The role of road safety in a sustainable urban mobility: An econometric assessment of the Spanish cities

1 THE ROLE OF ROAD SAFETY IN A SUSTAINABLE URBAN MOBILITY: AN ECONOMETRIC ASSESSMENT OF THE SPANISH NUTS-3 CASE. Authors Mercedes Castro-Nuño ([email protected]). Applied Economics & Management Research Group, University of Seville (Spain) José I. Castillo-Manzano ([email protected]). Applied Economics & Management Research Group, University of Seville (Spain) Xavier Fageda ([email protected]). Dept. of Economic Policy, University of Barcelona (Spain) ABSTRACT: There has been a structural change in mobility in major Spanish cities in recent decades, with a switch to the pattern followed in other countries in the area. A shift has taken place from a traditional Mediterranean model to a North American city stereotype, with uncontrolled motorization and major implications for public health. This article specifically analyzes negative road safety related externalities that result from this process, given that the trend seems to show a steady decline in road safety accidents on urban roads in Spain, with major differences among NUTS-3 provinces. The objective is to evaluate the factors that empirically explain these differences for the 2003-2013 period using a panel data analysis. Results show that a key role is played by urban development variables, such as population density and improvements in health services, with advances linked to more accessible and sustainable urban transportation, such as the Smart City concept. Not only does this article close a gap in the literature, but the findings can also serve as a practical guide for the development and implementation of 2 urban mobility and road safety plans, and reveals the special needs of the most vulnerable groups. Keywords: Road Safety, Urban Areas, Smart City, Panel Studies, NUTS-3 provinces. JEL Codes: C33, O18, R41. Authors confirm that neither the manuscript (nor any part of it) has been published previously and is not under consideration for publication elsewhere. 1. Introduction The analysis of current and future road transport trends in large cities is a research topic of growing international importance (Archer and Vogel, 2000). High levels of traffic congestion remain one the most visible transport problems in any developed city, as a consequence of the increased number of private vehicles that come from the uncontrolled motorization (Yigitcanlar et al., 2015), which, according to a general consensus, may also be related to both the lack of an effective/attractive public transit system and opportunities for walking and cycling (Beirao & Cabral, 2007; Julsrud & Denstadli, 2017). Factors such as the mass movement of population from rural areas to cities, high human density, urban sprawl and the consequent increase in daily commutes by private cars (Dumbaugh and Rae, 2009; Vassallo et al., 2012) have turned urban space into a scenario not only with environmental externalities (air pollution, noise) (Kersys, 2015), but also with other health negative risks such as traffic accidents (Shekari et al., 2016). 3 In recent decades, European cities have moved on from the traditional Mediterranean model, in which economic and leisure activities were clustered together in the city center, and journeys were predominantly made on foot or on public transport (Salvati and Gargiulo Morelli, 2014). There has been a proliferation of the North Americaninfluenced city stereotype, characterized by uses and services being widespread and the migration of population to metropolitan residential areas, forcing citizens to invest more time in their journeys and triggering a modal change toward motorized means of transport (Dura-Guimera, 2003). Nevertheless, despite of the relationship between urban mobility and road safety, most efforts follow an accessibility and sustainability approach that focuses on environmental issues, and the literature that has addressed urban sustainability from a safety dimension is both more recent and less abundant. This literature shows that accidents that occur on urban roads present differentiated risk factors from interurban road accidents, with urban morphology having special repercussions (Gomes, 2013; Ma et al., 2010; Moeinaddini et al., 2014; Scott et al., 2016), especially road intersections (Ferreira and Couto, 2013). In addition, the wide variety of collectives should not be neglected: pedestrians, cyclists, young drivers, motorcycles and scooters, private cars, transportation vehicles and public transit vehicles all share a limited amount of urban space and this complicates mobility (Chang et al., 2016; Jevtić et al., 2015; Maestracci et al., 2012; Tay et al., 2011). Researchers such as Rakauskas et al. (2009) and Zwerling et al. (2005) analyze factors that could explain accidents in urban areas. For example, on highways drivers travel longer distances, weather conditions and the state of the roads can have a greater effect on driving, speeds are higher and, above all, the speed and quality of post-accident 4 trauma attention play a crucial role (the “golden hour”, Castillo-Manzano et al., 2014; Sánchez-Mangas et al., 2010). Other studies (Archer and Vogel, 2000; Moeinaddini et al., 2015; Vorko-Jović et al., 2006) point to factors such as traffic congestion, greater frequency of journeys (greater exposure to risk), different alcohol/drug consumption patterns that are especially acute at weekends, and drivers’ different attitudes and behaviors. Studies by Clark (2003), Clark and Cushing (2004), Ewing et al. (2003, 2016), Suhkai and Jones (2014) and Yeo et al., (2015) are especially significant. These authors analyzed the influence of variables such as the degree of urban sprawl, with an almost exponential increase in the number of private vehicle journeys from metropolitan areas to inner cities. This is the current article’s object of study for the specific case of urban roads in Spanish provinces (EU statistical classification NUTS-3 regions), for which to date there has been a marked lack of literature addressing the issue of urban accidents disaggregated on the territorial scale. This article seeks to fill the gap in the literature on urban traffic accidents in Spain, as the precedents that explore the problem on the regional and provincial levels (Albalate et al., 2013; Gómez-Barroso et al., 2015; Redondo-Calderón et al., 1999; Rivas-Ruiz et al., 2007; Úbeda et al., 2016; TolónBecerra et al., 2009; 2013) only consider the effects of accidents on inter-city roads, or address accidents in urban areas in a single province or city (De Oña et al., 2011, 2013; García-Altés and Pérez, 2007; Kanaan et al., 2009; Melchor et al., 2015; Prat et al., 2015). The main contribution of this article is to determine the statistically significant factors that explain the different urban road safety outcomes at the provincial level in Spain, 5 subsequently focusing on the main and most populated cities (provincial capitals). A negative binomial multivariate model is formulated and applied to original panel data for the 2003-2013 period. The results are an essential step forward in the ability to tackle the issue of traffic accidents on Spanish city roads with a greater guarantee of success. The article is structured as follows: following this Introduction, the second section presents the database used and the methodology applied; the third section gives the results and the corresponding discussion; and lastly, a conclusions section is included and the paper is completed with the bibliography. 2. Empirical framework: data, variables and methodology Using available official data, an original database has been constructed for urban road safety that is unprecedented in Spain. Taking into account that the aim of the current study is to assess road safety in urban settings at the provincial level in Spain, panel data were constructed ad-hoc for a sample made up of the 50 provinces into which Spain is politically and administratively organized (EU statistical classification NUTS-3 regions), excepting Ceuta and Melilla. The two Autonomous Cities of Ceuta and Melilla in North Africa were excluded from the analysis due to the skew that they would have introduced into the sample due to their small size and special characteristics compared to the other cities. The period of analysis spans from 2003 (when Law 57/2003 concerning Large Cities was passed) to 2013 (the most recent year for which standardized data can be found). Figures 1 and 2 show differences in urban road safety in the 50 Spanish provinces considered in the analysis to facilitate understanding of the database. 6 [PLEASE INSERT FIGURE 1 NEAR HERE] [PLEASE INSERT FIGURE 2 NEAR HERE] Table 1 describes the variables considered (same naming as in the model, below) and the sources from which they are taken. The unit of observation is the province-year pair. Variables can be split into two groups: 1. Endogenous. Accidents and fatalities within 30 days (following the Vienna Convention’s international criterion) recorded on urban roads in Spanish NUTS-3 provinces. 2. Explanatory. Sample economic, social-demographic, geographic and mobility heterogeneity is captured with different categories of variables: demographic (population density and population structure by age); economic (economic activity and sectoral structure); health network (hospital density); geographic (latitude and longitude); mobility (motorization); and accessibility/sustainability conditions affecting urban transit (Smart City, alternative non-motorized public transit modes). [PLEASE, INSERT TABLE 1 NEAR HERE] This panel has been econometrically treated with the STATA package using an econometric model that takes province i during period t in the expression (1): Yit = α + βkXit + γkZit + ν’Yearit + εit, (1) where Yit are explained road safety variables for either the total number of urban traffic accidents or the number of urban traffic fatalities per accident (within the following 30 days, according to the Vienna Convention); Xit contains the vector of attributes of each province (the explanatory variables relating to demographic, economic, motorization, 7 geographic and health characteristics); Zit are dummy variables that identify the accessibility and sustainability conditions in which urban mobility takes place (regarding Smart City status and the availability of a subway and/or urban rail system). Dummies are also included for Year to capture the common time trend in all the provinces; and εit is the mean zero random error. The explanatory variables considered in the two models (one for each explained variable: urban accidents and fatalities) are based on factors typically analyzed in previous road safety studies (Albalate and Bel, 2012; Castillo-Manzano et al., 2013, 2014, 2015, 2016; Dee,1999; and Tolón-Becerra et al., 2013 for Spanish NUTS-3 provinces, but in relation to accidents on intercity roads), and the few precedents that exist on urban road safety (for example, Vorko-Jović et al., 2006, and Yannis et al., 2015 for the case of the principal European cities). With respect to their description, as can be observed in Table 1, Gross Domestic Product (GDP) per capita is used to test for any relationship between economic development and road fatalities. According to the prior literature, a priori it is not clear what the sign of the coefficient related to this variable should be. On the one hand, the road accident fatality rate could rise along with a country’s economic development due to greater risk exposure (Kopits and Cropper, 2005), but on the other hand, the relationship between economic development and the road accident fatality rate could reduce and the trend could even reverse once a certain income level has been reached (Bishai et al, 2006). In this article, both GDP per capita and GDP per capita squared are considered as exogenous variables. Following prior studies (Albalate and Bel, 2012; Castillo-Manzano et al., 2013, 2015; Kopits and Cropper, 2005), Table 1 also includes a variable for the motorization rate. 8 However, the relationship that should be expected for this variable is not very clear, either. On the one hand, higher motorization levels could imply greater exposure to traffic accidents but, on the other, more developed countries would benefit from better infrastructure and better vehicles, more progressive policies and more beneficial social attitudes toward road safety (Castillo-Manzano et al, 2013). It should be mentioned that the GDP per capita and motorization variables might capture similar effects, although the correlation found between the two (see Table 3, below) is not sufficiently high to consider the existence of a collinearity problem. From the economic point of view, Table 1 also shows that variables have been considered for the relative importance that the manufacturing sector and the construction sector have on total employment in the province. Both of these variables seek to capture the approximate impact that heavy truck traffic flows due to industrial and construction activities may have on road safety in Spanish urban areas. According to earlier studies (Castillo-Manzano et al., 2015, 2016; Dablanc et al., 2013; Dong et al., 2014; Nuzzolo and Comi, 2014; Nuzzolo at al., 2016), these trucks’ special technical features and the difficulty that their maneuvering presents in urban conditions suggest that a negative outcome on road safety should be expected. From the demographic point of view, two control variables have been considered. First, following earlier studies on urban road traffic accidents, a variable for Population Density has been included. According to what has gone before, in principle a negative sign should be expected for the influence that this has on urban road safety, in the sense that, the more concentrated the population (and, therefore, the less disperse it is throughout the metropolitan area), the more the number of fatalities caused by road 9 accidents should decrease. Previous studies agree that this relationship can be explained by greater traffic congestion in more densely populated cities incontrovertibly contributing to a significant reduction in driving speeds and, as is well known, this is one of the basic risk factors that leads to traffic accidents (Clark, 2003; Ewing et al., 2003, 2016; Graham and Glaister, 2003; Sukhai and Jones, 2014; Yannis et al., 2015; Yeo et al., 2015). On the other hand, greater population density could also be expected to entail more traffic accidents, as there is a rise in the degree of vehicle exposure to accidents. Also with respect to the population, a control variable is included for the mean age of the population in each province with the purpose of capturing the demonstrated greater vulnerability that certain collectives present. In this regard, according to certain authors such as Kanaan et al. (2009), Keall et al. (2004), Prat et al. (2015), Stevenson and Palamara (2001) and Williams et al. (1998), for example, the youngest drivers are a major risk collective that contribute by raising the likelihood and severity of accidents, especially in urban areas, as a result of drugs and alcohol DUI behaviors. However, the relationship between road safety outcomes and mean age is complex. In this regard, the number of fatalities as well as the number of accidents per mile driven (by car) has a “hammock” shape, with very high accident numbers per kilometer traveled for people below 25 and above 80. Despite this, motorist accidents are rare below driving age. For pedestrian crashes, it seems that the “hammock” is shifted toward younger ages, and continuously worsens for all age groups above 30, and even more so when a person reaches 80 or 85. The same is true for cyclists; accident rates rise sharply after the age of 80, although the trend starts at about 70. Regarding the special characteristics of each of the provinces, first consideration has been given to each provincial capital’s geographic location and atmospheric conditions 16 consequent brake that they put on the speed of inner city traffic may result in a lower road mortality rate. With respect to cities’ geographical locations, it can be observed that the variable for Latitude is positive and significant in the estimates for accidents, while it is also positive and significant for fatalities although in this latter case this result only holds for the larger capitals subsample in the fatalities regressions. This result is in line with previous studies (Golob and Recker, 2003; Ivan et al., 2015; Jaroszweski and McNamara, 2014) and might indicate that urban traffic accidents are more frequent in cities in the north of Spain, where the climate is more adverse. However, in this context it is important to note that although bad weather can be predicted to lead to worse road safety, this does not mean that there are greater numbers of traffic accidents and fatalities in all northern areas in general. In fact, international experience shows that Northern European countries such as Sweden, Finland, the United Kingdom, Norway, etc. have been implementing optimal strategies and actions for many decades, as a result of which they are now considered world leaders in road safety, and models to imitate (see CastilloManzano et al., 2014). With respect to the Longitude variable, a positive and significant coefficient has been obtained in the larger provincial capitals subsample regressions (both for fatalities and accidents), which indicates that there are more accidents and fatalities on urban roads in provinces in the east of the country. Geographically-speaking, the provinces in question coincide with the main tourist areas in Spain and, in some cases, in all Europe (specifically, in the case of the Mediterranean coastal provinces of Barcelona, Alicante and Valencia, and Majorca in the Balearic Islands); these are provinces with intense visitor traffic and a great deal of leisure and night life. Earlier literature has already shown that highly concentrated tourism can be associated with an increase in traffic 17 accidents in these particular areas of Spain, although only some specific areas have been evaluated, such as the Balearics (see e.g., Rosselló and Saenz-de-Miera, 2011; Saenzde-Miera and Rosselló, 2012). As a consequence of this finding, testing for a possible correlation between tourist activity and urban accidents in all Spanish NUTS-3 provinces is a line of investigation that remains to be addressed in future research. Results for the main explanatory variables that can be more related to levels of urbanization or urban development (population density, hospital density, Smart City status, subway and/or urban rail availability) suggest that the concentration of population and services in large cities can lead to greater road safety in net terms. To be specific, in line with earlier studies such as Castillo-Manzano et al. (2013) and Clark and Cushing (2004), the sign of the hospital density per kilometer variable is negative, but it is only statistically significant for explaining urban road mortality in general for the entire sample. This suggests that when there are more hospitals available, there is a reduction in accident severity (due to a fall in the time for postaccident medical treatment to be administered, as A&E services are in closer proximity). However, as might be expected, the relationship is not as strong with the likelihood that an accident will occur. In the case of the larger provincial cities (provincial capitals) subsample, no statistical significance is detected because of limited variability among the provinces. With respect to the population density variable, it can be observed in Tables 4 and 5 that a positive sign and statistical significance are obtained in accident regressions, and a negative sign in the case of road mortality. The prior literature can be used to explain this: it is logical that more people are actively mobile in very dense urban areas and, 18 therefore, there is a greater risk of accidents; however, accidents are less severe due to greater inner city traffic congestion, which limits driving speeds (see, for example, Clark and Cushing, 2004; Ewing et al., 2003, 2016; Graham and Glaister, 2003; Sukhai and Jones, 2014; Yannis et al., 2015). In other regards, the variable representing availability of a well-developed public transit system in the form of a subway and/or urban light rail system seems to reduce the number of fatalities in traffic accidents, although statistical significance is only obtained at 10% in one of the regressions for the larger provincial capitals subsample. The number of accidents also seems to be lower for larger cities with a more developed public transit system, although the variable is not statistically significant. These results corroborate what is suggested by other authors, such as Kersys (2015), Redman et al. (2013) and Yannis et al. (2015). The variable that captures Smart City status can be observed to maintain a negative and generally significant correlation in the fatalities regression, while it is not significant in the accidents regression. This novel result can be explained by the fact that the development of a technology-based smart urban transit system can make a clear contribution to reducing the most serious consequences of urban traffic accidents, as has also been concluded by recent research by Agarwarl et al. (2015), Guayante et al. (2014); Krishnan and Balasubramanian (2016), Medvedev et al. (2015) and Zhuhadar et al. (2017). In other respects, the lack of significance of the reduction in the number of accidents can be explained by the fact that the concept of Smart City is, broadlyspeaking, still in the embryonic stage in Spain. In short, this article’s research results for urban road safety in Spanish provinces are in the same line as earlier studies that find that higher levels of urban development and 19 greater concentrations of activities and population result in a lower urban road traffic accident rate. However, although the findings show that integrating road safety into a smart transportation system developed within the Smart City framework enables gains to be made in urban road mortality, it does not justify abandoning more traditional accident prevention strategies, such as those that are educationor traffic supervision or control-based (see Leden et al., 2014). 4. Concluding remarks. The inexorable growth of urban mobility in developed countries in recent decades, underpinned by the intensive use of private motorized vehicles, has resulted in increased numbers of privately owned vehicles and their excessive usage for personal journeys. This has given rise to a variety of negative externalities that compromise their environmental, social and energy sustainability, with road accidents standing out. In this context, this paper applies econometrics to determine the factors that explain differences in urban road safety in Spanish NUTS-3 provinces. For this, unprecedented panel data were compiled for the 2003-2013 period and a differently specified negative binomial model applied for the endogenous variables of traffic accidents and fatalities in urban areas. Novel results have been obtained that cover a gap in the literature on this issue, mainly on the level of Spain. In particular, the findings reveal a positive correlation between the level of urban development (approximated by variables for economics, industrial activity, healthcare, more advanced urban transit systems, and smart network implementation to connect the city’s functional subsystems) and improvements to urban road safety. More specifically, it seems that in the case of cities in Spanish provinces, a wider urban spread inevitably leads to more severe traffic accidents, whereas road 20 mortality is lower in urban areas with denser populations (thus more densely concentrated). In view of the obtained estimations, a negative correlation is found between population density and road mortality. Consequently, it can be concluded that urban dispersion seems to be a risk factor for particularly severe urban traffic accidents. As journey volume and distance (and, therefore, driving speeds) are unavoidably higher in a more widespread urban environment, urban dispersion seems to be inversely related to road mortality, which makes its management and planning a question of public health. For all the above reasons, it can be stated that population can be considered to be a sufficiently significant predictor of urban road mortality in the Spanish provinces. Following Litman (2015), publicly subsidized transport policies in large cities, with more sustainable means of transport can, on occasion, unintentionally give rise to a perverse effect in the form of excessive city expansion toward outlying metropolitan areas. The results obtained in this article indicate that, from the perspective of road safety, it might be more sociably desirable to give incentives to promote urban concentration rather than the suburban model that has predominated in recent decades, with the urban development of dormitory towns attached to larger Spanish cities. At the same time, following Alper et al. (2015), the firm belief exists that sub-central level road safety management can also be an efficient means of creating a safer context for all urban traffic users, and for the most vulnerable users, especially. 21 Acknowledgement The authors would like to express their gratitude to the Directorate General of Traffic (DGT) for the human and economic resources that it made available for this study, which was conducted as part of grant funded research project ref: SPIP2015-1702. References Abdel-Aty, M. A., Radwan, A.E. (2000). Modeling traffic accident occurrence and involvement. Accident Analysis and Prevention, 32, 633-642. Agarwal, P. K., Gurjar, J., Agarwal, A. K., Birla, R. (2015). Application of artificial intelligence for development of intelligent transport system in smart cities International Journal of Transportation Engineering and Traffic System, 1, 20-30. Albalate, D., Bel, G. (2012). Motorways, tolls, and road safety: Evidence from Europe’. SERIEs: Journal of the Spanish Economic Association 3, 457-473. Albalate, D., Fernández, L., Yarygina, A. (2013). The road against fatalities: Infrastructure spending vs. regulation? Accident Analysis and Prevention, 59, 227239. Alper, D., Sinuany-Stern, Z., Shinar, D. (2015). Evaluating the efficiency of local municipalities in providing traffic safety using the Data Envelopment Analysis. Accident Analysis and Prevention, 78, 39-50. Archer, J., Vogel, K. (2000). The Traffic Safety Problem in Urban Areas. Research Report CTR2000-04, Royal Institute of Technology, Stockholm, Sweden. Beirão, G., Cabral, J. S. (2007). Understanding attitudes towards public transport and private car: A qualitative study. Transport Policy, 14, 478-489. Bishai, D., Quresh, A., James, P., Ghaffar, A. (2006). National road casualties and economic development. Health Economics, 15, 65-81. 22 Buck, N. T., While, A. (2015). Competitive urbanism and the limits to smart city innovation: The UK Future Cities initiative. Urban Studies, 1-19, doi: 0042098015597162. Castillo-Manzano, J.I., Castro-Nuño, M., Fageda, X. (2013). Can health public expenditure reduce the tragic consequences of road traffic accidents? The EU-27 experience. European Journal of Health Economics, 15, 645-652. Castillo-Manzano, J.I., Castro-Nuño, M., Fageda, X. (2014). Could being in the European Union save lives? An econometric analysis of the Common Road Safety Policy for the EU-27. Journal of European Public Policy 21, 211-229. Castillo-Manzano, J. I., Castro-Nuño, M., Fageda, X. (2015). Can cars and trucks coexist peacefully on highways? Analyzing the effectiveness of road safety policies in Europe. Accident Analysis and Prevention, 77, 120-126. Castillo-Manzano, J. I., Castro-Nuño, M., Fageda, X. (2016). Exploring the relationship between truckload capacity and traffic accidents in the European Union. Transportation Research Part E: Logistics and Transportation Review, 88, 94-109. Chang, Y.S., Lee, W.J., Lee, J.H. (2016). Are there higher pedestrian fatalities in larger cities?: A scaling analysis of 115 to 161 largest cities in the United States. Traffic Injury Prevention, 17, 720-728. Chin, H. C., Quddus, M. A. (2003). Applying the random effect negative binomial model to examine traffic accident occurrence at signalized intersections. Accident Analysis and Prevention, 35, 253-259. Clark, D. E. (2003). Effect of population density on mortality after motor vehicle collisions. Accident Analysis and Prevention, 35, 965-971. Clark, D. E., Cushing, B. M. (2004). Rural and urban traffic fatalities, vehicle miles, and population density. Accident Analysis and Prevention, 36, 967-972. 23 Coruh, E., Bilgic, A., Tortum, A. (2015). Accident analysis with the random parameters negative binomial panel count data model. Analytic Methods in Accident Research, 7, 37-49. Constantinou, E., Panayiotou, G., Konstantinou, N., Loutsiou-Ladd, A., Kapardis, A. (2011). Risky and aggressive driving in young adults: Personality matters. Accident Analysis and Prevention, 43, 1323-1331. Dablanc, L., Giuliano, G., Holliday, K., O'Brien, T. (2013). Best practices in urban freight management: Lessons from an international survey. Transportation Research Record: Journal of the Transportation Research Board, 2379, 29-38. Dee, T. S. (1999). State alcohol policies, teen drinking and traffic fatalities. Journal of Public Economics, 72, 289-315. de Oña, J., Mujalli, R. O., Calvo, F. J. (2011). Analysis of traffic accident injury severity on Spanish rural highways using Bayesian networks. Accident Analysis and Prevention, 43(1), 402-411. de Oña, J., López, G., Mujalli, R., Calvo, F. J. (2013). Analysis of traffic accidents on rural highways using Latent Class Clustering and Bayesian Networks. Accident Analysis and Prevention, 51, 1-10. Dirección General de Tráfico (DGT) (2015). Anuario estadístico de accidentes. Ministerio del Interior. Madrid. Spain. Available at: http://www.dgt.es/es/seguridadvial/estadisticas-e-indicadores/publicaciones/anuario-estadistico-accidentes/. Dong, C., Clarke, D. B., Richards, S. H., Huang, B. (2014). Differences in passenger car and large truck involved crash frequencies at urban signalized intersections: An exploratory analysis. Accident Analysis and Prevention, 62, 87-94. 24 Dumbaugh, E., Rae, R. (2009). Safe urban form: Revisiting the relationship between community design and traffic safety. Journal of the American Planning Association, 75, 309-329. Dura-Guimera, A. (2003). Population deconcentration and social restructuring in Barcelona, a European Mediterranean city. Cities, 20, 387-394. Ewing, R., Schieber, R. A., Zegeer, C. V. (2003). Urban sprawl as a risk factor in motor vehicle occupant and pedestrian fatalities. American Journal of Public Health, 93, 1541-1545. Ewing, R., Hamidi, S., Grace, J. B. (2016). Urban sprawl as a risk factor in motor vehicle crashes. Urban Studies, 53, 247-266. Ferreira, S., Couto, A. (2013). Traffic flow-accidents relationship for urban intersections on the basis of the translog function. Safety Science, 60, 115-122. García-Altes, A., Pérez, K. (2007). The economic cost of road traffic crashes in an urban setting. Injury Prevention, 13, 65-68. Golob, T. F., Recker, W. W. (2003). Relationships among urban freeway accidents, traffic flow, weather, and lighting conditions. Journal of Transportation Engineering, 129, 342-353. Gomes, S. V. (2013). The influence of the infrastructure characteristics in urban road accidents occurrence. Accident Analysis and Prevention, 60, 289-297. Gómez-Barroso, D., López-Cuadrado, T., Llácer, A., Suárez, R. P., Fernández-Cuenca, R. (2015). Análisis espacial de los accidentes de tráfico con víctimas mortales en carretera en España, 2008-2011. Gaceta Sanitaria, 29, 24-29. Graham, D. J., Glaister, S. (2003). Spatial variation in road pedestrian casualties: the role of urban scale, density and land-use mix. Urban Studies, 40, 1591-1607. 25 Guayante, F., Díaz-Ramírez, A., Mejía-Alvarez, P. (2014). Detection of Vulnerable Road Users in Smart Cities. In 2014 Eighth International Conference on Next Generation Mobile Apps, Services and Technologies (pp. 307-312). IEEE. Hamann, C., Peek-Asa, C. (2013). On-road bicycle facilities and bicycle crashes in Iowa, 2007–2010. Accident Analysis and Prevention, 56, 103-109. Ivan, K., Haidu, I., Benedek, J., Ciobanu, S. M. (2015). Identification of traffic accident risk-prone areas under low-light conditions. Natural Hazards and Earth System Sciences, 15, 2059-2068. Jaroszweski, D., McNamara, T. (2014). The influence of rainfall on road accidents in urban areas: A weather radar approach. Travel Behaviour and Society, 1, 15-21. Jevtić, V., Vujanić, M., Lipovac, К., Jovanović, D., Pešić, D. (2015). The relationship between the travelling speed and motorcycle styles in urban settings: A case study in Belgrade. Accident Analysis and Prevention, 75, 77-85. Johansson, P. (1996). Speed limitation and motorway casualties: a time series count data approach. Accident Analysis and Prevention, 28, 73-87 Julsrud, T. E., Denstadli, J. M. (2017). Smartphones, travel time-use and attitudes to public transport services. Insights from an explorative study of urban dwellers in two Norwegian cities. International Journal of Sustainable Transportation, 11, 602-610. Kanaan, A., Huertas, P., Santiago, A., Sánchez, J. A., Martínez, P. (2009). Incidence of different health factors and their influence on traffic accidents in the province of Madrid, Spain. Legal Medicine, 11, S333-S336. Karlaftis, M.A, Tarko, A.P. (1998). Heterogeneity considerations in accident modeling. Accident Analysis and Prevention, 30, 425–433. 32 Equality DSmart_city Dummy variable (fictitious) that takes a value of 1 if the cities in each Spanish NUTS-3 region have Smart City status and 0 otherwise Spanish Smart City Network (RECI) Dsubway_rail Dummy variable (fictitious) that takes a value of 1 if the cities in each Spanish NUTS-3 region possess a subway and/or urban light rail system and 0 otherwise Spanish Observatory of Metropolitan Mobility Importance of manufacturing Importance for employment of industrial sector manufacturing activity over total provincial employment in thousands of persons Spanish Regional Accounts (National Statistical Institute, INE) Importance of construction Importance for employment of construction sector over total provincial employment in thousands of persons Spanish Regional Accounts (National Statistical Institute, INE) Motorization Number of private cars per 1000 inhabitants in NUTS-3 provinces Directorate General of Traffic (DGT) TABLE 2. Variables descriptive statistics Variables Mean Standard Deviation Minimum Maximum Entire sample Provincial Capitals >200,000 inhabitants Entire sample Provincial Capitals >200,000 inhabitants Entire sample Provincial Capitals >200,000 inhabitants Entire sample Provincial Capitals >200,000 inhabitants Accidents 986.57 2156.12 2412.39 3712.61 5 25 14840 14480 Fatalities 12.72 22.82 18.87 27.81 0 0 149 149 GDP per capita 20.80 21.48 4.50 5.02 12.16 12.94 38.10 38.10 Age 42.15 40.99 2.77 2.21 36.29 36.29 49.44 46.9 Longitude 4.88 6.87 6.10 9.42 0.21 0.22 41.39 41.39 Latitude 39.91 38.66 3.14 4.24 28.27 28.27 43.75 43.21 Population density 134.41 264.58 175.00 223.51 8.9 51.1 806.4 806.4 Hospital density 6.21e-06 4.97e-06 2.67e-06 1.37e-06 2.45e06 2.45e-06 0.000018 8.18e-06 DSmart_city 0.16 0.20 0.36 0.40 0 0 1 1 Dsubway_rail 0.18 0.5 0.38 0.50 0 0 1 1 Importance of manufacturing 0.24 0.23 0.08 0.10 0.09 0.09 0.51 0.51 Importance of construction 0.17 0.16 0.05 0.05 0.07 0.08 0.3 0.3 Motorization 666.12 647.23 70.47 67.98 492 503 957 864 33 TABLE 3a. Correlations matrix (Entire sample) Variables Fatalities Accidents GDP Age Longitude Latitude Pop_dens Hospitals Smart Subway/Rail Manufacturing Construction Motorization Fatalities 1 Accidents 0.86 1 GDP per capita 0.23 0.30 1 Age -0.26 -0.20 0.06 1 Longitude -0.01 -0.06 -0.09 0.03 1 Latitude -0.001 0.02 0.42 0.60 -0.23 1 Population density 0.66 0.72 0.33 -0.31 0.05 -0.27 1 Hospital density -0.35 -0.31 0.006 0.34 -0.13 0.19 -0.43 1 DSmart_city -0.06 0.07 0.006 0.12 0.01 0.04 0.02 -0.01 1 Dsubway_rail 0.47 0.52 0.25 -0.22 0.001 -0.16 -0.31 -0.31 0.12 1 Importance manufacturing -0.001 0.009 0.55 0.23 -0.08 0.62 -0.08 0.14 0.001 -0.04 1 Importance construction 0.05 -0.09 -0.27 -0.18 -0.02 -0.16 -0.13 -0.03 -0.51 -0.12 -0.32 1 Motorization -0.09 -0.06 0.23 0.14 -0.21 -0.03 -0.08 0.19 0.05 -0.02 -0.23 -0.03 1 34 TABLE 3b. Correlations matrix (Provincial Capitals >200,000 inhabitants) Variables Fatalities Accidents GDP Age Longitude Latitude Pop_dens Hospitals Smart Subway/Rail Manufacturing Construction Motorization Fatalities 1 Accidents 0.86 1 GDP per capita 0.29 0.40 1 Age -0.15 -0.08 0.29 1 Longitude -0.14 -0.18 0.02 0.10 1 Latitude 0.17 0.20 0.46 0.66 -0.18 1 Population density 0.60 0.70 0.37 -0.21 -0.12 -0.21 1 Hospital density -0.49 -0.51 -0.14 0.17 0.09 0.17 -0.6 1 DSmart_city -0.11 0.06 -0.02 0.24 -0.02 0.08 -0.0 1 Dsubway_rail 0.32 0.42 0.35 -0.06 -0.20 0.05 0.40 -0.39 0.12 1 Importance manufacturing 0.06 0.08 0.55 -0.17 0.04 0.69 -0.20 0.20 0.01 0.05 1 Importance construction 0.05 -0.12 -0.37 0.13 -0.07 -0.19 -0.09 -0.14 0.59 -0.19 -0.09 1 Motorization 0.03 0.06 0.02 -0.39 -0.21 -0.40 0.18 -0.01 0.03 0.23 -0.58 0.13 1 35 TABLE 4. Estimation results (panel data model, mean population with negative binomial distribution) Endogenous variable: Number of urban traffic accident fatalities Independent Variables (1) Entire sample with GDP per capita squared as explanatory variable (2) Entire sample without GDP per capita squared as explanatory variable (3) Provincial Capitals >200,000 inhabitants with GDP per capita squared as explanatory variable (4) Provincial Capitals >200,000 inhabitants without GDP per capita squared as explanatory variable GDP per capita 0.21 (0.06)*** 0.03 (0.01)*** 0.23 (0.11)** -0.005 (0.02) GDP per capita2 -0.003 (0.0012)*** - -0.004 (0.002)** - Age 0.03 (0.001)** 0.03 (0.001)** -0.09 (0.02)*** -0.08 (0.02)*** Longitude 0.001 (0.003) 0.04 (0.004) 0.009 (0.004)** 0.01 (0.004)*** Latitude 0.01 (0.02) 0.01 (0.02) 0.05 (0.01)*** 0.07 (0.01)*** Population density -0.0005 (0.0002)** -0.0004 (0.0002)* 0.0004 (0.0003) 0.0007 (0.0004)* Hospital density -49311.49 (12901.44)*** -49175.93 (13383.51)*** 44912.9 (38109.73) 38499.83 (43882.75) DSmart_city -0.24 (0.15)* -0.24 (0.14)* -0.61 (0.26)*** -0.69 (0.27)*** Dsubway_rail -0.07 (0.10) -0.09 (0.10) -0.21 (0.14) -0.22 (0.13)* Importance manufacturing -0.84 (0.52)* -0.63 (0.59) 1.23 (1.03) 1.44 (1.31) Importance construction -0.34 (1.08) -0.21 (0.14) 2.99 (2.27) 2.69 (2.18) Motorization 0.0008 (0.0005) 0.001 (0.0005)*** 0.0006 (0.0011) 0.0017 (0.001) Intercept -16.09 (0.93)*** -14.75 (0.78)*** -13.03 (1.33)*** -12.17 (1.38)*** Year fixed effects YES YES YES YES Wald test (joint sign.) 425.05*** 448.55*** - - BreuschPagan/CookWeisberg test for heterogeneity (Ho: Constant variance) 1191.10*** 1165.16*** 128.25*** 129.18*** Wooldridge test – autocorrelation (Ho: No first order autocorrelation) 8.75*** 8.76*** 6.90*** 6.91*** ADF test – nonstationarity (Ho: nonstationarity) -0.60*** -0.60*** -0.53*** -0.53*** Doornik-Hansen test for multivariate normality 12237.21*** 10903.34*** 1878.44*** 1608.89*** No. Observations 600 600 216 216 No. Provinces 50 50 18 18 Note 1: standard errors in brackets, robust to heteroscedasticity and grouped by province. Regressions specify a withingroup AR(1) correlation structure for panels. Population used as exposure variable. Note 2: Statistical significance at 1% (***), 5% (**), 10% (*), respectively. STATA does not give the Wald test in the regressions for the subsample of most highly populated provincial capitals. 36 TABLE 5. Estimation results (panel data model, mean population with negative binomial distribution) Endogenous variable: Number of urban traffic accidents Independent Variables (1) Entire sample with GDP per capita squared as explanatory variable (2) Entire sample without GDP per capita squared as explanatory variable (3) Provincial Capitals >200,000 inhabitants with GDP per capita squared as explanatory variable (4) Provincial Capitals >200,000 inhabitants without GDP per capita squared as explanatory variable GDP per capita -0.018 (0.05) -0.01 (0.02) -0.12 (0.06)* -0.001 (0.02) GDP per capita2 0.00016 (0.0009) - 0.002 (0.001)** - Age -0.007 (0.03)** -0.07 (0.03)** -0.16 (0.09)* -0.15 (0.09) Longitude 0.005 (0.007) 0.005 (0.006) 0.01 (0.004)*** 0.01 (0.003)*** Latitude 0.06 (0.03)* 0.06 (0.03)* 0.12 (0.04)*** 0.11 (0.03)*** Population density 0.0013 (0.0004)*** 0.001 (0.0004)*** 0.001 (0.0004)*** 0.001 (0.0004)*** Hospital density -15107 (12638.63) -15248 (12605.72) 29066.86 (35490.57) 27829.07 (35362.14) DSmart_city 0.09 (0.06) 0.09 (0.06) 0.15 (0.09) 0.14 (0.09) Dsubway_rail 0.02 (0.01) 0.02 (0.1) -0.14 (0.13) -0.13 (0.15) Importance manufacturing 1.83 (1.10)* 1.80 (1.06)* 3.56 (1.50)*** 2.89 (1.33)** Importance construction -0.19 (1.54) -0.20 (1.53) 1.45 (2.74) 1.02 (2.75) Motorization 0.0003 (0.0009) 0.0003 (0.0009) 0.004 (0.001)*** 0.003 (0.001)*** Intercept -7.07 (1.47)*** -7.11 (1.45)*** -8.06 (3.99)** -8.34 (3.92)** Year fixed effects YES YES YES YES Wald test (joint sign.) 198.25*** 189.73*** - - Breusch-Pagan/CookWeisberg test for heterogeneity (Ho: Constant variance) 1369.67*** 1359.94*** 60.69*** 66.13*** Wooldridge test – autocorrelation (Ho: No first order autocorrelation) 73.88*** 79.43*** 41.62*** 45.03*** ADF test – nonstationarity (Ho: nonstationarity) -0.55** -0.55** -0.52** -0.52** Doornik-Hansen test for multivariate normality 12861.74*** 11553.73*** 1808.81*** 1549.32*** No. Observations 600 600 216 216 No. Provinces 50 50 18 18 Note 1: standard errors in brackets, robust to heteroscedasticity and grouped by province. Regressions specify a withingroup AR(1) correlation structure for panels. Population used as exposure variable. Note 2: Statistical significance at 1% (***), 5% (**), 10% (*), respectively. STATA does not give the Wald test in the regressions for the subsample of most highly populated provincial capitals. 37 FIGURES Figure 1. Number of traffic accidents for Spanish provinces Source: Authors. Figure 2. Lethality/severity index for Spanish provinces (no. fatalities/no. accidents) Source: Authors.