classified into four categories: grid-iron, warped parallel, loops and lollipops, and mixed patterns (refer to Fig. 1). This classification divided the city into different precincts. The results of this research showed that, compared to other street patterns, the loops and lollipops design increased the probability of an injury but reduced the probability of fatalities and property damage-only incidents in the event of a crash. However, it should be considered that the areas of these precincts may affect the results regardless of street patterns.
115
Marshall and Garrick (2011) focused on how some aspects of overall community design might affect road safety at city level. The aim in this research was to identify how the characteristics of the street network, such as street network density, street connectivity and street network patterns, can have an impact on road safety outcomes. This study concluded that road safety outcomes are affected by street network characteristics. They proposed a spatial analysis of crash data in 24 medium-sized California cities. Noland (2003) also found that increases in the total lane miles, the average number of lanes on collector roads, and the percent of collectors with lanes 12 feet or wider was correlated with an increase in total fatalities in the US. Some researchers have developed empirical road safety planning tools and macro-level collision prediction models at neighborhood level (e.g., Hadayeghi et al., 2003; Lovegrove, 2007; Lovegrove and Sayed, 2006a). Hadayeghi et al. (2010b) and Hadayeghi (2009) considered different types of data, including street networks, land use, traffic demands, socioeconomic and demographic characteristics, dwelling units, and employment, in the city of Toronto to reduce the number of collisions and, consequently, improve safety. Hadayeghi (2002) proposed that the number of households, major road kilometers, vehicle kilometers traveled, intersection density, posted speed and volume/speed ratio can impact accident occurrence. He proposed models for predicting the number of accidents per zone in the city of Toronto. Wei and Lovegrove (in press) proposed an empirical tool based on four groups of data as independent variables including exposure, socio-demographics, transportation demand management and road network in evaluating the safety of cyclists. These variables are not usually considered at the micro level (Khondakar et al., 2010). Detailed descriptions of these four groups are given
Fig. 1. Examples of street patterns (Rifaat et al., 2011).
116
M. Moeinaddini et al. / Safety Science 62 (2014) 114–120
in different studies, such as Lovegrove (2007) and Lovegrove and Sayed (2006a). In addition, Guevara et al. (2004) found that the significant factors of injury and property-damage crash models in Tucson, Arizona, are the number of employees, population density, intersection density, percentage of miles of principal arterial, percentage of miles of minor arterials, and percentage of miles of urban collectors. Hadayeghi et al. (2006) examined the temporal transferability of the zonal accident prediction models for the city of Toronto in 1996 and 2001. Hadayeghi et al. (2010a) considered land use, traffic, demographic, dwelling units, employment types and street networks, including number of rail stations, total rail kilometer, number of schools, total arterial road, expressway, collector, laneway, local road and ramp kilometers, number of 4-legged and 3-legged signalized intersections, as independent variables in developing planning level transportation safety tools using a geographically weighted Poisson regression. Although these studies evaluated different effective factors, they cannot show street network and transport fatality relationships across different cities. There have also been other efforts to classify street patterns. Lovegrove (2007) proposed that, in residential areas, cul-de-sac road networks are popular for limiting through-traffic but accessibility for emergency vehicles and transit is significantly less than a grid network. Lovegrove and Sayed (2006b) suggested that 3-way offset road networks are safer than grid networks and cul-de-sac networks. Sun (2009) proposed that the 3-way offset and Fused Grid road networks were the safest road patterns. The 3-way offset pattern also features significantly better road safety and accessibility than other street patterns (Ogden, 1996). Wei and Lovegrove (2012) reviewed some studies of neighborhood road patterns that can save the lives of vulnerable road users. However, these efforts cannot illustrate urban street patterns across different cities. Many researchers have proposed traditional micro-level collision prediction models, but less attention has been given to the transferability of macro-level collision prediction models (Khondakar et al., 2010). Although the effects of different factors have been explored in previous research, limited research has been conducted on exploring the impacts of street network factors on transport fatalities in different contexts (at the city level) with a macro-level scale. As a part of continuing efforts in this area, this study examines how these factors, including length of roads, blocks per area, nodes per selected areas and length of motorways, contribute to crash fatalities. This study uses data from 20 cities around the world to estimate this relationship in different contexts.
2. Materials and methods This paper focuses on the relationship between street network factors and transport fatalities in different contexts with a macro-level scale by considering street network variables, such as blocks per area, nodes per selected areas and length of roads and motorways, as independent variables and the number of passenger transport fatalities per million inhabitants as the dependent variable (refer Table 1). The data that are used in this study are collected from the mobility in cities database (UITP, 2006) and the Open Street Map (OSM) web site. The UITP database has 120 urban mobility indicators in 52 cities from various parts of the world, but the number of urban street network indicators is limited to 20 cities in this database. The cities are chosen based on availability of data for all indicators after removing the outliers (refer Table 1). In addition, there are limited indicators for street networks in the UITP data source. There are only four indicators for street networks at the city level that indicate the length of roads and length of motorways in different units (per hectare and per inhabitant). Other UITP street indicators, such as road investments and speed,
cannot be illustrated in the street network at the city level. These indicators also cannot represent the shape of real urban street networks. Because the real shapes of street networks are illustrated in street maps, this research also uses the OSM data source to include indicators of urban street network shape. Because block size and intersection density are important factors in street networks, connectivity and shape indicators are illustrated in this study by blocks per area and nodes per selected areas (refer Fig. 2). The number of blocks per area is the number of polygons that are bounded by street lines in a selected area. UITP database defined Metropolitan and central business district for each city in the database. The selected area is chosen based on UITP Metropolitan definition. The number of nodes per selected areas is the number of nodes in a selected area per the number of polygons that are bounded by street lines in that area. The number of blocks per area can show the size of the street network blocks, and the number of nodes per selected areas can introduce the shape of them. Blocks per area and nodes per selected areas can be achieved for urban streets by modifying and analyzing OSM data. OSM data are digital maps of street networks with various layers. Because only the street layer is needed in this research, these maps should be edited in JOSM software (Java open street map). JOSM is a free program for modifying OSM maps that is provided by the OSM web site. This software is user friendly, and the streets layer can be selected easily. This selected layer can be saved as a new .gpx file, and then this file can be analyzed in ArcGIS software. There are three different layers in the .gpx file when it is opened in ArcGIS. These layers are displayed as streets that are lines, intersections and junctions that are nodes and the selected area that is a polygon. Nodes also include the points at which a street changes direction (mid-block nodes in Fig. 2 are used to show these nodes). The streets layer is converted to polygons using Arc Toolbox to create polygons that are bounded by street lines. The intersect tool in Arc Toolbox can provide street polygons in the selected area and the number of nodes in that area. Therefore, the block (polygon) density and nodes per selected areas are the result of open street map analysis in ArcGIS software after editing the maps in JOSM software. Fig. 2 illustrates the result of converting the OSM data to block density and nodes per selected areas. The number of blocks per area can illustrate the size of street blocks, and nodes per selected areas show the average road segment density for blocks. Although the type of connection (e.g., 4-way, 3-way, interchange, etc.) is not considered in this macrolevel study, more nodes can indicate more routes. Therefore, these indicators also represent connection density. The length of motorway per ten thousand inhabitants, the length of road per million inhabitants and the number of passenger transport fatalities per million inhabitants are collected from UITP. There are also data on the length of road and length of motorway per urban hectare in the UITP database. Because the dependent variable is transport fatalities per inhabitant, the length of road and length of motorway per inhabitant are also selected in order to illustrate the relationship between transport fatalities and street length in the context of population. This issue shows the impacts of population growth, length of road and length of motorway on road safety. In addition, blocks per area and nodes per selected areas can indicate land development. The roads in this research are different from motorways. The roads can be used by motorized vehicles, pedestrians and cyclists but the motorways can be used by motorized vehicles while walking and cycling users do not experience benefits from them. Table 1 presents the data that are used in this study. Different methods have been successfully used by previous researchers in different approaches to crash studies. Clark and Cushing (2004) used linear regression. Linear regression with a
117
M. Moeinaddini et al. / Safety Science 62 (2014) 114–120 Table 1 Research data. Cities
Blocks per area
Nodes per selected areas
Length of road per million inhabitants (m)
Length of motor way per ten thousand inhabitants (m)
Passenger transport fatalities per million inhabitants
Amsterdam Athens Barcelona Berlin Bern Bilbao Bologna Brussels Budapest Chicago Clermont Ferrand Copenhagen Dubai Dublin Geneva Ghent Glasgow Graz Hamburg Helsinki Hong Kong Krakow Lille Lisbon London Lyons Madrid Manchester Marseilles Melbourne Milan Moscow Munich Nantes Newcastle Oslo Paris Prague Rome Rotterdam Sao Paulo Seville Singapore Stockholm Stuttgart Tallinn Tunis Turin Valencia Vienna Warsaw Zurich
53.6 – 54.97 – – – 18.2 – – – – 21.96 – – 26.96 – 35.26 20.03 – – 18.08 – – 37.61 53.16 34.96 59.48 – 23.14 – – – 52.5 32.16 – – 46.12 – 58.03 62.11 – 52.55 – 38.08 29.06 – – 35.09 – 46.37 20.38 –
6.71 – 4.89 – – – 8.55 – – – – 6.48 – – 8.26 – 6.91 9.8 – – 9.51 – – 7.01 8.48 5.11 5.43 – 8.52 – – – 6.3 10.44 – – 6.12 – 6.89 5.88 – 5.8 – 7.83 11.77 – – 4.24 – 5.77 8.77 –
2.75 2.31 2.1 1.57 3.92 4.36 2.49 1.94 2.43 4.77 3.4 3.85 3.1 4.26 4.9 5.48 5.8 4.4 – 3.61 0.284 1.45 3.48 0.889 2.03 2.47 4.87 3.7 1.63 – – 0.406 1.83 5.41 4.12 5.86 1.98 2.91 2.8 4.07 1.96 2.02 0.94 – 1.19 2.17 – 2.71 2.87 1.81 1.68 4.7
– 3.9 8.97 2.02 15.7 12 15.2 3.63 1.36 9.53 4.17 12.5 39.5 – 10.3 14.6 11.1 7.96 – 9.08 1.65 2.24 14.4 8.68 0.99 6.81 9.83 7.09 4.13 – – 1.14 4.78 12.3 12.9 17 6.95 6.44 11.1 – 2.02 11.6 4.52 26.9 4.92 2.51 – 17.2 2.87 2.64 4.5 10.9
33 77 84.3 19.2 34.8 54.3 99.1 45.6 63.1 79.7 114 46.9 203 47.2 40.5 – 52.7 39.8 34.5 20.6 30 – 24.5 73.1 41.8 22.1 71 42.2 62.5 77.4 – 47.1 22.3 64.9 23.2 38.7 65.9 57.5 108 33.9 109 103 58.4 36.4 57.2 65.2 – 97.9 33.8 26.4 – 44.6
spatial lag parameter was used by Levine et al. (1995). Chatterjee et al. (2004) utilized Poisson models. Negative binomial (NB) models were also used by numerous studies (e.g., Lovegrove and Sayed, 2006b; Wei and Lovegrove, in press; Abdel-Aty and Radwan, 2000; Hadayeghi et al., 2003 and Hadayeghi et al., 2007; Maher and Summersgill, 1996; Miaou and Lord, 2003; Mitra and Washington, 2007; Noland and Quddus, 2004; Memon, 2006). Zero-inflated Poisson and NB models (ZIP/ZINB) were used by Qin et al. (2004) and Shankar et al. (1997). Guevara et al. (2004) utilized simultaneous estimation of NB models. Full Bayes hierarchical models with spatial effects were used by Aguero-Valverde and Jovanis (2006) and Quddus (2008). Hadayeghi et al. (2003) proposed a geographically weighted regression (GWR). Log-linear models using ordinary least squares (OLS) estimation were used by Washington et al. (2006) and Wier et al. (2009). Road collisions are random and rare events. Moreover, count data are non-negative and discrete. Therefore, conventional linear
regression models with a normally distributed error structure are not suitable for modeling road collisions. Poisson and NB regressions in the generalized linear model (GLM) framework have been more successfully adopted for this issue (Hadayeghi, 2009). Kim et al. (2002) found that Poisson Gamma and Poisson lognormal models can be more successful for the prediction of collisions. The Poisson distribution has been shown to be suitable for modeling crash data for a single site, but crash data for different sites often exhibit large variances and small means that lead to overdispersion when there is a variance-to-mean ratio greater than 1 (Hadayeghi, 2009). Therefore, the NB distribution, also known as the Poisson–gamma distribution, has become the most commonly used method for modeling crash data at a series of sites (Hadayeghi, 2009). In previous collision prediction models studies, GLMs were commonly used. They have been successful in interpreting random, rare, sporadic, and non-negative collision data (Lovegrove et al., 2010; Wei and Lovegrove, in press).
118
M. Moeinaddini et al. / Safety Science 62 (2014) 114–120 Table 5 Parameter estimates. Parameter
(Intercept) Blocks per area Nodes per selected areas Length of road per million inhabitants Length of motor way per ten thousand inhabitants (Scale)
Fig. 2. An illustration of blocks and nodes.
Table 2 Continuous variable information. Indicators
N
Minimum
Maximum
Mean
Std. deviation
Passenger transport fatalities per million inhabitants Blocks per area Nodes per selected areas Length of road per million inhabitants Length of motor way per ten thousand inhabitants
20
22
108
60.47
27.435
20 20 20
18.1 4.24 .28
59.5 11.77 5.80
37.785 7.3140 2.7732
14.1208 2.02433 1.56602
20
.99
17.20
8.4805
4.40316
B
Std. error
95% Wald confidence interval
Hypothesis test
Lower
Upper
Wald ChiSquare
df
Sig.
1.922 .018 .114
.5343 .0054 .0429
.874 .008 .030
2.969 .029 .198
12.933 11.329 7.058
1 1 1
.000 .001 .008
.140
.0453
.229
.051
9.547
1
.002
.114
.0181
.079
.150
39.563
1
.000
.068a
.0214
.037
.126
Dependent variable: passenger transport fatalities per million inhabitants. Model: (Intercept), blocks per area, nodes per selected areas, length of road per million inhabitants, length of motor way per ten thousand inhabitants. a Maximum likelihood estimate.
Table 6 Omnibus testa. Likelihood ratio Chi-Square
df
Sig.
23.742
4
.000
Dependent variable: passenger transport fatalities per million inhabitants. Model: (Intercept), blocks per area, nodes per selected areas, length of road per million inhabitants, length of motor way per ten thousand inhabitants. a Compares the fitted model against the intercept-only model. Table 3 Case processing summary.
Included Excluded Total
N
Percent (%)
20 32 52
38.5 61.5 100.0
Table 4 Correlation matrix.
Table 7 SD and Pearson Chi-Square goodness of fit.
Deviance Scaled deviance Pearson Chi-Square Scaled Pearson Chi-Square
Value
df
Value/df
1.382 20.228 1.356 19.851
15 15 15 15
.092 .090
Pearson correlation
R
M
P
N
Dependent variable: passenger transport fatalities per million inhabitants. Model: (Intercept), blocks per area, nodes per selected areas, length of road per million inhabitants, length of motor way per ten thousand inhabitants.
R M P N
1 .510 .052 .012
.510 1 .101 .276
.052 .101 1 .554
.012 .276 .554 1
the NB model, the Gamma model assumes a log link (refer to Eq. (1)):
R = length of road per million inhabitants. M = length of motorway per ten thousand inhabitants. P = blocks per area. N = nodes per selected areas.
In this study, the dependent variable is the number of passenger transport fatalities per million inhabitants. It is considered to be similar to count data, where the number of deaths only takes positive and discrete values. On the other hand, this variable is a scaled factor because it is divided by inhabitants. Lognormal and Gamma with log link models are used to scale data in the GLM framework. Lognormal is not in the exponential family (e.g., Poisson, Gamma, Binominal, Inverse, Negative Binominal, etc.). Therefore, GLMs with a gamma distribution are recommended because they performed slightly better than the log transformed model. Similar to
i X Y ¼ EXP b0 þ bi X i
! ð1Þ
i¼1
where Y is the dependent variable, i the subscript showing the number of independent variables, X the independent variable, b0 the constant, calculated in the calibration process and bi is the coefficient of the independent variable, calculated in the calibration process of the model. 3. Results In this section, the results of the street network GLM analysis are discussed. Tables 2 and 3 indicate descriptive statistics for the variables included in the model. Table 4 indicates that there is no strong correlation between the independent variables
M. Moeinaddini et al. / Safety Science 62 (2014) 114–120
included in the model. Table 5 indicates the variables included in the model, their parameter estimates, and the significance of the parameters (5% level). The Omnibus test, likelihood ratio ChiSquare test statistics, the scaled deviance (SD) and the Pearson Chi-Square statistic show the model goodness of fit (refer to Tables 6 and 7). Therefore, the final model can be defined as follows:
PF ¼ EXPð1:992 0:140R þ 0:018P þ 0:114N þ 0:114MÞ
ð2Þ
where PF is the passenger transport fatalities per million inhabitants, R the length of road per million inhabitants, P the blocks per area, N the nodes per selected areas and M is the length of motorway per ten thousand inhabitants This model shows that passenger transport fatalities in this model are significantly affected by street network indicators. Among these indicators, the number of nodes per selected areas and the length of motorway per inhabitant have higher positive parameters, so these indicators have greater effects on passenger transport fatalities in this model. The third effective indicator with a positive relationship is blocks per area. The length of road per inhabitant has negative parameter and thus it is inversely associated with transport fatalities in this model. Overall, fewer nodes per selected areas and blocks per area, shorter motorways per inhabitant and more roads per inhabitants are effective urban street network indicators that are associated with fewer transport fatalities. This suggests that macro-scale street network planning that reduces block density, nodes per selected areas and motorway density and increases road density could contribute to fewer transport fatalities across different cities. 4. Discussion and conclusions Macro-scale urban street network indicators as independent variables and passenger transport fatalities as the dependent variable are analyzed in this research using GLM analysis. From this study, it can be concluded that the length of road per million inhabitants, blocks per area, nodes per selected areas and length of motorway per ten thousand inhabitants are the urban street network factors that are associated with passenger transport fatalities. The model shows that more blocks per area are correlated with more passenger transport fatalities. Distributing more blocks in urban areas can produce more nodes. These nodes produce more conflicts. On the other hand, more and smaller blocks create more connectivity, and this arrangement produces more direct routes for alternative travel modes. However, the model shows that more blocks are also correlated with more passenger transport fatalities involving vehicle transport. This suggests that considering more connected routes to be used exclusively by non-motorized modes is more effective at allowing fewer conflicts and transport fatalities involving vehicle transport than creating more blocks and nodes to increase connectivity for all modes. In addition, the model shows that more nodes per selected areas are associated with more passenger transport fatalities. Consequently, more simple, large blocks with fewer nodes are correlated with fewer fatalities. The model also indicates that longer motorway per inhabitant is correlated with more passenger transport fatalities, while road length has a negative relationship with fatalities. More motorways can increase convenience for private motorized vehicle users. This type of road belongs to motorized vehicles users with high speed, while walking and cycling users do not experience benefits from them. Therefore, streets patterns with more facilities, such as motorways for high speed car users, do not contribute to fewer fatalities. This study evaluates the relationship of urban street indicators and passenger transport fatalities in different cities within various contexts at the city level. Therefore, the results of this study can
119
illustrate this relationship regardless of context effects. In addition, this study also attempts to use some factors, such as nodes and blocks, to describe the shapes of the street networks at a macro scale. Moreover, it attempts to address street network indicators at the city level (instead of the neighborhood level) that have not been sufficiently studied in the literature. Currently, more sustainable urban areas are needed. To promote these kinds of areas, having fewer traffic-related fatalities is an important goal. Therefore, this research tries to indicate the relationship between urban street indicators and transport fatalities. This relationship can be used in fatality reduction strategies in different cities. In addition, because the urban street network is the result of macro-scale planning decisions, considering the relationship of urban street indicators and fatalities can lead to better planning decisions for new cities. This research is limited to selected UITP and open street map web site data in selected cities. However, the universal scale relationship models may be replicated by future studies for other urban structure indicators in order to identify effective factors for sustainable travel patterns regardless of context effects. In addition, this research uses OSM data sources that are free editable maps because we had financial limitations. Further studies may enhance the model by using the original versions of urban maps. Moreover, urban structure indicators change rapidly; future studies may evaluate these changes by updating their data sources and examining the relationship between urban structure indicators and fatalities in various parts of the world to address the changing results.
References Abdel-Aty, M., Chundi, S., Lee, C., 2007. Geo-spatial and log-linear analysis of pedestrian and bicyclist crashes involving school-aged children. J. Saf. Res. 38, 571–579. Abdel-Aty, M.A., Radwan, A.E., 2000. Modeling traffic accident occurrence and involvement. Accident Anal. Prevent. 32, 633–642. Aguero-Valverde, J., Jovanis, P.P., 2006. Spatial analysis of fatal and injury crashes in Pennsylvania. Accident Anal. Prevent. 38, 618–625. Augustus, A., 2012. Determinants of road traffic accident occurrences in Lagos state: some lessons for Nigeria. Int. J. Humanities Soc. Sci. 2, 252–259. Chatterjee, A., Everett, J.D., Reiff, B., Schwetz, T.C., Seaver, W.L., Wegmann, F.J., 2004. Tools for assessing safety impact of long-range transportation plans in urban areas, In: 80th Annual Meeting of the Transportation Research Board, TRB, Washington, DC. Clark, D.E., Cushing, B.M., 2004. Rural and urban traffic fatalities, vehicle-miles and population density. Accident Anal. Prevent. 36, 967–972. Dee, T.S., 1999. State alcohol policies, teen drinking, and traffic fatalities. J. Publ. Econ. 72, 289–315. De Leur, P., Sayed, T., 2001. Using claims prediction model for road safety evaluation. Can. J. Civ. Eng. 28, 804–812. Eluru, N., Bhat, C., Hensher, D., 2008. A mixed generalized ordered response model for examining pedestrian and bicyclist injury severity level in traffic crashes. Accident Anal. Prevent. 40, 1033–1054. Evans, L., 1986. The effectiveness of safety belts in preventing fatalities. Accident Anal. Prevent. 18, 229–241. Grimm, M., Treibich, C., 2012. Determinants of Road Traffic Crash Fatalities across Indian States. Health Economics. Guevara, F.L.D., Washington, S.P., Oh, J., 2004. Forecasting crashes at the planning level: simultaneous negative binomial crash model applied in Tucson, Arizona. Transp. Res. Rec. 1897, 191–199. Hadayeghi, A., 2002. Accident Prediction Models for Safety Evaluation of Urban Transportation Network. Department of civil engineering. University of Toronto, Ontario, Canada. Hadayeghi, A., 2009. Use of Advanced Techniques to Estimate Zonal Level Safety Planning Models and Examine Their Temporal Transferability. Department of Civil Engineering. University of Toronto, Ontario, Canada. Hadayeghi, A., Shalaby, A., Persaud, B., 2003. Macro-level accident prediction models for evaluating the safety of urban transportation systems. Transp. Res. Rec. 1840, 87–95. Hadayeghi, A., Shalaby, A., Pesuad, B., 2007. Safety prediction models: a proactive tool for safety evaluation in urban transportation planning applications. Transp. Res. Rec. 2019, 225–236. Hadayeghi, A., Shalaby, A., Persaud, B., 2010a. Development of planning level transportation safety tools using geographically weighted poisson regression. Accident Anal. Prevent. 42, 676–688.
120
M. Moeinaddini et al. / Safety Science 62 (2014) 114–120
Hadayeghi, A., Shalaby, A., Persaud, B., 2010b. Development of planning-level transportation safety models using full bayesian semiparametric additive techniques. J. Transp. Saf. Security 2, 45–68. Hadayeghi, A., Shalaby, A., Persaud, B., Cheung, C., 2006. Temporal transferability and updating of zonal-level accident prediction models. Accident Anal. Prevent. 38, 579–589. Khondakar, B., Sayed, T., Lovegrove, G., 2010. Transferability of community-based collision prediction models for use in road safety planning applications. J. Transp. Eng. 136, 871–880. Kim, H., Sun, D., Tsutakawa, R.K., 2002. Lognormal vs. gamma: extra variations. Biometrical J. 44, 305–323. Kopits, E.-A., 2004. Traffic Fatalities and Economic Growth. Department of Economics. University of Maryland, College Park. Lapparent, M., 2005. Individual cyclists’ probability distributions of severe/fatal crashes in large French urban areas. Accident Anal. Prevent. 37, 1086–1092. Levine, N., Kim, K.E., Nitz, L.H., 1995. Spatial analysis of Honolulu motor vehicle crashes. II. Zonal generators. Accident Anal. Prevent. 27, 675–685. Lovegrove, G., 2007. Road Safety Planning, New Tools for Sustainable Road Safety and Community Development. Verlag Dr, Müller, Germany. Lovegrove, G., Lim, C., Sayed, T., 2010. Community-based, macro-level collision prediction model use with a regional transportation plan. J. Transp. Eng. 136, 120–128. Lovegrove, G., Sayed, T., 2006a. Macro-level collision prediction models for evaluating neighbourhood traffic safety. Can. J. Civ. Eng. 33, 609–621. Lovegrove, G., Sayed, T., 2006b. Using macro-level collision prediction models in road safety planning applications. Transp. Res. Rec.: J. Transp. Res. Board 1950, 73–82. Maher, M.J., Summersgill, I., 1996. A comprehensive methodology for the fitting of predictive accident models. Accident Anal. Prevent. 28, 281–296. Mann, R.E., MacDonald, S., Stoduto, G., Bondy, S., Jonah, B., Shaikh, A., 2001. The effects of introducing or lowering legal per se blood alcohol limits for driving: an international review. Accident Anal. Prevent. 33, 569–583. Marshall, W.E., Garrick, N.W., 2011. Does street network design affect traffic safety? Accident Anal. Prevent. 43, 769–781. McCarthy, P., 2001. Effect of speed limits on speed distributions and highway safety: a survey of recent literature. Transp. Rev. 21, 31–50. Memon, A.Q., 2006. Road Accident Prediction Models Developed from a National Database: Poisson and Negative Binomial Regressions, In: Transportation Research Board 85th Annual Meeting CD-ROM Washington DC, United States. Miaou, S.P., Lord, D., 2003. Modeling traffic crash-flow relationships for intersections: dispersion parameter, functional form, and Bayes versus empirical Bayes. Transp. Res. Rec. 1840, 31–40. Mitra, S., Washington, S., 2007. On the nature of over-dispersion in motor vehicle crash prediction models. Accident Anal. Prevent. 39, 459–468. Noland, R.B., 2003. Traffic fatalities and injuries: the effect of changes in infrastructure and other trends. Accident Anal. Prevent. 35, 599–611. Noland, R.B., Quddus, M.A., 2004. Improvements in medical care and technology and reductions in traffic-related fatalities in Great Britain. Accident Anal. Prevent. 36, 103–113. Ogden, K.W., 1996. Safer Roads: A Guide to Road Safety Engineering. Avebury Technical.
Permpoonwiwat, C.K., Kotrajaras, P., 2012. Pooled time-series analysis on traffic fatalities in Thailand. World Rev. Business Res. 2, 170–182. Qin, X., Ivan, J.N., Ravishanker, N., 2004. Selecting exposure measures in crash rate prediction for two-lane highway segments. Accident Anal. Prevent. 36, 183– 191. Quddus, M.A., 2008. Modelling area-wide count outcomes with spatial correlation and heterogeneity: an analysis of London crash data. Accident Anal. Prevent. 40, 1486–1497. Rifaat, S.M., Tay, R., De.Barros, A., 2011. Effect of street pattern on the severity of crashes involving vulnerable road users. Accident Anal. Prevent. 43, 276–283. Ruhm, C.J., 1996. Alcohol policies and highway vehicle fatalities. J. Health Econ. 15, 435–454. Sass, T.R., Zimmerman, P.R., 2000. Motorcycle helmet laws and motorcyclist fatalities. J. Regul. Econ. 18, 195–215. Shankar, U., 2003. Pedestrian Roadway Fatalities. National Center for Statistics and Analysis Advanced Research and Analysis. U.S. Department of Transportation National Highway Traffic Safety Administration. Shankar, V., Milton, J., Mannering, F., 1997. Modeling accident frequencies as zeroaltered probability processes: an empirical inquiry. Accident Anal. Prevent. 29, 829–837. Snyder, D., 1989. Speeding, coordination, and the 55 MPH speed limit: comment. Am. Econ. Rev. 79, 922–925. Sun, J., 2009. Sustainable Road Safety: Development, Transference and Application of Community-Based Macro-Level Collision Prediction Models, Civil Engineering. University of British Columbia, Okanagan. UITP, 2006. Mobility in cities Database. International Association of Public Transport (UITP). Wagenaar, A., Maybee, R., Sullivan, K., 1988. Mandatory seat belt laws in eight states: a time-series evaluation. J. Saf. Res. 19, 51–70. Washington, S., Schalwyk, I.V., Meyer, M., Dumbaugh, E., Zoll, M., 2006. Incorporating Safety into Long-Range Transportation Planning. NCHRP Report 546, TRB, Washington, DC. Wei, F., Lovegrove, G., (in press). An empirical tool to evaluate the safety of cyclists: community based, macro-level collision prediction models using negative binomial regression. Accident Analysis and Prevention. Wei, V.F., Lovegrove, G., 2012. Sustainable road safety: a new (?) neighbourhood road pattern that saves VRU lives. Accident Anal. Prevent. 44, 140–148. Whetten-Goldstein, K., Sloan, F., Stout, E., Liang, L., 2000. Civil liability, criminal law, and other policies and alcohol-related motor vehicle fatalities in the United States: 1984–1995. Accident Anal. Prevent. 32, 723–733. Wier, M., Weintraub, J., Humphreys, E.H., Seto, E., Bhatia, R., 2009. An area-level model of vehicle-pedestrian injury collisions with implications for land use and transportation planning. Accident Anal. Prevent. 41, 137–145. Wilkinson, J.T., 1987. Reducing drunken driving: which policies are most effective? Southern Econ. J. 54, 322–334. World Health Organization (WHO), 2009. Global Status Report on Road Safety. Time for Action. World Health Organization, Geneva.