Centralization versus decentralization of traffic law enforcement governance in Spain

Full Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journalInformation?journalCode=cres20 Regional Studies ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/cres20 Centralization versus decentralization of traffic law enforcement governance in Spain Jose I. Castillo-Manzano, Mercedes Castro-Nuño, Lourdes Lopez-Valpuesta & Jesús Boby To cite this article: Jose I. Castillo-Manzano, Mercedes Castro-Nuño, Lourdes Lopez- Valpuesta & Jesús Boby (2022) Centralization versus decentralization of traffic law enforcement governance in Spain, Regional Studies, 56:11, 1976-1988, DOI: 10.1080/00343404.2021.2020743 To link to this article: https://doi.org/10.1080/00343404.2021.2020743 View supplementary material Published online: 16 Feb 2022. Submit your article to this journal Article views: 157 View related articles View Crossmark data Centralization versus decentralization of traffic law enforcement governance in Spain Jose I. Castillo-Manzano a , Mercedes Castro-Nuño b , Lourdes Lopez-Valpuesta c and Jesús Boby d ABSTRACT The current paper deals with the debate over centralization versus decentralization in road safety management, and specifically in traffic law enforcement. We address the case study of Spain, where a single traffic law acquis is enforced by two systems: road policing decentralized for some regions and national road policing for all others. Our econometric findings reveal that the Spanish regions with centralized traffic policing present better safety figures, which might be useful for reinforcing the coordinated performance of road safety policy in other traditionally decentralized countries. Furthermore, our evidence indicates that cautious analyses have to be carried out on a caseby-case basis. KEYWORDS road safety; law enforcement; traffic police management; centralization; decentralization; multilevel governance JEL K42, L91, R41 HISTORY Received 15 July 2020; in revised form 1 December 2021 INTRODUCTION Decentralization and the multilevel management of public administration is a growing trend in most developed countries (Koprićet al., 2017; Steiner et al., 2018). According to the World Bank (1999), decentralization can be defined as the delegation of political, fiscal and administrative powers to subnational levels. Scientific research into decentralization was popularized in the second half of the 20th century with Oates’(1972)Decentralization Theorem, which outlined a framework to identify the appropriate sharing of national, regional and local public services. As a consequence, vast numbers of scholars have addressed the impact of decentralizing public services, exploring approaches, models and effectiveness. Nevertheless, the debate is still open because there are outcomes supporting both centralization and decentralization depending on the specific policy area (Kuhlmann & Wayenberg, 2016) as each offers potential advantages and disadvantages. First, decentralization can be justified by the Theory of Fiscal Federalism, which assumes that decentralized government is more responsive to citizen preferences and, therefore, more effective at solving public problems (Oates, 1972). In fact, a great deal of evidence shows that decentralization enables a more efficient service to be provided to citizens, as well as increased public participation and accountability (Kuhlmann & Wayenberg, 2016) in both developing and developed countries. In this context, it has even been demonstrated that decentralization may lead to a reduction in inequalities between regions (Ezcurra & Rodríguez-Pose, 2013; Kyriacou et al., 2017). Effectiveness is assumed to increase due to regional and local governments’ability to better adapt to specific local needs (Balaguer-Coll et al., 2010) and respond more quickly to demands on account of their smaller size. However, there is also empirical evidence to the contrary: for certain authors, decentralization may result in the provision of poorer services due to access to personnel © 2022 Regional Studies Association CONTACT a (Corresponding author) [email protected] Applied Economics & Management Research Group. Universidad de Sevilla, Seville, Spain. b [email protected] Applied Economics & Management Research Group. Universidad de Sevilla, Seville, Spain. c [email protected] Applied Economics & Management Research Group. Universidad de Sevilla, Seville, Spain. d [email protected] Universidad de Sevilla, Seville, Spain. Supplemental data for this article can be accessed at https://doi.org/10.1080/00343404.2021.2020743 REGIONAL STUDIES 2022, VOL. 56, NO. 11, 1976–1988 https://doi.org/10.1080/00343404.2021.2020743 with lower competency levels and reduced expertise, fewer monitoring and control artifacts, and the more complex coordination and implementation of reforms (Segal, 1997; Treisman, 2002), which, therefore, reduces the effectiveness and quality of the public services offered. In this regard, a centralized administration could create economies of scale in which fixed costs are not repeated, whereas multiple fixed costs can exist when administration is decentralized (Steiner et al., 2018). Furthermore, competition between regional and local governments can lead to other risks such as corruption or control by local political elites (Cai & Treisman, 2004), especially in developing countries, where it can even result in slower economic development (Davoodi & Zou, 1998). The debate on the most appropriate level of distribution of public administration competencies is also ongoing in the area of social care services, with the specific governance level chosen for each type of public service according to the characteristics of the specific country or region (Aslim & Neyapti, 2017). Some examples of public services that are frequently decentralized are education (Mwinjuma et al., 2015), healthcare (Jiménez-Rubio & García-Gómez, 2017), or justice and security (McCluskey et al., 2014). Police system services may also be devolved to different tiers of government to achieve the best outcomes (Lowatcharin & Stallmann, 2019) and this is the case of the current paper’s specific object of study: road traffic policing. Traffic law enforcement management is decentralized in several European countries such as Belgium, Germany, the UK, Sweden and the Netherlands, with the responsibility to implement targets and strategies shared between the central and regional authorities (SafetyNet, 2009). In general terms, it has been empirically demonstrated that law enforcement is one of the most effective strategies for changing driver behaviour (Adler et al., 2012), especially regarding excessive speeding (Aarts & Van Schagen, 2006), non-use of seat belts (Acosta-Rodriguez et al., 2020), drugs driving (Penning et al., 2010) and drink driving (Stringer, 2019). Together with engineering and education, police enforcement has been demonstrated to be a traditional approach to achieving reductions in the prevalence and incidence of traffic accidents (Stanojević et al., 2013) by encouraging road users to comply with basic traffic rules and regulations and deterring traffic violations (Feng et al., 2020), specifically on interurban roads (Delavary Foroutaghe et al., 2020), and for Spain (Castillo-Manzano et al., 2019), both in the case study and on the kind of road analysed in the current paper. This broad treatment of the effectiveness of traffic police enforcement on road safety outcomes fully justifies the appropriateness of our analysis as we study a novel topic based on the safety impact of the decentralization of traffic law enforcement governance on interurban roads in a case study of Spain, a country which is an outstanding illustration of decentralization in general (Moreno, 2002). Based on the previous literature that addresses the decentralization of public policies (Kuhlmann & Wayenberg, 2016), it can be argued that the level of decentralization of the traffic police in Spain could have either a positive or negative influence on road traffic safety. As stated above, the smaller size of a regional police force and its greater proximity to citizens, which allows it to better adapt to the area’s specific culture, related behaviours and geography, should enable it to be more agile (Lowatcharin & Stallmann, 2019). However, the national approach should give the police access to greater assets and resources and a higher level of expertise and enable them to promote the use of management best practices (in the sense of the so-called ‘economies of scale’), among other things (McCluskey et al., 2014). Unfortunately, there are still no world-generalizable studies in the literature on comparative levels of traffic policing. The purpose of our research is, therefore, to explore the impact of the coexistence of two tiers of traffic police governance on road traffic safety and to identify the sign of the impact (should one exist) for the case of Spain, using data for Eurostat statistical territorial classification NUTS-3 regions (considered to be the provinces in Spain) and a 16-year period (2003–18). This research paper’s analysis of the state of the art makes two new contributions to the scientific literature. First, it adds a new field of study to the public administration decentralization debate, specifically on traffic law enforcement, which, to our knowledge, has not been empirically assessed previously, and offers new practical evidence. Second, it provides the safety literature with a new analysis of the impact that the decentralization of traffic police enforcement has had in Spain in terms of traffic accident and fatality figures for interurban roads. In view of this background, the remainder of the paper is structured as follows. The next section explains the Spanish context for the decentralization process and road safety. The third section includes the methodology applied and the variables and panel data used for the analysis. The fourth section gives the results of the econometric analysis and a discussion, including a comparison with other scientific bibliography. Finally, the fifth section presents the main conclusions of the research. There is also the supplemental data online. TRAFFIC POLICE DECENTRALIZATION PROCESS AND ROAD SAFETY IN SPAIN Territorially, Spain is organized into 17 Eurostat NUTS-2 regions (according to Eurostat terminology) (i.e., a firstlevel political and administrative division per the Spanish Constitution of 1978), which is arranged around the framework of a ‘State of Autonomies’(Moreno, 2002), and two North African cities (Ceuta and Melilla), which have also been granted the status of self-governed territories. Each NUTS-2 region is, in turn, formed of one or several NUTS-3 regions (in Eurostat terminology, which are known as provinces in Spain), with a total of 50 in all. Administratively, Spain is not a federation but a decentralized unitary state (Colomer, 1998), with the Centralization versus decentralization of traffic law enforcement governance in Spain 1977 REGIONAL STUDIES widespread roll-out of devolution of governance to regional governments throughout Spain’s democratic era, since the 1978 Constitution came into force (Balaguer- Coll et al., 2010), until the current situation was reached in which approximately 80% of the public budget is directly managed by regional and local administrations (Spanish Ministry of Finance data). This decentralization process has not been homogeneous (Martí-Costa et al., 2017), that is, some regions have higher levels of decentralization than others in many services and depending on policy areas considered. For example, nowadays, while the process of transferring full competencies in social welfare issues such as healthcare (Costa-Font & Rico, 2006; Costa-Font & Turati, 2018) or education (Alegre, 2010; Salinas & Solé-Ollé, 2018) has been completed, conversely, traffic and road safety management has followed a different decentralization model and pace. In this regard, the 1978 Spanish Constitution stated that the state reserves for itself some exclusive legal and executive functions regarding traffic and the circulation of motor vehicles. However, based on some historical antecedents, some subsequent legislation expressly recognized that the Basque Country (1982) and Catalonia (1997) regions should exercise full executive competencies for the supervision and control of national traffic regulations on interurban roads in their respective territories. In other words, while the devolution process in areas such as education and health has been rolled out during a period of over two decades to the benefit of the regions and only recently concluded, with all the regions whose autonomy is recognized in the Constitution being given competencies in these areas, in the case of traffic management, which had received previous recognition, it has been reduced to the issues analysed in the present paper (supervision of compliance with the regulations in force) and then, only in some specific regions. As a consequence, traffic police management is centralized for most Spanish regions and self-governed in only a few. Three levels of decentralization exist (Table 1): .Centralized traffic police. This is the most common approach with road traffic controlled by the Agrupación de Tráfico de la Guardia Civil (ATGC –Civil Guard Traffic Group). .Fully decentralized policing. In the regions of Catalonia and the Basque Country, the policing of road traffic comes under the full authority of the regional governments, with road traffic controlled by the Mossos d’Esquadra and Ertzaintza, respectively. .Partially decentralized policing. In the region of Navarre, road traffic policing is partially decentralized, with the ATGC sharing responsibility with the regional police (Policía Foral, in Spanish). In Spain, local/municipal police departments are in charge of traffic enforcement on urban roads, while the national or regional traffic police are responsible for interurban roads. It is also important to highlight that the traffic legal acquis is the same for all the regions in the country. Road safety strategies typically included in the literature, such as alcohol consumption limits (Krüger, 2013), the implementation of a points-based driving licence system (Rodríguez-López et al., 2016), and the speed limits in force (Lloyd et al., 2015) are implemented in all the Spanish regions and so do not contribute to any differences in safety figures. Focusing on safety figures, recorded traffic accidents have improved significantly in Spain since the beginning of the 21st century, to the extent that a 70.9% reduction has been achieved in the period 2000–13) (Organisation for Economic Co-operation and Development (OECD), 2015). It can, therefore, be considered a country to learn from (Wegman, 2017), particularly in the European context (Castillo-Manzano et al., 2014a,2015). By way of example, Table 2 shows a comparison of traffic fatality figures in Spain and in the EU-28 in which the first and last years of the time series analysed in the present paper have been taken as references. As can be observed, although both the number and rate (per 100,000 inhabitants) of fatalities were high in Spain in 2003, 15 years later, in 2018, the figures had improved significantly. When the data for Spain are broken down by NUTS-3 region for the case of interurban roads, the same positive evolution in the figures can be seen in the case of both fatalities and accidents in NUTS-3 and with centralized and decentralized traffic enforcement (Figure 1; and see Table A1 in Appendix A in the supplemental data online). These data have turned Spain into an example of success in road safety policies in the EU framework (Castillo- Manzano et al., 2014c), including the implementation of points system-based driving licences (Gras et al., 2014), innovative mass-media public advertising campaigns (Nicolás Ojeda et al., 2015) and hardening punitive strategies by reforming the Penal Code (Castillo-Manzano et al., 2011). In this context of reforms aimed at improving road safety, we consider that the decentralization of traffic enforcement in some Spanish regions should be studied in depth in order to be able to assess its effects on road accidents and fatalities. EMPIRICAL FRAMEWORK: DATA, VARIABLES AND METHODOLOGY Original panel data were constructed ad hoc using available official data for a sample based on the 50 Spanish NUTS-3 regions (provinces). The study period is from 2003 to 2018. The beginning of this period corresponds to the coming into force of Royal Decree-Law No. 1428/2003, which established the current General Driving Regulation in Spain, while 2018 was the most recent year for which we have found a complete dataset for the regions and variables considered. This time period also allows an analysis of the influence of external macroeconomic, social 1978 Jose I. Castillo-Manzano et al. REGIONAL STUDIES and political changes that have impacted road traffic safety. Two models have been estimated using both traffic accidents and fatalities on interurban roads as the endogenous variables and a set of exogenous variables detailed in Table 3. Accidents and fatalities on urban roads (in municipalities) have been excluded from the analysis as, in Spain, road traffic safety on these roads comes under the responsibility of the local police (which, as stated above, is fully decentralized). The panel data have been econometrically treated with the STATA package using two econometric models that take NUTS-3 region (province) iduring period t: 1 ln(E[Yit]) = a + z ·Decentralizationit ·Yeart + b kXit +ntYeart+ d iNUTS3 −Regioni+1it, (1) where Y it are the endogenous road safety variables, that is, the number of road traffic accidents/road traffic fatalities on interurban roads (within a period of 30 days following the accident, in accordance with the Vienna Convention). As explained in the following section, total figures have been chosen instead of per capita variables, as the total number of drivers was used as the exposure variable (Usman et al., 2011). Decentralization it is a dummy variable used to capture the decentralization of traffic police enforcement governance for some Spanish regions. This dummy variable takes a value of 1 if traffic policing is fully/partially decentralized in the region, and 0 if traffic policing depends on the central authorities. We include the decentralization variable corrected by time trend to avoid collinearity with regional fixed effects. Note that road traffic enforcement decentralization is constant for the entire assessment period, which might represent a limitation for our analysis. This is an adaptation of difference-in-differences (DiD) techniques in which we have introduced the intervention effect over time. The variable Decentralization it ·Year t is called ‘Decentralized trend’in the corresponding tables. X it are control variables that measure: (1) Economic and healthcare issues, more specifically the unemployment rate and hospital density, respectively. The previous literature broadly demonstrates pro-cyclical behaviour between the economic situation and road traffic safety, so the unemployment rate is expected to show a negative sign with road traffic safety outcomes (as in Scuffham & Langley, 2002). Hospital density is included in the model as a proxy of healthcare system capacity. It is, therefore, estimated a priori that hospital density and traffic fatalities will show a negative correlation, following earlier studies such as Truong et al. (2016) and Castillo-Manzano et al. (2014b). (2) Demographic parameters related to the average age of the population and the ratio of capital’s population in the total regional population. The average age variable aims to model the difference in driving behaviours for different Table 1. Traffic police decentralization process in Spain. Spanish NUTS-2 regions Spanish NUTS-3 regions (provinces) Type of governance (year) Legislation Traffic police departments Basque Country Alava Fully decentralized (1982) Royal Decree 3256/1982 of 15 October Ertzaintza Gipuzkoa Vizcaya Catalonia Barcelona Fully decentralized (2000) Organic Law 6/1997 of 15 December Mossos d’Esquadra Girona Fully decentralized (1998) Lleida Fully decentralized (1999) Tarragona Fully decentralized (2000) Navarre Navarre Partially decentralized Organic Law on the Reintegration and Improvement of the Navarre Regional Regime, 1982 Navarre Police Force (Policía Foral) and Civil Guard Traffic Group Source: Authors. Table 2. People killed in road accidents (EU-28 versus Spain). Number Rate (number per 100,000 persons) 2003 2018 % 2003 2018 % EU-28 50,989 25,178 −50.62% 10.4 4.9 −52.88% Spain 5373 1806 −66.39% 12.7 3.9 −69.29% Source: Eurostat. Centralization versus decentralization of traffic law enforcement governance in Spain 1979 REGIONAL STUDIES ages. According to the existing scientific literature (Prat et al., 2015), the youngest drivers present more reckless driving behaviour with more frequent consumption of drugs and alcohol while driving, which may contribute by increasing the likelihood and severity of accidents. However, elderly drivers normally present deteriorated physical and cognitive capabilities that can lead to an increase in both accidents and mortality (Li et al., 2003). The ratio of the capital’s population in the total population is a proxy of population concentration. There is no consensus on the influence of this variable on road traffic safety: in some cases, higher population concentration is expected to lead to an increase in accidents (Castro-Nuño et al., 2018), which can be explained by an increase in travel from metropolitan areas to access essential services mostly located in the capital city; but better road safety has also been found in other cases (Tolón-Becerra et al., 2013)as a result of a reduction in commuting in regions with a higher population concentration. (3) Mobility and infrastructure parameters, specifically, the motorization rate and the density of highways on all roads. The motorization rate is used as a proxy of the exposure risk factor; thus, it is initially expected that a higher stock of vehicles and a greater use made of them may lead to higher accident and fatality rates (Noble et al., 2015). A negative relationship is also initially expected between the accident and fatalities and highway ratio, as better infrastructure is shown to improve road traffic safety (Aparicio-Izquierdo et al., 2013; Sánchez-González et al., 2018). Finally, the total number of tourists visiting the region is included as a proxy of exposure and risk behaviours. Tourism has previously been demonstrated to increase traffic congestion Figure 1. Fatality rate (a) and accident rate (b) per 100,000 persons on Spanish interurban roads in NUTS-3 regions in 2018. Sources: Authors from the Spanish General Traffic Directorate (DGT) and National Statistics Institute (INE). 1980 Jose I. Castillo-Manzano et al. REGIONAL STUDIES and to be associated with an increase in alcohol consumption and other imprudent behaviour, so it is initially expected to contribute to a deterioration in road traffic safety (Castillo-Manzano et al., 2020). Meteorological variables through the volume of rainfall. There is empirical evidence that indicates that weather conditions affect road Table 3. Variables used in the empirical analysis. Variable Description Source Accidents on interurban roads Traffic accidents on interurban roads in each of the Spanish NUTS-3 regions Spanish General Traffic Directorate Yearbook (https:// www.dgt.es/menusecundario/dgt-en-cifras/dgt-en- cifras-resultados/?tema=accidentes-de- trafico&categoria=informe-anual&pag=1&order= DESC) Fatalities on interurban roads Fatalities within 30 days of the accident on interurban roads in each of the Spanish NUTS- 3 regions Spanish General Traffic Directorate Yearbook (https:// www.dgt.es/menusecundario/dgt-en-cifras/dgt-en- cifras-resultados/?tema¼accidentes-de- trafico&categoria¼informeanual&pag¼1&order¼DESC) Decentralization Dummy variable equal to 1 if traffic policing is partially or fully decentralized in the region; 0 otherwise Spanish Ministry of Internal Affairs (http://www.interior. gob.es/en/portada) Unemployment rate Unemployment rate during the fourth quarter of the period in each of the Spanish NUTS-3 regions Spanish National Statistics Institute (INE) (https://www. ine.es/en/index.htm) Hospital density Number of hospitals per 1000 km 2 in each of the Spanish NUTS-3 regions National Hospitals Catalog –Spanish Ministry of Health, Consumer Affairs, and Social Welfare (https://www. mscbs.gob.es/en/ciudadanos/prestaciones/ centrosServiciosSNS/hospitales/home.htm) Average age Average age of the population living in each of the Spanish NUTS-3 regions as of 1 January INE (https://www.ine.es/en/index.htm) Ratio of population in capital city Ratio of the population living in the main city in the region with respect to the total population in each of the Spanish NUTS-3 regions INE (https://www.ine.es/en/index.htm) Motorization rate Vehicles per 1000 inhabitants in each of the Spanish NUTS-3 regions INE (https://www.ine.es/en/index.htm) Highways ratio Ratio between the total length of highways and the total length of all types of roads in each of the Spanish NUTS-3 regions Spanish Ministry of Transport, Mobility, and Urban Agenda (https://www.mitma.gob.es/carreteras/ catalogo-y-evolucion-de-la-red-de-carreteras) Tourists Total number of tourists who stayed at least one night in each of the Spanish NUTS-3 regions INE Precipitation Average monthly rainfall by region and year in each of the Spanish NUTS-3 regions Spanish State Weather Agency (http://www.aemet.es/ en/eltiempo/observacion/ultimosdatos? k¼esp&datos¼img&w¼2) Per capita GDP Gross domestic product per inhabitant in each of the Spanish NUTS-3 regions INE Population density Million inhabitants per km 2 in each of the Spanish NUTS-3 regions Eurostat (https://ec.europa.eu/eurostat/databrowser/ view/tps00003/default/table?lang=en) Driver census (exposure variable) Valid driving licences in each of the Spanish NUTS-3 regions Spanish General Traffic Directorate Yearbook (https:// www.dgt.es/menusecundario/dgt-en-cifras/dgt-en- cifras-resultados/?tema=conductores|accidentes-de- trafico&categoria=informe-anual&pag=1&order= DESC) Centralization versus decentralization of traffic law enforcement governance in Spain 1981 REGIONAL STUDIES traffic safety; specifically, we expect an unfavourable impact of rainfall on the road traffic safety variables (Usman et al., 2011). To capture evolution over time, first, a time trend is used. An evolution of the specification with time fixed effects is included in Table A3 in Appendix A in the supplemental data online as a robustness check. NUTS-3Region it are fixed effects per region according to the outcomes of the Hausman tests (Table 5). Regional fixed effects are included in the model to quantify all the specific characteristics of each NUTS-3 region not measured by the remaining variables. The inclusion of regional fixed effects is a common approach when working with panel data in road traffic research projects (Toffolutti & Suhrcke, 2014). Finally, ε it is the mean zero random error. Table 3 describes the variables considered and the sources from which they have been taken. Table 4 gives the units and the descriptive statistics. Given the nature of panel data, we have used a fixed effects Poisson model for our estimations. Pearson goodness-of-fit chi-squared tests have been calculated for the individual models for each NUTS-3 region (province). The null hypothesis of NUTS-3 region-level overdispersion in both endogenous variables is rejected in all cases, whereby the Poisson is a good fit in these cases. In addition, the standard deviation is lower than the mean at the NUTS-3 region level for both endogenous variables. The only possible over-dispersion comes from between- NUTS-3 region differences and has been controlled by using robust standard errors clustered by NUTS-3 region as in Wu et al. (2014). Results are not included here for the sake of simplicity but are available from the authors upon request. RESULTS AND DISCUSSION Table A2 in Appendix A in the supplemental data online shows the correlation matrix of the variables used in the analysis. As can be observed, there are no high correlations among the exogenous variables used, which implies that there are no multicollinearity problems. Table 5 gives the estimation results for the model proposed in the previous section. A normality test (Shapiro– Wilk) has been performed on the endogenous variables. The endogenous variables cannot be inferred to follow a Table 4. Variables: descriptive statistics. Variable Unit of measurement Observations Mean SD Minimum Maximum Accidents on interurban roads Annual number of road accidents with casualties 800 808.79 772.35 83.00 5622.00 Fatalities on interurban roads Annual number of people killed in a traffic accident 800 44.56 34.19 4.00 241.00 Decentralization –800 0.16 0.37 0.00 1.00 Unemployment rate Annual % of unemployed people per total labour force 800 16.49 7.96 2.87 42.34 Hospital density Annual number of hospitals per surface area (km 2 ) 800 0.059 0.264 0.000 2.186 Average age Years 800 42.61 2.85 36.29 50.49 Ratio of population in capital city Ratio 800 0.32 0.13 0.09 0.76 Motorization rate Vehicles per 1000 inhabitants 800 0.678 0.080 0.072 0.957 Highways ratio Ratio 800 0.07 0.08 0.004 1.192 Tourists Annual millions of persons 800 0.905 0.868 0.121 6.560 Precipitation mm/100 800 486.50 305.16 4.00 1867.90 Per capita GDP €, current prices 800 20,243.91 4898.90 10,111.00 37,675.00 Population density Million inhabitants/km 2 800 131.62 173.49 8.58 842.89 Driver census Annual number of drivers 800 506,112.40 589,328.10 50,172.00 3,520,477.00 Year 800 2010.50 4.61 2003.00 2018.00 1982 Jose I. Castillo-Manzano et al. REGIONAL STUDIES normal distribution, which also indicates that a count data distribution seems to be a good choice. Estimations can present problems of heteroscedasticity and temporal autocorrelation in the error term if they are not correctly treated. The modified Wald test for groupwise heteroscedasticity is given in Table 5 and demonstrates that model estimations without any corrections may be affected by this issue. The Levin– Lin–Chu test is also performed on the endogenous variables and supports the selection of a stationary model (Levin et al., 2002). In addition, the entire model is significant at 1% as shown by Wald chi-squared statistics. Finally, some modifications to the models included in Table 5 are included as robustness checks in Table A3 in Appendix A in the supplemental data online: .Models III and IV introduce time fixed effects into the original specification. .Models V and VI remove one NUTS-3 region (Navarre) where traffic law enforcement is partially decentralized to test whether this special situation leads to any bias in the results. .Models VII and VIII introduce new variables into model as proxies of exposure and activity to analyse the influence that they might have. .Models IX and X add differentiated slopes for each NUTS-3 region to test whether these individual trends should be added to the original model. .Models XI and XII introduce a general pooled ordinary least squares (OLS) specification. Results of multivariate correlations show that NUTS-3 regions where traffic law enforcement is decentralized (exercised by the regional traffic police department) record, ceteris paribus, higher numbers of interurban traffic accidents and fatalities compared with those where traffic enforcement governance is centralized under the national traffic police force. This finding is statistically significant at 1% in most cases and is found in both the original models and the robustness checks (with the only exceptions of models X and XI in Table A3 in Appendix A in the supplemental data online). This suggests that, in the Spanish case and during the analysed period, the ATGC (Civil Guard) has been more effective at preventing crashes than the regional police forces (Mossos d’Esquadra, Ertzaintza and Policía Foral) when applying the same traffic laws and considering the typical socio-economic, mobility and environmental variables that may impact on road traffic safety as control variables. This outcome is in line with the previous literature that has analysed the impact of decentralization on other public policies (Segal, 1997), and demonstrates, in this case, that a centralized administration could lead to positive results as it enables access to larger quantities of better resources and methods, minimizes competition between administrations, and reinforces monitoring and control practices as a consequence of the so-called ‘economies of scale’ that can also be found in the centralization of other public policy areas (Karjalainen, 2011). This would also all be reinforced by the Spanish interurban road network’s characteristic ‘network economies’as analysed in the current paper, because it serves all road users who circulate in Spanish territory (both transit from one region to another and continuous transit generated by tourist movements or the so-called Paso del Estrecho 2 during the summer months), irrespective of the point of origin. There is no doubt that this enhances the role of the centralized Table 5. Estimation results (panel data model, population averaged Poisson with fixed effects). Endogenous variables/ exogenous variables (I) Fatalities on interurban roads (II) Accidents on interurban roads Decentralized trend 0.01277** (0.0063) 0.05010*** (0.0192) Unemployment rate −0.02462*** (0.0022) −0.00798*** (0.0019) Hospital density −0.2194*** (0.0591) −0.1484*** (0.0356) Average age 0.05492*** (0.0198) 0.04624 (0.0532) Ratio of population in capital city 2.1736** (0.9587) −0.6156* (0.3615) Motorization rate −0.4431** (0.2075) −0.0650 (0.1697) Highways ratio −0.1157** (0.0589) −0.00485 (0.0165) Tourists −0.0595* (0.0317) 0.1503* (0.0797) Precipitation 0.0972** (0.0391) 0.0193 (0.0212) Year −0.0994*** (0.00543) −0.0539*** (0.0213) Time fixed effects No No Regional fixed effects Yes Yes AR(1) Yes Yes Modified Wald test for groupwise heteroscedasticity 629.86*** 4903.23*** Hausman test Chi 2 26.39*** 88.29*** Levin–Liu test adjusted t* −1.5752* −2.7598*** Shapiro–Wilk test for normality W 0.8277*** 0.6848*** Joint significance test (Wald Chi 2 ) 7462.85*** 15,883.73*** Note: Standard errors are shown in parentheses, robust to heteroscedasticity and grouped by NUTS-3 region. Regressions specify a within-group AR(1) correlation structure for panels. Driver census used as an exposure variable. Statistical significance at 1% (***), 5% (**) and 10% (*). Centralization versus decentralization of traffic law enforcement governance in Spain 1983 REGIONAL STUDIES