Worker mobility and the purchase of low CO₂ emission vehicles in France: A datamining approach

Boroumand, Raphaël Homayoun; Goutte, Stéphane; Péran, Thomas; Porcher, Thomas Article Worker mobility and the purchase of low CO₂ emission vehicles in France: A datamining approach The European Journal of Comparative Economics (EJCE) Provided in Cooperation with: University Carlo Cattaneo (LIUC), Castellanza Suggested Citation: Boroumand, Raphaël Homayoun; Goutte, Stéphane; Péran, Thomas; Porcher, Thomas (2019) : Worker mobility and the purchase of low CO₂ emission vehicles in France: A datamining approach, The European Journal of Comparative Economics (EJCE), ISSN 1824-2979, University Carlo Cattaneo (LIUC), Castellanza, Vol. 16, Iss. 2, pp. 171-205, https://doi.org/10.25428/1824-2979/201902-171-205 This Version is available at: https://hdl.handle.net/10419/320147 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/4.0/ The European Journal of Comparative Economics Vol. 16, no. 2, pp. 171-205 ISSN 1824-2979 http://dx.doi.org/10.25428/1824-2979/201902-171-205 Worker mobility and the purchase of low CO2 emission vehicles in France: a datamining approach* Raphaël Homayoun Boroumand**, Stéphane Goutte***, Thomas Péran****, Thomas Porcher** Abstract This paper seeks to pattern a non-driven geographical classification of French departmental territorial units based on both mobility behavior and passenger car fleet composition. With no mathematical regression analysis but applying datamining methodology to behavior, consumption and geography variables, we have grouped French territorial units into 8 clusters with similar characteristics. The main results reveal that commuters’ behavior with respect to the choice of transport mode varies substantially across clusters (rural and highly rural, urban and highly urban clusters, ...). Conversely, the structure of the French vehicle fleet and French car purchases in terms of engines, tax horsepower and CO2 emissions are similar. This finding should enable state organizations to implement differentiated public policies for environmental and industrial sectors. Alternatively, our paper should help industrial groups to better deploy their economic strategies in line with environmental concerns. JEL classification: O18, P18, R11, R40 Keywords: CO2 emissions, Datamining, Cluster, Worker mobility, Passenger cars, France Graphical Abstract * The authors would like to thank Prof. Thomas Stocker (University of Bern, Former Co-Chair of Working Group 1 of the UN IPCC), Prof. Julien Chevallier (University Paris 8) and Dr. Simon Porcher (University Paris 1 Sorbonne) for enriching discussions on decarbonizing policies in France. ** Paris School of Business (PSB), 59 rue Nationale 75013 Paris France. *** CEMOTEV, Université Versailles Saint-Quentin en Yvelines, Paris-Saclay, France. **** Director of ROLACC Geneva, Associate Researcher University Paris Descartes (Laboratoire CMH) EJCE, vol. 16, no. 2 (2019) Available online at http://eaces.liuc.it 172 1. Introduction The emissions of greenhouse gases (GHG) from transport have grown substantially in recent years, and projections for 2020-2030 are pessimistic (Pérez-Suárez and López-Menéndez, 2015). In many countries of the European Union (EU) such as France, this growth may even thwart the aim of complying with the 3x20 environmental rule 1 ; it is a central component of the EU’s growth strategy for 2020 (EU, 2010). Moreover, it is a long-established finding that CO2 emissions from the most polluting vehicles are directly responsible for a growing number of cancers and respiratory diseases (Dockery et al., 1993; Fullerton et al., 2008). During the year 2012 alone, 7 million premature deaths were attributed to air pollution exposure (WHO, 2014). Particular attention must also be devoted to the emission of fine particles from diesel vehicles (Longhin et al., 2016). Studies linking health and air pollution report that particulate matter (PM2.5) and black carbon emissions from incomplete fuel combustion are responsible for long-term mortality (Bond et al., 2013), and cindynics 2 has provided insights on this matter that can be incorporated into risk analysis models (Fann et al, 2012; Smith et al., 2013). Designed as a component of the new environmental science, the study of territories and, specifically, the negative externalities related to car use in urban areas have already generated good results and helped to provide strong conclusions (Parry et al., 2007). Since the 1990s 3 , the environmental protection of territories has fully included new binding practices for reducing waste, protecting the environment and maintaining an optimal quality of life, which includes the preservation of air quality (IPCC, 2014). The dominant stream of literature focuses on describing the positive externalities associated with control over transport. Indeed, while industry, agriculture and services are all responsible for the growth of CO2 emissions, the most important component of 1 Based on the EU’s 1990 pollution level, this rule calls on Member States to reduce GHG emissions by 20%, have renewable energy represent 20% of the energy mix and achieve a 20% energy savings. 2 The science of risk analysis. 3 From the adoption of the United Nations Framework Convention on Climate Change at the Earth Summit in Rio de Janeiro (June 1992) to the inclusion, for example, of the Environmental Charter (2004) into the French Constitutionality Corpus, a strong chain of hierarchically organized national and international legal standards devoted to the preservation of the environment has been established. In France, the French constitutional judge, since Constitutional Act No. 2005-205 of March 1, 2005 (Article 2), can invalidate national law that contravenes an environmental obligation and fully apply the precautionary principle. R. H. Boroumand et al., Worker mobility and the purchase of low CO2 emission vehicles in France Available online at http://eaces.liuc.it 173 air quality degradation comes from transport (Curiel-Esparza et al., 2016; Rustemoglu and Andrs, 2016; Davydova-Belitskaya and Skiba, 2003). Studies that go deeper into identifying the causes of carbon emissions stress the key role played by factors such as population change (Schellnhuber and Svirejeva- Hopkins, 2008), economic growth per capita, regional energy intensity, the contribution of regional fuel mix and energy and carbon intensity (Gonz‡lez et al., 2014; Remuzgo and Sarabia, 2015). Other specific papers over the 1995-2015 period strictly demonstrate the strong relationship between the per capita income of urban territorial units and the pollution burden they generate4: Boyce, 1994; Torras and Boyce, 1998; Scruggs, 1998; Boyce et al., 1999; McGranahan and Satterthwaite, 2002; Hedenus and Azar, 2005; White, 2007; Kovacs et al., 2013; Zwickl et al., 2014; Holian and Kahn, 2015. If we restrict our attention to the analysis of air pollution factors in territorial units (regions or sub-regional units), in addition to socioeconomic data collection to establish a typography of transport users’ polluting practices in urban areas (Buchs and Schnepf, 2013; Chancel, 2014), it is crucial to have data on the composition of the private car fleet (Agyemang-Bonsu et al., 2010; Kholod et al, 2016). This difficulty is not always easy to overcome because state censuses of private parking facilities, such as the COPERT 4 4 databases for Europe and parts of America and Asia, are heterogeneous and incomplete due to their practice of compiling public data that are not published on a regular basis. Alternative approaches proposed in various studies such as photographs and videos of roads and parking lots (Kholod et al., 2016) provide valuable additional information, but their accuracy is insufficient if the aim is to evaluate the level of pollution caused by a country’s passenger car fleet. For example, data from the COPERT 4 file distinguish among passenger cars, light commercial vehicles, trucks, buses, motorcycles and mopeds. However, the file contains no data on the tax horsepower or engine types of passenger cars. However, several studies claim that among the factors affecting pollution from transport emissions in urban areas, it is most important to consider the composition of the fleet and parking restrictions (Kholod et al., 2016), if possible through composite indicators (Kilkis, 2016) or multidisciplinary studies (Venkatesh et al., 2014). 4 The COPERT 4 program (Computer Program to Calculate Emissions from Road Transport) is a global model for estimating air pollution from road transport. Developed in coordination with the European Environment Agency, it calculates gas emissions estimates. EJCE, vol. 16, no. 2 (2019) Available online at http://eaces.liuc.it 174 Only a public census over a long period of time enables the effective application of data relating to fleets and behavior to measure progress in addressing air pollution from transport. Recall that a 2010 estimate indicated that the share of transport in total global greenhouse gas emissions was 23%, with the corresponding figure for the US being 25% (Sims et al., 2014). Automobile manufacturers in the EU have already implemented innovations to significantly reduce the fuel consumption of passenger cars (Fontaras and Dilara, 2012). However, the overall volume of fuel combustion and corresponding pollution from transport may continue to grow if forecasts indicating a substantial increase in the number of vehicles per resident of a territorial unit, city, region or state continue to hold (IEA/OECD, 2010). Indeed, forecasts from the IPCC indicate that the number of vehicles will double by 2030 and triple by 2050, bringing the total number of light-duty vehicles worldwide to 2 billion (Kahn Ribeiro et al., 2007). Given these uncertainties, two political interventions can be recommended: first, to encourage lower-emissions transport choices related to individual mobility as carpooling, widespread public transportation use, rail system use, and alternative scenarios (Vermote et al., 2013); and, second, promote technological innovation to optimize vehicles and make them less polluting (Palencia et al., 2012). However, in the 1990s, the EU initiated a new policy: moving from a petrol-driven 5 car market to a diesel-driven one. Studies have demonstrated that the potential gains from this industrial policy choice are small and overcompensated by, first, the massive CO2 emissions attributed to the associated supply chain and, second, the resulting increase in black carbon particles because the first generation of diesel vehicles were not equipped with particulate filters (Cams and Helmers, 2013). Worse, the dieselization of the European fleet allowed buyers of passenger cars with diesel engines to affect future savings on their vehicle purchases. With an average of -35% in fuel combustion efficiency for a type-A diesel vehicle compared to gasoline vehicles of the same type, buyers can afford to buy more powerful and therefore more polluting cars (Schipper and Fulton, 2009). On a purely behavioral level, studies seeking to better understand the determinants of the choice of commuter transport modes are numerous. Some incorporate into model inputs regarding the possible degree of leisure during the trip, comfort, ease of postponing travel or the cost related to the daily commute to work 5 In 1990, the proportion of diesel vehicles in the total European fleet did not exceed 10%. R. H. Boroumand et al., Worker mobility and the purchase of low CO2 emission vehicles in France Available online at http://eaces.liuc.it 175 (Wang et al. 2013; Chowdhury et al., 2015; Peer et al., 2016). Others identify the quality and the security of route changes as levers that could encourage the use of public transport instead of private cars (Kingham et al., 2001). It appears that commuters’ age and income influence passenger car purchasing behavior and, therefore, their choice of transport mode (Liu et al., 2016). However, if the ease of access to transport infrastructure and the route practicability are clearly identified as factors affecting the satisfaction of workers with mobility requirements (Wang and Wang, 2016), the literature rarely includes the exact distribution of daily commuting methods used by a population based on geographic data at the level of the territorial unit. There are studies demonstrating the difficulties in commuting to work related to geographical factors (Mackett and Thoreau, 2015) and others that offer a high degree of detail on the modes of transport used daily by commuters (St-Louis et al., 2014). However, few studies are able to combine behavioral data, national vehicle fleet data and geographical data. In an original approach using datamining and clustering, justified in section 2.2, our paper fills a major gap in the literature by providing a highly detailed study and results regarding the level of pro-environmental behavior by French commuters when making a car purchase. Global and national studies highlighting the cause and effect relationship between fuel combustion and the environmental and health problems are now sufficiently numerous and precise to promote environmental awareness among consumers. However, few of these studies are directly applicable to France, where the administrative division of the territory is an intricate and historical legacy with high impact in the implementation of energy policies. Our working hypothesis is that in spite of the great geographical and behavioral heterogeneity of the French territorial units, it is possible to establish a new basic system to implement environmental and industrial policies respectful of the diversity of the population (i.e., departmental units in our paper). The first objective of this research is to document and highlight the specific behaviors and trends among the French in terms of labor mobility, that is to say, mobility linked to job requirements (distance, frequency, infrastructure, etc.). The second objective of this paper is to shape a non-driven geographical classification of French departmental territorial units, based on both mobility behavior and passenger car fleet composition. EJCE, vol. 16, no. 2 (2019) Available online at http://eaces.liuc.it 176 The paper is structured as follows. Section 1 discusses the level of detail available in extant studies on worker mobility. Section 2 describes the conceptual framework and specifies the data and methodology. Section 3 develops the empirical results and economic discussions. Section 4 concludes and provides policy recommendations. 2. Data and Methods 2.1. Data sources Studies on workers’ mobility have increased in France since the 1970s, when the government decided to create a Ministry of Environment 6 . Therefore, the government and the National Statistics and Economic Studies Service 7 (INSEE) have conducted relevant investigations and qualitative analyses. The production of behavioral data on commuters was institutionalized in the late twentieth century, and thus general behavioral facts have been highlighted and widely reported in the academic literature. This research has established that the most disadvantaged workers are residents of medium-sized cities (Villeneuve, 1970). These cities are large enough to ensure a significant number of jobs and small enough to ensure proximity to employment. However, estimates indicate that in the late 1980s, one out of two commuters leaves his town of residence to go to work, meaning that commuting is longer and entails higher speeds (Terrier, 1986). Since that time, given the understanding of what the application of statistical approaches to workers’ mobility can provide in terms of defining and strengthening public policies, institutional data and French government data on "mobility and work" have become abundant. Such data can now be used to highlight new commuter behaviors or reveal the acceleration of existing trends. However, one difficulty is that many of these data are produced and published by public departments that do not communicate in a coordinated manner. The aim of the present study is to combine data from various public institutional sources into a single dataset to restate and study them more consistently. In total, the database that we construct and use in this article contains 6 The Ministry of Environment (otherwise known as the Ministry of Ecology or the Sustainable Development Ministry) and the Ministry of Equipment coexisted in France until 2007, when the two were merged. 7 INSEE is the department responsible for the production and publication of official statistics under the Ministry of Economy. R. H. Boroumand et al., Worker mobility and the purchase of low CO2 emission vehicles in France Available online at http://eaces.liuc.it 177 more than 3.1 million observations distributed across 354 variables to study their evolution over time. Specifically, we present three categories of data: behavioral data, consumption data and geographical data 8 . 2.1.1. Commuter behavioral data (2006-2012) Since 2004 9 , INSEE has performed an annual national population census. This regular census should enable us to accurately count the French population to determine its demographic and social characteristics and, ultimately, develop a profile of French housing and work. This census uses municipalities as the territorial units. Under the terms of the 2002 law on local democracy, for municipalities with a population under 10,000 residents, censuses are comprehensive and are conducted annually over a fiveyear rolling window. For the other municipalities, a sample survey is conducted annually and the entire territory of these municipalities is surveyed at the end of the same fiveyear period. The institutional behavioral data we include in our database are exclusively mobility flow data for the active workforce. According to the definition employed by the censuses conducted by the State authorities since 2004, the active workforce is defined as any person 15 years of age or older reporting a profession (salaried or not) working full or part time, helping a person in his work as an apprentice or a paid trainee, being unemployed or engaged in reduced working time, or finally, being student or retired but still employed. Only the data concerning the workers actually bound by a contract of employment are kept. Mobility flows for year n are available in year n+3 and based on the population census from year n and the geographical division of year n+2. Furthermore, the average age of private cars included in the French fleet as of January 1, 2012, is 8.2 years 10 . For some datasets, including those related to commuter consumption and technological advances in vehicle engines, we considered the years 2013 and 2014. We therefore selected 2006-2012 as our period of study. The data studied are generally available from the institutional websites of the French government 11 or available on request from the 8 See the Appendix for the main municipal and departmental variables under consideration. 9 In accordance with French law no. 2002-276 of February 27, 2002, on local democracy. 10 http://www.ccfa.fr. The French automobile manufacturers committee (CCFA) is the leading French trade association for car manufacturers. 11 http://www.insee.fr; http://www.statistiques.developpement-durable.gouv.fr; http://www.observatoire-des-territoires.gouv.fr EJCE, vol. 16, no. 2 (2019) Available online at http://eaces.liuc.it 178 relevant government services and departmental directorates. The first step in constructing the database was to establish, for each of the 36,688 12 French municipalities, the working population (consisting of individuals over 15 years of age). For the latter number, we report the numbers of individuals using - in their daily commute to work with a starting point of the town of residence considered - a private car, public transportation, two-wheeled vehicles and walking. We then made our observations and drew conclusions based on our study of the variation in the municipallevel utilization rates of each means of transportation, including the numbers and rates of individuals working outside their municipality of residence. Of course, it will be necessary, first, to explain the variation in all of these rates over the period considered for a given territorial unit and second, to compare the differences across territorial units during the period of reference or within a single year. 2.1.2. Consumption and mobility (2006-2014) To avoid aggregation bias 13 , we applied our datamining and clustering methodology at the departmental level. Consequently, another important step in the construction of the final database was to reassemble the 100 French departments 14 by aggregating municipal data. The French departments, administered by General Councils, are the territorial reference units, both politically and geographically. This provides us with the departmental utilization rates of different means of transportation (private cars, public transport, two-wheeled vehicles and walking) and the per department rates of workers working outside their municipality of residence. We then inserted our "consumption" data. First, we include the department-level composition of private cars under 15 years old. This composite comprises variables on motor vehicles. We employed two databases in this regard: one detailing the engine power source and tax 12 This number represents the maximum number of territorial units over the period 2006-2012. In the initial construction of this database, we include the 112 municipalities in the French overseas communities, namely, Martinique, Guadeloupe, French Guyana and Reunion. 13 As the basic administrative level of the database is the municipality, we chose to aggregate at the departmental territorial unit. We assume that geographical heterogeneity is minor but meaningful at this level. In France, administrative levels in order of importance (ascending order) are municipality, department and region. 14 We exclude Mayotte because of the unreliability of the associated data due to the special status of the community. Mayotte has been a French department and an overseas region since 2011, following the referendum of March 29, 2009. R. H. Boroumand et al., Worker mobility and the purchase of low CO2 emission vehicles in France Available online at http://eaces.liuc.it 185 2.2.4. Optimal number of clusters It is necessary to determine the optimal number of clusters. To do so, we plot in Figure 3 the within-group sum of squares (i.e., error sum of squares (EES)), which we would like to be close to zero. Figure 3: Within groups sum of squares The result indicates that the choice of K=8 clusters (or more than 10) seems to minimize the ESS and thus to be a good number of clusters. We will choose K=8 instead of greater than 10 because it will be too difficult to capture, identify and interpret the structure of the departments in each cluster if we take too many groups. Of course, when KN, where N is the total number of departments, the ESS converge to zero. Another statistical test to choose the optimal number of clusters based on the dendrogram tree is the Duda-Hart (2001, sec. 10.10) Je(2)/Je(1) index. This rule is as follows: larger values indicate more distinct clustering. Presented with the Duda-Hart, Je(2)/Je(1) values are pseudo-T-squared values. Smaller pseudo-T-squared values indicate more distinct clustering. We clearly see that this test gives K=3 or K=8 as the optimal number of clusters. We choose K=8 instead of K=3 in the sequel because with only 3 clusters, the segmentation is not accurate enough to allow good economic and policy recommendations. EJCE, vol. 16, no. 2 (2019) Available online at http://eaces.liuc.it 186 Table: Duda-Hart statistical test Number of clusters Duda/Hart Je(2)/Je(1) pseudo T-squared 1 0.3494 175.04 2 0.2670 189.38 3 0.7409 8.04 4 0.3936 67.79 5 0.5967 10.14 6 0.5490 11.50 7 0.5433 10.93 8 0.7313 9.54 9 0.3840 36.89 10 0.3005 13.97 11 0.5624 3.89 12 0.5959 4.07 13 0.0517 18.33 14 0.3190 36.29 15 0.3301 6.09 2.2.5. Clustering projection on principal component We plot in Figure 4 the departmental cluster projection with K=8 in the principal components obtained via PCA. We observe that the non-supervised segmentation divides the projected departments into well separated clusters. This means that our clustering is efficient in separating the data in our dataset. Figure 4: Cluster segmentation for K=8 We put the clustering results obtained with K=3 and K=5 in the Appendix. R. H. Boroumand et al., Worker mobility and the purchase of low CO2 emission vehicles in France Available online at http://eaces.liuc.it 187 2.2.6. Clustering projection on French map We plot in Figure 5 the departmental cluster projection with K=8 on the French geographical map. Figure 5: Cluster segmentation for K=8 We put the clustering map results for each cluster separately in the Appendix. 3. Results and Discussion Before evaluating the determinants of and the links between French mobility and consumer behaviors, a geographical map of possible territorial unit combinations should be made. Table 1 reports the composition of the 8 clusters and the departments associated with them. We present a description of geographical realities highlighted by clustering and the behavioral tendencies of the active workforce over 15 years old in daily commuting to work. Furthermore, we depict a typical private car purchasing structure in each cluster. The discussion section aims to connect these figures and offers an explanation of the environmental awareness of French workers. EJCE, vol. 16, no. 2 (2019) Available online at http://eaces.liuc.it 188 Table 1: Composition of the 8 clusters: Geographical clustering content (departmental based) Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6 Cluster 7 Cluster 8 Essonne Charente Bouches-du- Rhône Allier Aisne Alpes- Maritimes Ain Alpes-de-Haute- Provence Gironde Deux-Sèvres Hauts-de-Seine Ardèche Charente- Maritime Bas-Rhin Calvados Ariège Haute-Garonne Dordogne Nord Ardennes Côte-d'Or Finistère Gard Cantal Isère Eure-et-Loir Paris Aube Côtes-d'Armor Haute-Savoie Haut-Rhin Corse-du-Sud Loire-Atlantique Haute-Vienne Rhône Aude Doubs Hérault Loire Creuse Pas-de-Calais Landes Aveyron Drôme Ille-et-Vilaine Loiret Gers Seine-et-Marne Pyrénées- Orientales Cher Eure Moselle Maine-et-Loire Haute-Corse Seine-Maritime Savoie Corrèze Indre-et-Loire Oise Meurthe-et-Moselle Haute-Marne Seine-Saint- Denis Tarn Haute-Loire Manche Var Morbihan Hautes-Alpes Val-d'Oise Vienne Haute-Saône Marne Puy-de-Dôme Lot Val-de-Marne Vosges Hautes- Pyrénées Saône-et-Loire Pyrénées- Atlantiques Lozère Yvelines Yonne Indre Sarthe Vendée Meuse Jura Somme Nièvre Loir-et-Cher Vaucluse Territoire de Belfort Lot-et-Garonne Mayenne Orne Tarn-et- Garonne R. H. Boroumand et al., Worker mobility and the purchase of low CO2 emission vehicles in France Available online at http://eaces.liuc.it 189 3.1. Geographical, topographical and urbanization description of the clusters Clustering and datamining processes clearly reveal two types of territorial units: urban and rural units. Due to its high level of detail, our study also allows us to distinguish highly urban clusters and highly rural clusters. The following findings are based on the number of cities located in the territorial units and their population density. Specifically, cluster 1 and cluster 3 are urban clusters, but the degree of urbanity is higher for cluster 1. The departments belonging to cluster 1 include most of what we call the "major French regional cities," including Toulouse, Bordeaux, Grenoble, Nantes, Rouen and Versailles. The three largest French cities with the highest population density (Paris, Marseille and Lyon) are in cluster 3. Cluster 4 and cluster 8 are rural clusters, and the degree of rurality is higher for cluster 8. Cluster 1 and cluster 3 are at the top of the cluster ranking, both in terms of urban density and the total number of large cities. Specifically, 58.33% of cluster 1 departments and 100% of cluster 3 departments contain cities of category 2 or 3. The only city in category 1 (Paris) is located in cluster 3. By contrast, 0% of cluster 4 and cluster 8 departments contain cities of categories 1, 2 and 3. If one classes all French departments according to the total workforce population 18 , Paris 19 , the Nord, the Bouches-du-Rhône, the Rhône and the Hauts-de-Seine are the leading departments and are in cluster 3. The geographical and topographical features of the departments contained in cluster 1 and cluster 3, on the one hand, and in cluster 4 and cluster 8, on the other, are also identifiable. Urban clusters are the least mountainous in France 20 , with only 16.67% of cluster 1 departments containing mountains of over 1,500 meters against 0% of cluster 3 departments. In contrast, 50% of cluster 8 departments (22.22% of cluster 4 departments) contain mountains over 1,500 meters. Cluster 8, the most rural, is thus distinguished by the presence of mountains, an isolating factor. Note also that the French notion of isolation also requires identifying opportunities for maritime openings and waterways. In this regard, it is clear that the most rural clusters also have the least 18 Taking, for example, 2012 figures. 19 The French capital is both a department and a city. 20 The city of Grenoble, which exceeds 160,000 residents (since 2013) although lying in the mountainous region of the Alps is an exception. EJCE, vol. 16, no. 2 (2019) Available online at http://eaces.liuc.it 190 coastal geography 21 . Among the 8 clusters, cluster 4 (5.56%) and cluster 8 (14.29%) are again at the bottom of the cluster ranking with respect to coastal departments. Logically, cluster 1 (33.33%) and cluster 3 (40%) are at the top of the same ranking. Finally, the results of our study should also allow us to assess the impact of borders on French mobility and consumption behaviors. While cluster 1 (bordering Spain) and cluster 3 (bordering Belgium) include 8.33% and 20% of border departments, respectively, cluster 4 (bordering Belgium, Switzerland and Spain) and cluster 8 (bordering Italy, Spain, Andorra, Belgium and Switzerland) comprise 16.67% and 35.71% of border departments, respectively. Cluster 6 includes the largest share of border departments at 44.44%. 3.2. Behavioral outcomes and mobility practices in France Interesting results regarding mobility practices in France emerge from the analysis of our 8 clusters. We will present these results with respect to the use of private vehicles for commuting, the rate of public transport use, the rate of walking, the rate of twowheeled vehicle use, and the share of employees who work outside of their home municipality. The classification of clusters based on the rate of private vehicle use for commuting and the public transport utilization rate exhibits near-perfect asymmetry. Table 2 shows that cluster 1 and cluster 3 have among the lowest rates of private vehicle use for commuting at 64.34% and 68.68%, respectively, while simultaneously having higher transport utilization rates, at 22.17% for cluster 1 and 16.08% for cluster 3. Conversely, in cluster 8, the rate of private vehicle use for commuting is the highest (75.39%) while the rate of transit use is the lowest (7.99%). Cluster 2 has a fleet utilization rate of 72.18%, which is relatively close to that of cluster 4 (72.89%), but has a public transit usage rate (13.11%) well above that of cluster 4 (10.94%) 22 . 21 Metropolitan France has more than 3,400 kilometers of coastline (this figure increases to over 4,000 if one includes overseas communities). 22 To neutralize the numbering effect in the interpretation of our results, behavioral data are normalized by relating the figures to the total working population of the territorial units over time; consumption data are normalized by relating them to the total vehicle fleet for the categories and periods studied. R. H. Boroumand et al., Worker mobility and the purchase of low CO2 emission vehicles in France Available online at http://eaces.liuc.it 191 Table 2: Utilization rates of means of transport for daily commuting (2016-2012)* Clusters Private car Public transportation Walking Two wheels Work outside municipality of residence 1 64,34% 22,17% 6,16% 3,79% 72,63% 2 72,18% 13,11% 6,87% 3,53% 67,04% 3 68,68% 16,08% 7,22% 3,80% 63,55% 4 72,89% 10,94% 7,70% 3,60% 62,35% 5 74,08% 9,65% 7,71% 3,65% 62,30% 6 74,29% 9,46% 7,68% 3,76% 62,28% 7 75,02% 8,91% 7,58% 3,73% 62,76% 8 75,39% 7,99% 8,01% 3,59% 61,31% * Cluster average over the period Our results show that when individuals choose between public transport and private vehicles, the decision is explained by the availability of transport infrastructure. The two most urban clusters show a lower use of private vehicles in commuting and greater use of public transport. Meanwhile cluster 8, the most rural cluster, exhibits the highest rate of vehicle use in commuting to work and the lowest rate of transport use. Importantly, the share of commuters walking to work has an obvious relationship with the utilization rates for private vehicles and public transport. The rates of walking range from 6.16% (cluster 1) to 8.01% (cluster 8). Cluster 1 exhibits the lowest rate of walking, as it uses fewer cars and more public transport. Conversely, in cluster 8, residents walk more, meaning that there is less use of public transport and more car use. This suggests a negative correlation between public transport use and walking. However, other clusters do not support this correlation. Cluster 3 exhibits greater public transport use than cluster 2 (16.08% and 13.11%, respectively) and a higher walking rate (7.22% and 6.87%, respectively). Clusters 4, 5, 6 and 7 are similar in the majority of our findings (with respect to the rates of private vehicle use, public transport use and walking). The rates of two-wheeled vehicle use for commuting are very similar, ranging from 3.53% (cluster 2) to 3.80% (cluster 3) and therefore have no connection with the differences in the use of private vehicles (excluding motorcycles) or public transport in commuting. Regarding the rate of workers working outside their municipality of residence, cluster 1 and cluster 2 have rates of 72.63% and 67.04%. Clusters 3, 4, 5, 6, 7, and 8 have fairly similar rates, ranging between 61.31% and 63.55%. EJCE, vol. 16, no. 2 (2019) Available online at http://eaces.liuc.it 192 3.3. The structure of the vehicle fleet and the choice of cars purchased Despite substantial differences in mobility behavior (Table 2), Tables 3 and 4 report very small differences across clusters in the structure of the fleet and in the choice of cars purchased. In terms of gasoline engines, the proportion of cars purchased with engines of less than 6 horsepower per cluster ranges as follows: 19.26% for cluster 5 to 20.30% for cluster 1. The figure for engines between 6 and 7 horsepower ranges from 13.52% (cluster 8) to 14.22% (cluster 1). For vehicles with more than 8 horsepower, the figure ranges from 5.28% (cluster 2) to 5.79% (cluster 3). There is also little difference in the proportion of diesel engines for engines under 6 horsepower, with the share ranging from 22.91% (cluster 1) to 24.07% (cluster 8) and diesels with more than 6 horsepower, ranging from 36.35% (cluster 1) to 37.31% (cluster 5). Finally, for uncommon engine types such as LPG vehicles 23 with less than 6 horsepower, the proportions by cluster are equally similar, ranging from 0.24% (clusters 6, 7 and 8) to 0.27% (cluster 1). For LPG vehicles over 6 horsepower, the figures are between 0.30% (cluster 6, 7 and 8) and 0.32% (clusters 2, 3, 4 and 5). For electric cars, the share is the same in all clusters at 0.02%. A more detailed analysis of the share of vehicles by engine type yields similar proportions across clusters. For example, for vehicles with 1 to 4 horsepower, the share ranges from 18.55% (cluster 8) to 18.92% (cluster 1). The pattern is similar for the purchase of vehicles with 7 or 9 horsepower, where the shares are between 16.15% (cluster 8) and 16.88% (cluster 1) and 3.61% (cluster 8) and 3.75% (cluster 1), respectively. Table 4 presents the details of purchases by horsepower and cluster. We find that the shares of purchases for the various vehicle types across clusters are very similar. 23 Liquefied petroleum gas R. H. Boroumand et al., Worker mobility and the purchase of low CO2 emission vehicles in France Available online at http://eaces.liuc.it 193 Table 3: Fleet of private cars under 15 years (on january 1st of each year) by energy source and administrative power class (2006-2014)* Clusters Gasoline <6 horsepower Gasoline [6;7] horsepower Gasoline ≥8 horsepower Diesel <6 horsepower Diesel ≥6 horsepower Gasoline LPG <6 horsepower Gasoline LPG ≥6 horsepower Electric cars Undetermined energy source and/or adminitrative power 1 20,30% 14,22% 5,60% 22,91% 36,35% 0,27% 0,31% 0,02% 0,03% 2 19,69% 13,71% 5,28% 23,67% 37,04% 0,26% 0,32% 0,02% 0,03% 3 19,99% 14,00% 5,79% 23,21% 36,38% 0,26% 0,32% 0,02% 0,03% 4 19,51% 13,65% 5,53% 23,74% 36,95% 0,26% 0,32% 0,02% 0,03% 5 19,26% 13,48% 5,31% 24,02% 37,31% 0,25% 0,32% 0,02% 0,03% 6 19,48% 13,61% 5,45% 23,87% 37,00% 0,24% 0,30% 0,02% 0,03% 7 19,40% 13,57% 5,36% 23,98% 37,10% 0,24% 0,30% 0,02% 0,03% 8 19,34% 13,52% 5,34% 24,07% 37,14% 0,24% 0,30% 0,02% 0,03% * Cluster average over the period Fleet of private cars under 15 years (on january 1st of each year) by administrative power class (highly detailed, 2006-2014)* Clusters [1;4] horsepower 5 horsepower 6 horsepower 7 horsepower 8 horsepower 9 horsepower [10;11] horsepower ≥12 horsepower Undetermined horsepower 1 18,92% 24,78% 23,07% 16,88% 6,26% 3,75% 3,47% 2,86% 0,01% 2 18,78% 25,04% 23,53% 16,56% 6,21% 3,65% 3,43% 2,79% 0,01% 3 18,79% 24,87% 23,21% 16,53% 6,20% 3,73% 3,56% 3,09% 0,01% 4 18,64% 25,07% 23,60% 16,37% 6,16% 3,66% 3,50% 2,99% 0,01% 5 18,59% 25,16% 23,83% 16,31% 6,16% 3,60% 3,44% 2,91% 0,01% 6 18,62% 25,18% 23,64% 16,22% 6,18% 3,65% 3,49% 3,00% 0,01% 7 18,57% 25,26% 23,75% 16,20% 6,17% 3,62% 3,46% 2,95% 0,01% 8 18,55% 25,30% 23,79% 16,15% 6,17% 3,61% 3,47% 2,95% 0,01% * Cluster average over the period EJCE, vol. 16, no. 2 (2019) Available online at http://eaces.liuc.it 194 Table 4: Total number of new cars registered in the year by CO2 emissions (2006-2014)* Clusters ≤100 g [101;120] g [121;130] g [131;160] g [161;165] g [166;200] g [201;250] g >250 g Undedermined carbon emissions 1 6,85% 35,92% 11,22% 34,50% 1,44% 7,49% 2,05% 0,51% 0,02% 2 6,91% 36,26% 11,21% 34,27% 1,45% 7,35% 2,02% 0,51% 0,01% 3 7,14% 35,73% 11,29% 34,07% 1,45% 7,55% 2,14% 0,61% 0,02% 4 7,09% 35,98% 11,28% 34,05% 1,45% 7,44% 2,10% 0,59% 0,02% 5 7,08% 36,16% 11,29% 34,05% 1,45% 7,33% 2,06% 0,57% 0,02% 6 7,08% 36,33% 11,34% 33,88% 1,51% 7,26% 2,02% 0,56% 0,01% 7 7,04% 36,38% 11,34% 33,95% 1,50% 7,22% 2,00% 0,56% 0,01% 8 7,07% 36,43% 11,33% 33,93% 1,50% 7,18% 1,99% 0,55% 0,01% * Cluster average over the period This difference calls into question the limited effectiveness of tax incentives in affecting vehicle purchasing decisions, especially given that the share of new cars purchased with CO2 emissions between 101g and 120g varies between 35.7% (cluster 3) and 36.4% (cluster 8). We therefore find that in terms of CO2 emissions, the differences across clusters in the choice of vehicle types are small but the differences among vehicle categories are important and do not respond to tax incentives implied by the "bonus-malus". Differences in the mobility of the French, as well as tax incentives for purchasing vehicles, appear to have little influence on vehicle choice. Therefore, one should consider the question of the supply of vehicles in France and the industrial policies of France and the EU with respect to the automotive industry. R. H. Boroumand et al., Worker mobility and the purchase of low CO2 emission vehicles in France Available online at http://eaces.liuc.it 201 Table 6: Consumption main variables' table Variables available for departmental and regional territorial units* Fleet of private cars under 15 years (on january the 1st of each year) by energy source and administrative power class (2006-2014): Gasoline <6 horsepower Gasoline [6;7] horsepower Gasoline ≥8 horsepower Gasoline LPG <6 horsepower Gasoline LPG ≥6 horsepower Diesel <6 horsepower Diesel ≥6 horsepower Electric cars Gasoline + super ethanol <6 horsepower Gasoline + super ethanol [6;7] horsepower Gasoline + super ethanol ≥8 horsepower Electricity + gasoline <6 horsepower Electricity + gasoline ≥6 horsepower Electricity + diesel Undetermined energy source and/or adminitrative power class Fleet of private cars under 15 years (on january the 1st of each year) by administrative power class (highly detailed, 2006-2014): [1;4] horsepower 5 horsepower 6 horsepower 7 horsepower 8 horsepower 9 horsepower [10;11] horsepower ≥12 horsepower Undetermined horsepower Total number of new cars registered in the year by CO2 emissions (2006-2014): ≤100 g [101;120] g [121;130] g [131;160] g [161;165] g [166;200] g [201;250] g >250 g Undedermined carbon emissions Total number of new cars registered in the year (2006-2014) Total number of diesel new cars registered in the year (2006-2014) * Annual and periodic data EJCE, vol. 16, no. 2 (2019) Available online at http://eaces.liuc.it 202 Table 7: Geographical main variables' table (by size of the territorial unit) Communal variables Municipality's geographic code (INSEE) Municipality's name Department's number Department's name Indicator of border department Region's name Telephone indicator Departmental variables Number of cities in category 1 (Paris, >2 million inhabitants) Number of cities in category 2 (from 200 000 to 1 999 999 inhabitants) Number of cities in category 3 (from 100 000 to 199 999 inhabitants) Number of cities in category 4 (<100 000 inhabitants) Department's number Department's name Indicator of border department Region's name Telephone indicator Regional variables Number of cities in category 1 (Paris, >2 million inhabitants) Number of cities in category 2 (from 200 000 to 1 999 999 inhabitants) Number of cities in category 3 (from 100 000 to 199 999 inhabitants) Number of cities in category 4 (<100 000 inhabitants) Region's name Indicator of border region Telephone indicator R. H. Boroumand et al., Worker mobility and the purchase of low CO2 emission vehicles in France Available online at http://eaces.liuc.it 203 Figure 6: Cluster segmentation for K=3 and K=5 K=3 K=5 EJCE, vol. 16, no. 2 (2019) Available online at http://eaces.liuc.it 204 Figure 7: Cluster segmentation for K=8 for each cluster (ordered Cluster 1 to 4) Cluster 1 Cluster 2 Cluster 3 Cluster 4 R. H. Boroumand et al., Worker mobility and the purchase of low CO2 emission vehicles in France Available online at http://eaces.liuc.it 205 Figure 8: Cluster segmentation for K=8 for each cluster (ordered Cluster 5 to 8) Cluster 5 Cluster 6 Cluster 7 Cluster 8