1Although criticism of the car has a very long history, the systematization of such criticism has been relatively recent. It was indeed only in the 1970s that the car, through the identification of its human costs—road deaths and pollution—became a public problem, in the sense given to it by Joseph Gusfield (1981). [1] The criticism of the car could then build on what Daniel Miller (2001) calls a “literature of externalities,” which sets out to identify all costs involved in the use of the automobile—as seen particularly in the work of Richard C. Porter (1999). [2] The work of Alfred Sauvy, then a professor at the College de France, ironically titled Les quatre roues de la fortune (The four wheels of fortune) (1968) highlighted the many costs generated by the spread of the automobile; in Germany, the essay by Hans Dollinger (1972) pursued the same objective. The car seems to be viewed as a polymorphic problem, [3] while very few voices still laud it (Lomasky 1997).

2The great interest of this work is that it restores the collective dimension of the cost of automobility. But the methods developed to understand these costs at the national level imperfectly reflect the social stratification of the costs generated. [4] Falling within the framework of national accounting, the analyses proposed in these works essentially divide owners from non-owners, which is insufficient in societies where over 80% of households own at least one car and where the cars of different households are quite dissimilar (Coulangeon and Petev 2013). [5]

3However this stage in “car consciousness” (Flink 1972)—that of its creation as a public issue—only makes sense if one situates it in terms of the chronology of the diffusion of a good that was initially an elite one, and then gradually became a mass commodity. As noted in the mid-1970s by Luc Boltanski in a much cited work, the critique of the car emerges at exactly the same time as the working classes gained access at a mass level to a good from which they had previously been to some extent excluded (Boltanski 1976). Yet this criticism largely neglects the social determinants of the harm done by the car to focus only on that aspect which is concerned with the area of choice and individual behaviour (Comby and Grossetête 2012). One of the aims of this article is to recognise the importance of these determinants through the study of the social distribution of the two main collective costs of the automobile: road deaths and pollution.

4This article intends to follow and to discuss the approach initiated by Boltanski— in the very different context of the contemporary period and based on recent data (Box 1)—concerning the forms of competition between social groups that are played out in road interactions (Boltanski 1975, 1976). Boltanski emphasized at the time that the class struggle on the road took a primarily symbolic form. As conflict on the road is too dangerous, and since that danger potentially affects all social groups in the same way, competition on the road focused on driving styles and vehicle characteristics.

5Yet it seems that the symbolic angle from which Boltanski considered the competition between different social groups over car use is no longer the most relevant one today. Other questions that are more closely related to public issues are now more important than symbolic conflict. These issues are indeed more acute in a context where, firstly, road safety is becoming a major political issue and, secondly, governments now try to define the qualities of a good car, a “green (or town) car”. [6] The recent initiative of publicising the palmares de la voiture citoyenne (“winner of the Green Car prize”) by the French Ligue contre la violence routière (League against Road Violence) also attempts to connect these two types of externalities. However, by concerning itself only with new vehicles, and ignoring a car’s level of use, this type of classification remains of limited value. Taking as its starting-point the hypothesis that competition on the road is not only symbolic, this article, on the other hand, makes it possible to assess and articulate such externalities at the level of the entire stock of cars owned by households, while understanding differentiations in the use of vehicles.

6The aim here is to reveal the social structures of road safety and the environmental sustainability of car models, by arguing that relatively homogenous lifestyle choices serve to differentiate between very unequally polluting and dangerous cars. The car thus makes it possible to grasp the social logic of a twofold relationship to risk, to road risks and environmental safety risks that are traditionally approached as two quite distinct public issues. Such an approach makes it possible to evaluate the different contributions of social groups to two classic externalities—the study of which, traditionally the purview of economics, neglects the social factors which are nonetheless central.

7A two-part hypothesis structures the analysis presented below. The first hypothesis is that the car, which makes one of the largest contributions to the carbon footprint of households, is the source of relatively few environmental concerns for those same households [7]. Secondly, and against the idea of an enlightened ruling class that is post-materialistic and has high awareness of road safety issues and the environment (Inglehart 1993; Wallenborn and Dozzi 2007), it is predicted that the most polluting and dangerous models, but also those that are the safest, are in the hands of the categories with the highest volume of capital.

8These questions are addressed here using data from the French “National Survey of Transport and Travel- ENTD” (Enquête nationale transports et déplacements) conducted in 2007–2008 (Box 1), which describes fairly accurately the car population of households. The aim is to link driver characteristics with those of their vehicles, by connecting the social distribution of two types of environmental externalities of the automobile, the injuries caused to people by the different categories of vehicles on the one hand (although that accident data is limited in ENTD 2008), and air pollution attached to their use, on the other. From a detailed analysis of the characteristics of the vehicles identified in the 2008 ENTD, this article discusses the rationales structuring the social distribution of different types of car models, in relation to these two types of externalities.

9After reviewing the scope and limitations of the hypotheses proposed in the pioneering work by Boltanski, the analysis in this article is concerned, firstly, with the social distribution of the environmental and safety performance of vehicles. Next the article addresses the construction of the social space of the dangerous and polluting car, taking into account the rationales for the use of vehicles: the fundamentally multidimensional nature of the automobile clearly calls for an analysis that makes it possible to simultaneously embrace the different characteristics of vehicles.

10Proceeding then to a typology in order to move beyond the wide variety of models, the last part of the article explores the social factors in car choice and attempts to show how several social rationales for the choice of a car model make a contribution differentiated by social group to the human and environmental costs of the car.

Competition over safety and level of pollution

11When Boltanski wrote his landmark article in 1975 he was working in a context of much higher mortality than today [8]—more than 15,000 people on average were then being killed every year on the roads of France [9]. Studying the social uses of the car, he put forward the idea that its democratisation opened the way to a form of competition for space between different categories of owners of vehicles and drivers, one specifically involving the safety of different classes of road users.

12With the exception of the autoroute (motorway), [10] the road could at that time seem to be a meeting place for widely different social groups, thus combining spatial proximity with social distance (Chamboredon and Lemaire 1970). The road very likely still remains a less segregated space today than some other types of spaces. Unlike other places, roads are not ones where avoidance and relegation strategies are practised. Although one can choose one’s neighbourhood according to one’s neighbours, one cannot choose one’s road according to the types of drivers that one might meet on it. The road falls within the category of “ownerless property” (Boltanski 1976), which, like all public goods, cannot be privatized. Competition then moves from the monopolization of space to the protection of motorists from other users, a protection that works through the specific characteristics of the vehicles of different categories of users. In other words, since one cannot avoid the encounter, one needs to arm oneself to confront it; however not all social groups are equally equipped for this confrontation (Boltanski 1975).

13Although it was considered in its infancy as a form of conspicuous consumption [11], the mass-market car links social groups whose car models are very dissimilar in terms of their potential danger to other road users. In this context, the choice of car is not a private matter, since it determines the frames of interaction in which social groups are engaged while driving (Boltanski 1975), especially in relation to the space occupied by different vehicles. The road—the unique arena where socially situated owners encounter each other, and have little control over their choice of interaction partners—opens up from this point of view a space of conflict and competition, where road safety is one of the main externalities.

14The context in which Boltanski envisioned strategies for the occupation of space in the mid 1970s (particularly in terms of the symbol of the “big car,” imposing docility on other slower and more vulnerable users) and forms of road interaction between social groups, is very different from the current one. At the time, manufacturers as much as public authorities and road users were beginning to regard road safety as a key issue. Forty years later, the issue of driver safety seems by contrast to be an omnipresent concern [12].

15Accident statistics have been used to show that the social characteristics of drivers are correlated with their behaviour on the road; the determinants of sex, age and how long the driver has held a licence, among others, have been particularly well explored [13]. Matthieu Grossetête shows that the social group dimension was often overlooked by research, even though there exists a strong social basis to road deaths (Grossetête 2010). But we can also imagine that this relationship between social group and road mortality may be mediated by another variable: the lethal and protective powers of vehicles—properties that are not independent of the social characteristics of the drivers (Coulangeon and Petev 2013). Although Grossetête links the higher accident rates of the lower social classes to the characteristics of a mobility rooted in local travel, which is by far the most dangerous, [14] the existing literature on the subject seems rather to hide the material bases of the “social roots of road deaths,” hypothesising that drivers are driving similar vehicles, but that some of them stand out because of their dangerous behaviour. This obscurantism is the result of several factors. Firstly, it is conceivable that car manufacturers do not spontaneously encourage the dissemination of data on the determinants of road deaths when they are related to vehicle characteristics. Furthermore, the nature of the data in question is complex, and in the recording of accident data the finer characteristics of damaged vehicles tend to disappear in favour of the description of drivers and passengers. The multiplicity of models and the very diversity of accidents (involving one or more cars) complicate the relationship between the accident and the models involved. Boltanski, even though he shows that the automotive tastes of the different social classes are part of a system, only envisages the over-representation of the lower classes among accident victims within the perspective of their lesser socialization in driving practice. “If the chances of a driver being involved in a traffic accident and, if involved, of being killed or injured, are—annual mileage being equal—at their highest among drivers under the age of twenty-five, and secondly, rise the lower a driver is in the social hierarchy, this is because the increase in the number of cars in circulation and the social heterogeneity of drivers […] particularly penalizes those who have most recently entered the road space market.” (Boltanski 1975, p. 41). However, thirty years later, Grossetête still finds that there is a strong relationship between road deaths and social group, even though the spread of the automobile is now very broad [15]. The final difficulty lies in the fact that different organizations measure occupant protection, not the dangerousness of the model: road safety tests act as if the driver had a wall in front of him, devoid of all social properties, and not another driver, even though in 2009 accidents involving two passenger vehicles accounted for more than half of all fatal road accidents (ONISR 2011) [16]. It is thus not possible to understand with any degree of precision the differing levels of danger that drivers are exposed to when they are at the wheel of one vehicle or another. The analyses presented here attempt to resolve the significant skewing of the data by differentiating in particular between vehicle dangerousness and protection capacity.

16Beyond the differentiated perception of risks and bodies (Boltanski 1971), the social roots of road deaths have perhaps something to do with the degree of protection offered by the different categories of vehicles, which are unevenly distributed according to social group.

17While the issue of road safety began to emerge in France during the 1970s, interest in the issue of automobile pollution came later. The articulation, recently, of social and environmental concerns (Cornut et al 2007) makes it possible to distinguish different types of issues: uneven exposure to pollution-related nuisances, social effects of environmental policies, environmental effects of social policies or unequal contribution of social groups to environmental costs. It is this last issue that holds our attention here, through the case of automobile pollution, where all the indicators point to the fact that it now accounts for a significant share of overall pollution at the city level just as at the global level. Vehicle traffic produces carbon monoxide, nitrogen oxides and hydrocarbons. According to MacNeill (2010), about two thirds of carbon monoxide emitted in rich countries in 1980 came from road vehicles; the proportions were 47% for nitrogen oxide and 30% for hydrocarbons. According to the survey and statistical studies department of the French Ministry of Ecology, Sustainable Development and Energy, in 2005, 27% of greenhouse gases in France came from the transport of goods and of individuals (Longuar et al. 2010).

18In fact, French governments are now increasingly paying closer attention to the pollution caused by motor vehicle traffic [17], a problem that has led to two responses in France: the promotion of altermobilités [18] and improvement of the national vehicle stock. Among these measures, regular and obligatory vehicle safety and pollution testing was the pioneering measure in France; established in 1992, it has to be carried out within the six months before the fourth anniversary of the date of first registration of the vehicle and be renewed once every two years. Technical inspection now specifically includes testing of the vehicle’s pollution levels. On January 1, 2008 the ecological bonus-malus was set up, a measure promoted by the standing debate on the environment (“Grenelle de l’environnement”), a framework intended to sanction and deter acquisition of higher polluting new vehicles in favour of those vehicles considered cleaner.

19The definition of what makes a good car thus becomes a public issue, not just a private matter. A “good” car is not only a comfortable, spacious, reliable and powerful car since it must also, in terms of environmental constraints, be a modest and economical vehicle. What do we learn in this respect from the choice of vehicles, according to their energy performance, about the behaviour of different social groups, and their relationship to the environment?

The social distribution of the energy-saving and safety performance of vehicles

Owning a big car

20The protective power, dangerousness and energy performance of vehicles can be measured by their weight and the list of active and passive safety features with which they are equipped. [19] How are such characteristics distributed among social groups?

21Vehicle weight is a fairly reliable indicator of the dangerousness and the protection offered by different car models, although it is not known to or directly perceptible by users [20]: it is above all the mass (or more accurately the differential mass) of the vehicles involved in an accident that is used to calculate the intensity of the shock to the passengers of each vehicle involved (IBSR 2009). This variable is also used by insurers (who are part of a safety and car repair association) to produce estimates of various costs (repairs, replacement and personal liabilities) associated with different vehicle models. Each vehicle is classified according to a ranking formula defined by three indicators. Among these, the “group,” which represents for the insurers the “inherent dangerousness” of the vehicle for others and determines the extent of civil liability, is calculated by factors based on variables such as the weight, power and top speed of vehicles [21]. The vehicle mass is also highly correlated with energy performance, although this also depends on the age of the vehicle. [22]

22The univariate statistics on vehicle weight according to socio-occupational status allow us to draw several conclusions (Table 1). The main result, not surprisingly, is that the weight of the vehicle increases with socio-occupational status. Among employees, it is managers who drive the heavier cars, while the self-employed, who also use their cars for a business that often requires the transport of goods (and therefore vans or estate cars which are bigger vehicles), own heavier models than the majority of employees.

Table 1

Statistics for the weight in kilograms of vehicles according to socio-occupational class and sex of driver

Men Women Mean Standard Deviation Mean Standard Deviation Farmers 1,223 500 1,277 402 Artisans & other self-employed 1,411 400 1,220 306 Public sector managers 1,215 321 1,165 344 Private sector managers 1,301 438 1,230 471 Intermediate occ. public sector 1,216 347 1,126 330 Intermediate occ. private sector 1,225 377 1,134 337 Public sector employees 1,194 471 1,122 278 Private sector employees 1,146 268 1,136 280 Skilled workers 1,186 382 1,136 317 Unskilled workers 1,158 280 1,113 276 All 1,232 385 1,146 324

Statistics for the weight in kilograms of vehicles according to socio-occupational class and sex of driver

Field: All persons owning a car whose characteristics are part of the survey.
Interpretation: In 2008, the average weight of cars owned by farmers was 1,223kg.

23If we look more closely at the PCS (professions et catégories socioprofessionnelles, socio-occupational categories), a fairly strong distinction between people employed in the public and private sectors can be observed (Singly and Thélot 1988). Among the most qualified employees, middle managers and professionals, it is the private sector employees who have relatively heavier vehicles than their public counterparts. Those exercising intermediate occupations in the private sector on average drive cars that are even heavier than those of public sector managers. The popular expression “grosse voiture “(‘big car’) has a literal meaning, demonstrating how Boltanski’s argument is relevant thirty years later, “To talk about a nice car, a luxury car, a grand tourer, a powerful car, […], a convenient car, economical, utilitarian […], a thoroughbred, a monster, […] is to use the categories of perception of material objects to designate classes of appropriators defined by age, sex and especially their position in the class structure.” (Boltanski 1975, p. 36).

24The statistics also highlight another and equally interesting trend, and one somewhat neglected in the automotive literature: equipment for women. What is notable and is intuitively easy to understand is that women on average drive relatively lighter cars than men. There is a fairly even distribution of the mean, around 1150 kilos, for vehicles belonging to female salaried workers, a phenomenon more accentuated for women than for men. Nonetheless, the weight of the cars belonging to women show the same social gradient as the vehicles belonging to men; thus, the cars of private sector female managers are on average heavier than all the cars of male employees, except for those of their counterparts. In other words, although sex is a determining factor of such a characteristic, social class remains influential. Although overall they are “less dangerous” and less protective because they are lighter, women’s vehicles remain subject to a social gradient of dangerousness that is as pronounced as it is for those of men.

The social distribution of security features: cars of low quality vs. high-specification cars?

25Possession of cars with active and passive safety systems varies greatly by the vehicle owners’ socio-occupational class (Table 2). However, the objective function of such features is not only a protective one: Sam Peltzman (1975) shows that increased security in fact leads drivers to take more risks because they feel better protected. The existence of security features thus reinforces the potential dangerousness of vehicles. [23]

Table 2

Frequency of different safety features according to sociooccupational class of driver

Active security Passive security ABS Anti-skid reduction Cruise control Driver airbag Passenger Airbag Farmers 71 26 11 40 34 Artisans & other self-employed 80 32 23 59 50 Public sector managers 82 27 18 70 63 Private sector managers 89 38 32 80 74 Intermediate occ. public sector 75 22 13 64 56 Intermediate occ. private sector 81 26 19 66 57 Public sector employees 75 21 11 59 49 Private sector employees 77 18 11 60 57 Skilled workers 73 18 9 49 41 Unskilled workers 67 12 8 46 38

Frequency of different safety features according to sociooccupational class of driver

Field: All persons owning a car whose characteristics are part of the survey.
Interpretation: Of the 100 cars owned by farmers, 71 are equipped with an ABS braking system.

26Here again, disparities go hand-in-hand with socio-occupational class; regardless of the type of security feature, it is majority-owned by the more affluent social classes. Where the features introduced from the 1990s are concerned, the differences remain pervasively present; thus, a car belonging to a private sector manager is two times more likely than the car of an unskilled worker to have a passenger airbag. Although the level of equipment of the newer features is lower, disparities remain in the same order of magnitude.

27However, it is likely that these “cars without qualities” are mostly older cars, as the household car population differs widely according to the age of vehicle [24]; here we are thus measuring less the differences in safety features than the relative obsolescence of the automotive equipment of different households. It is therefore important to compare what is comparable, in this case vehicles up to six years old in 2008. If features such as ABS and airbags have become commonplace, often equipping standard models (almost nine out of ten new vehicles are equipped with each of the two features), differences nonetheless persist. The more recent cars owned by private sector managers are nearly twice as likely to have anti-skid reduction (ASR) (46%) than vehicles of a comparable age driven by skilled workers. Regarding ABS, which is now standard equipment on most vehicles, the gap was 10 percentage points in favour of private sector managers compared to skilled workers. In short, the social differences in the protective capacities of cars cannot be explained only by an effect of the age of the fleet; the capacity for protection against the risk of accident and its effects is unequally distributed within the social space, whatever the age of the vehicles in question. The potential for protection and the dangerousness of vehicles appears to maintain a homological relationship with the space of social positions.

28Although Boltanski had indeed observed that the vehicles of each of the social groups were clearly differentiated, it should be added that this competition is not purely symbolic; the safety equipment features and vehicle weights also demonstrate inequalities in road risk, involving not only an uneven distribution of opportunities, but also an unequal distribution of misfortune.

Social distribution of energy consumption

29Let us now turn to the distribution of the energy consumption [25] of vehicles according to their owners’ socio-occupational class (Table 3).

Table 3

Statistics of consumption per 100km of all vehicles and diesel vehicles of less than 6 years old according to the socio-occupational class of the owner

All vehicles Diesel vehicles less than 6 years old Mean Median Mean Median Farmers 7.3 7 6.8 6.6 Artisans & other self-employed 7.7 7 7.5 7 Public sector managers 6.7 6.5 6.2 6 Private sector managers 7 6.8 6.5 6 Intermediate occ. public sector 6.8 6.5 6.3 6 Intermediate occ. private sector 6.7 6.5 6.1 6 Public sector employees 6.6 6 6.2 6 Private sector employees 6.6 6 6.2 6 Skilled workers 6.9 6.5 6.4 6 Unskilled workers 6.8 6.5 6.2 6

Statistics of consumption per 100km of all vehicles and diesel vehicles of less than 6 years old according to the socio-occupational class of the owner

Field: All persons owning a car whose characteristics are part of the survey.
Interpretation: In 2008, the mean consumption of a car owned by a farmer is 7.4 litres per 100 kilometres.

30If we look at the first column of the table, it appears that all the cars of employees appear on average to have similar energy consumption, oscillating between 6.6 and 7.7 litres per 100 kilometres, while all of the self-employed categories drive vehicles that consume about 10% more energy [26].

31This finding can be explained if we look more specifically at a particular category of vehicles, diesel vehicles under six years old. The results mentioned above are confirmed. Self-employed and qualified private sector employees appear to own relatively less efficient vehicles, while the most highly qualified public sector employees own more efficient vehicles—which brings them closer to the least skilled employees—both employees and manual workers.

32This first section shows that different social groups, according to their capital assets, will be using cars that are more or less dangerous—for them and for others—and more or less polluting. At this point, the most vulnerable vehicles (but also the least dangerous) are to be found more in the lower classes, who also drive relatively more polluting vehicles, because of the average age of their car. However, a vehicle that is very greedy in energy consumption terms and that is rarely driven will obviously pollute less than an efficient vehicle being driven a lot; equally, the level of dangerousness of a vehicle is related to the intensity of its use. We think in terms of fixed mileage (or kilometrage), although there are also differences between social categories in terms of their annual mileage. Also, bi-variate statistics have obvious limitations: the characteristics of the vehicles are not independent of one another; a new car is more likely to be equipped with the latest safety features. Technical progress, by improving energy efficiency and vehicle safety features, thus complicates the analysis; vehicles of varying ages according to social group are very likely to present very different levels of energy consumption and safety features. Furthermore, the analysis focuses on the characteristics of vehicles and not on their usage, which if it was taken into account would very likely change the conclusions advanced so far. It is thus necessary, beyond the apparent heterogeneity of the goods held by different social groups, to distinguish them more systematically, by reference to all of their characteristics. But above all, it is necessary to consider the use of vehicles. Correspondence analysis proves to be a particularly useful tool here.

From the social space of car models to the construction of a typology of models

Stratification of energy consumption and the dangerousness of vehicles

33In this section I construct a multiple correspondence analysis (MCA), which will allow us to encompass all the characteristics of different vehicles, a necessary method because it seems appropriate to the inherently multidimensional nature of the car. From the list of vehicles available to households, I selected the models that have a frequency of at least fifteen cases, [27] and excluded commercial vehicles. Then, for each model, I derived means of the different technical characteristics and usage. The resulting table gives the mean characteristics for each model, of all versions of the model concerned, and includes means for the average taxable horsepower, age, weight, urban consumption and annual mileage, and the percentages of the model purchased new and with diesel engines. I then transformed such continuous variables into categorical variables, dividing them into quartiles in order to obtain the variables that have an equal number of terms. In this way, each model is allotted a mean horsepower quartile, an average age, and an average weight, etc. A multiple correspondence analysis is then performed on the data. The MCA thus uses the following active variables: average age, average weight, mean horsepower, mean consumption, the percentage of new vehicles, the percentage of diesel vehicles, the mean annual distance driven, and the relative scarcity of models (see the Appendix for details on how the MCA was constructed).

34The relationship between factors 1 and 2 provides a stratification of vehicles according firstly to weight and power (axis 1), which can be read as an axis of dangerousness and, secondly, a combination of age of vehicle and their level of pollution (axis 2). In the following paragraphs I discuss the representation of individuals in the plane of the two axes (Figure 1. See Figure 4, in the Appendix, for the representation of the range of active variables in the plane.).

Figure 1

Representation of vehicles in terms of the plane of axes 1 and 2 of the correspondence analysis

Representation of vehicles in terms of the plane of axes 1 and 2 of the correspondence analysis

35Axis 1, 16% of the inertia, can be read as a gradient of the dangerousness of vehicles: on the left are heavier models with poor energy performance (Mercedes E Class, Renault Espace—furthermore these vehicles usually have diesel engines) against vehicle models of small engine displacement and small size (Renault R5, Peugeot 106) on the right. Concentrating 10% of the inertia, axis 2 appears to provide information on level of pollution. This axis also contrasts high and low mileage vehicles. At the top are the relatively high fuel consumption models (because they are older car models, as on the right quadrant of the graph—e.g. Renault Super 5, Peugeot 205, Citroen AX, etc.), or particularly energy-hungry newer models (e.g. Renault Espace, Toyota Rav4, etc.). The space thus defined features four clearly differentiated groups of vehicles according to their fuel consumption and their dangerousness. At the top, we find vehicles that are quite polluting because they are old (right-hand quadrant) or particularly powerful and dangerous (left-hand quadrant) as against the relatively less polluting models in the lower half of the graph. At the bottom of the graph the same division as before is repeated: with relatively unsafe newer vehicles on the left and older and smaller cars on the right.

36The representation of the additional variables concerning the socio-occupational class and sex of drivers in the previous graphic clarifies the similarities between the characteristics of the vehicles and those of their drivers (Figure 2)

Figure 2

Representation of the socio-occupational class and sex of drivers—supplementary variables—in axes 1 and 2 of correspondence analysis

Representation of the socio-occupational class and sex of drivers—supplementary variables—in axes 1 and 2 of correspondence analysis

37While men’s cars are found in the most dangerous and most polluting quadrant, women’s vehicles occupy positions in the opposite quadrant, that of small-engined cars that have relatively lower emissions and lower mileage. Managers and workers also occupy almost polar opposite positions; powerful new vehicles on one side, that are both dangerous and less polluting, and smaller-engined vehicles that are older and thus more polluting on the other. A final opposition, which effectively reproduces that between male and female drivers, concerns the self-employed versus the most feminised occupations (employees, middle management in the public sector); while the former are found to be driving vehicles that are both powerful and polluting, the latter have vehicles that one might describe as “town cars” (i.e. small cars, [translator’s note.])” This space helps to differentiate the relationship to environmental and road hazards and contrasts a car population subject to the risks of pollution and vulnerability, mainly due to the age of the vehicles being driven (northeast quadrant), with a car population in which pollution and dangerousness are relatively limited (southwest quadrant). It remains to be understood how such characteristics connect with the intensity of vehicle usage.

Typology of vehicles according to their fuel economy and dangerousness

38Correspondence analysis, despite the number and diversity of models (over 500 different models are listed in the survey), provides some measure of comparability between vehicles. It makes it possible to identify some relatively simple structuring principles of automotive space. However, the analysis generates a greater number of dimensions, which can be exploited by means of an inductive classification of models based on where these vehicles fall on these different dimensions. This means that behind the alleged incommensurability of models claimed by car manufacturers [28], there are vehicles which, marque and version aside, actually strongly resemble each other from the point of view of some of their technical characteristics. This inductive typology also allows us to escape manufacturers’ own rankings and propose a more “intelligent” typology, based on specific criteria (Boltanski 1970).

39From the coordinates of the different car models on the first five axes of the MCA [29], a hierarchical clustering is performed (CAH—classification ascendante hiérarchique). [30] The most refined typology identifies six classes, whose technical characteristics and owners are discussed in the following paragraphs and detailed in Tables 4 and 5.

Table 4

Descriptions of technical characteristics of each cluster

1 2 3 4 5 6 Frequency (in %) 30 29 16 14 6 5 Taxable horsepower 4.9 6.6 6.2 6.9 4.8 9.5 Weight in kg 976 1,272 1,054 1,389 1,066 1,793 Consumption per 100 km 6.2 6.9 7.2 6.7 5.9 9 Annual kilometrage 11,021 13,894 7,920 16,759 11,669 16,057 Vehicles bought as new (%) 43 38 20 56 78 39 Diesel vehicles (%) 32 63 48 81 43 85 French vehicles (%) 31 41 64 26 67 25 German vehicles (%) 6 1 6 33 8 29 Most frequent model Clio 306 205 407 C3 A6

Descriptions of technical characteristics of each cluster

Field: All models with over 15 occurrences in the household car fleet.

Table 5

Social distribution of vehicle clusters according to socio-occupational class of the owner

1 2 3 4 5 6 Farmers 32 33 16 10 8 1 Artisans & other self-employed 19 28 12 21 15 5 Public sector managers 26 32 5 22 10 6 Private sector managers 31 27 12 14 7 9 Intermediate occupations public sector 36 27 11 13 5 7 Intermediate occupations private sector 31 30 14 15 5 5 Public sector employees 35 29 15 12 4 4 Private sector employees 36 28 14 12 5 4 Skilled workers 24 32 26 11 6 2 Unskilled workers 28 30 27 9 3 2 All 30 29 16 14 6 5

Social distribution of vehicle clusters according to socio-occupational class of the owner

Field: All persons owning a car whose characteristics are part of the survey,
Note: In 2008, of 100 personal vehicles owned by farmers, 11 are in Class 1.

40The first segment, about 30% of all models in the typology, includes smallengined vehicles that have been in circulation for an average of 8 years. These are light and relatively less polluting vehicles. Their mileage is relatively moderate. They are mainly small cars of the late 1990s: the Peugeot 106 and 206, and the Renault Twingo and Clio form a significant proportion of the models surveyed. These are cars that are characterized by low dangerousness, and also driven less. They are mainly (60%) owned by women, who are often in lower grade occupations (e.g. manual workers and employees).

41The second cluster of vehicles, which account for about 29% of the car fleet described by the typology, includes larger-engined and heavier vehicles with an average age quite similar to that of the previous class. These vehicles have relatively higher mileage, and poorer energy performance. They are mid-size models of the late 1990s, often purchased as used vehicles. They include the Peugeot 306 and 406, Renault Mégane and Scenic and the Citroën Xantia. The socio-demographic profile of their drivers is quite similar to that of the previous drivers (aside from the fact that managers are under-represented) and men are in the majority.

42The third segment (16% of the models surveyed) contains vehicles of fairly average power, weight and energy consumption, that are distinguished mainly by being very old vehicles. The poor energy efficiency of such vehicles can be explained less by their engine size than by their age. These vehicles are driven less, but they also offer limited protection and have relatively few safety features. Models of the early 1990s, if not earlier, they typically comprise cars such as the Peugeot 205, Citroën AX and Renault 21. Two-thirds of their owners are men and 40% of these are manual workers and farmers.

43The fourth segment (14%) contains vehicles that have larger engines and tend to be newer. These are fairly heavy vehicles, with poor energy performance relative to their recent entry into the market. Often foreign-made, and for the most part German, their weight makes them rather dangerous vehicles, but their average fuel consumption means they are fairly low-emission vehicles; however, they are widely used. These are mid-range vehicles of the mid 2000s: Audi A3 and A4, Volkswagen Passat, Peugeot 407 and 607. 60% are owned by men and public-sector managers are over-represented amongst them.

44The fifth segment contains quite recent small cars (5%). Equipped with all the safety equipment, these are not very dangerous vehicles (their weight is low) and have the lowest average consumption. These are small-engined cars of the early 2000s, and they typically include the Peugeot 107 and 207, Citroën C1 and C3. These vehicles have a relatively low average annual kilometrage. Predominantly owned by women, such cars frequently belong to public sector managers or craftswomen and shopkeepers.

45The last segment includes the most powerful vehicles, These are heavy vehicles, and they range from family cars (MPVs) top of the range vehicles (Mercedes E Class, Audi A6) or SUVs (Sport Utility Vehicles). [31] They are distinguished by very high consumption, and greater power and weight than cars in other classes; they combine poor energy performance and high dangerousness with intensive use. They are often owned by men, who frequently belong to the most highly qualified fringes of the salaried classes and the self-employed. [32]

46This typology reveals several social rationales underlying the choice of a car. A high level of pollution can either derive from constraints imposed by need (cars of classes 2 and 3), or from an expensive choice that is not, prima facie, financially constrained (classes 4 and 6). These last two classes of vehicles are those that have the highest frequencies of cars using diesel fuel: this is the case for over 80% of car classes 4 and 6.

47The danger dimension helps to draw a fairly clear line between the lower classes, who have rather less protective and not very dangerous vehicles, and the more privileged classes, a similar opposition to that presented by sex. In other words, a taste based on need here seems to contrast with a taste for freedom, although some classes of vehicle are relatively well distributed throughout the social space. However, vehicles belonging to the smallest clusters are those of the wealthiest classes. More subtle differentiations emerge, finally; the taste of public sector managers for fairly new small-engined cars (class 4), is a reminder of the “aristocratic asceticism” Bourdieu identified among teachers, that is the opposite of the taste for big cars, both quite recent and a little older (classes 3 and 6): “the aristocratic asceticism of teachers (and public sector managers) […] who systematically migrate towards the least expensive and the most austere forms of leisure […] is the opposite of the luxury tastes of professional people who collect the most expensive forms of consumption (culturally and/or economically) and the most prestigious ones, […] attendance at concerts […] possession of pianos, art books, […] of foreign cars […].” (Bourdieu 1979, p. 325).

48Moreover, even if this typology distinguishes groups of vehicles in terms of highly differentiated levels of energy performance, dangerousness and safety protection, is it still necessary to assess the variation in levels of pollution and dangerousness of vehicles according to their annual kilometrage. Do the disparities associated with uses of these different categories of vehicles reinforce or qualify the contrasts revealed by the classification?

49To account for the intensity of vehicle usage, in the following paragraphs the level of dangerousness of a vehicle is defined as a variable taking into account its weight, taxable horsepower and annual kilometrage; as a contributor to pollution, it is simply the product of its average consumption multiplied by its annual mileage. [33] We find that the vehicle classes thus formed are empirically distinct in terms of energy consumption levels and dangerousness (Figure 3). If the relationship between level of risk and consumption is relatively linear, this is not the case for the link between dangerousness and protection level (Figure 4). In this case, technical progress makes a distinction between different groups of vehicles. These analyses in terms of dangerousness and pollution qualify the results presented in the second part of this article: the vehicles that are apparently the most polluting, mainly because of their age, in class 3, in which manual workers are over-represented, are actually driven very little and thus de facto contribute in a very small degree to the actual level of pollution. By contrast the most powerful and polluting vehicles in classes 4 and 6, owned more frequently by managers, are also driven more intensively thus further contributing to road and environmental risk. A double paradox is thus confirmed here: firstly, the groups most sensitive to environmental risk are those which contribute most significantly to it; secondly, these same groups, who use their cars a great deal, contribute very little to the lack of road safety, mainly because their own vehicles are relatively safe.

Figure 3

Distribution of the pollution and dangerousness levels of different classes of vehicles

Distribution of the pollution and dangerousness levels of different classes of vehicles

Figure 4

Distribution of the dangerousness and protection levels of different classes of vehicles

Distribution of the dangerousness and protection levels of different classes of vehicles

50The HCA does not, however, make it possible to separate the effects of different factors in choosing a particular model class. The effects of income, sociooccupational class and sex are inseparable here. The effect of income appears to be particularly strong and might well account for the difference between between the group of rather old cars (Classes 1, 2 and 3) and that of the more recent (classes 4, 5 and 6). However, for vehicles in similar price ranges, energy performance and dangerousness are clearly differentiated; it is therefore necessary in what follows to isolate the specific effects of other socio-demographic variables and understand how they are related.

Social factors of car features: a choice model

51This final section presents the results of two multinomial logistic models (see Box 2 for the methods and details of the models used and Table 6 for parameter estimation) estimating the probability of belonging to one of the three classes of the most dangerous and polluting models rather than to the other three classes. Classes were obtained by hierarchical clustering as presented in the previous section.

Table 6

Parameter estimation of multinomial logistic models Probability of owning a vehicle belonging to classes 2, 4 and 6 for 25% of the most active drivers and 25% of the least active drivers

Reference variable Classes 1, 3, 5 First quartile of kilometrage Last quartile of kilometrage Class 2 Class 4 Class 6 Class 2 Class 4 Class 6 Variable of reference Active variable Coef. Marginal effect % p Coef. Marginal effect % p Coef. Marginal effect % p Coef. Marginal effect % p Coef. Marginal effect % p Coef. Marginal effect % p Constant –0.84 30.3 *** –3.18 4.0 *** –2.21 9.9 *** –0.87 29.5 *** –2.12 10.7 *** –3.32 3.5 *** Sex Man Woman 0.11 2.3 *** –0.67 -1.9 *** –0.74 –4.9 *** 0.09 1.9 *** –0.21 –1.8 ** –0.35 –1.0 *** Income 2nd quintile –0.02 ns 0.32 ns –0.33 ns 0.19 ns 0.40 4.5 ** –0.29 ns First quintile 3rd quintile 0.1948 ns 0.2942 ns 0.1789 ns 0.1993 ns 0.4718 5.4 *** –0.0415 ns 4th quintile 0.0621 1.3 * 0.6628 3.5 *** –0.0996 ns 0.332 ns 0.7774 10.0 *** –0.1159 ns 5th quintile –0.4058 ns 0.8089 4.6 *** 0.2903 ns 1.0187 24.2 *** 1.0501 14.8 *** 0.1882 ns PCS Farmers 0.1395 ns 0.0996 ns –0.1512 ns 0.6871 15.9 *** 0.6632 ns 2.3108 23.3 *** Skilled workers Artisans, other selfemployed –0.4693 ns 0.8082 ns 1.0259 13.6 ** 0.0435 0.9 *** 1.0004 13.9 *** 2.3872 24.8 *** Private sector managers –0.0409 ns 0.8746 5.1 * 0.4912 ns –0.00868 –0.2 ** 0.7955 10.3 *** 1.2475 7.7 *** Public sector managers 0.2243 ns 0.6913 ns –0.7272 –4.9 * 0.4644 ns 0.00205 ns 1.1177 6.5 *** Private sector intermediate occupations 0.0135 ns –0.2926 ns 0.0981 ns 0.2467 ns 0.2796 ns 0.5505 ns Public sector intermediate occupations –0.377 ns 0.6792 ns –0.9322 ns 0.2149 ns –0.1609 ns 0.6712 3.1 *

Reference variable Classes 1, 3, 5 First quartile of kilometrage Last quartile of kilometrage Class 2 Class 4 Class 6 Class 2 Class 4 Class 6 Private sector employees 0.2547 ns 0.713 ns 0.4774 ns 0.3734 ns 0.1721 ns 0.4107 ns Public sector employees –0.1836 ns 0.6887 ns –0.00439 ns 0.2688 5.9 ** 0.1947 ns 0.9766 5.3 ** Unskilled workers –0.0476 ns 0.2711 ns 0.3229 ns 0.3988 ns –0.0341 ns 0.6913 3.3 * Retired –0.292 ns 0.247 ns –0.1833 ns 0.2615 5.7 ** 0.4384 5.0 ** 0.6251 ns Inactive –0.0109 ns –0.3119 ns 0.1358 ns –0.0132 ns 0.192 ns 1.5792 11.5 *** Stratum Rural area –0.0771 ns 0.211 ns –0.4277 –3.2 * –0.00178 ns –0.00865 ns 0.0369 ns UU > 20,000 UU < 20,000 0.2278 ns –0.0328 ns 0.1247 ns 0.0126 ns –0.0278 ns –0.2439 ns Greater Paris 0.1616 3.5 ** 0.4105 1.9 ** –0.1166 ns –0.00276 ns –0.3168 –2.7 *** 0.2756 1.1 * Inner Paris Suburbs 0.2935 ns 0.471 2.3 ** 0.245 ns 0.1666 ns –0.5452 –4.2 *** –0.1832 ns City of Paris –0.00364 ns 0.5715 2.9 ** 0.0886 ns 0.0117 ns –0.3324 ns –0.00901 ns Age 0 ns 0.000414 ns –0.0102 ns 0 0.0 *** 0.0174 0.2 *** 0.0239 0.1 ***

Parameter estimation of multinomial logistic models Probability of owning a vehicle belonging to classes 2, 4 and 6 for 25% of the most active drivers and 25% of the least active drivers

Box 2.—The multinomial logistic regression model being used Classical multinomial logit model

Classic multimonial logit model
Let {1, …, k} be the terms of the dependent variable Y and X = (1, X₁, …, X_p) the p explanatory variables. It seeks to model the probabilities P(Y = j |X = x) for j = 1, …, k – 1; group k is taken as reference. The model is:

Thus:

Specification of models
For the first and last quartiles of drivers, six models were estimated successively from which are measured the degradation of the quality of fit, respectively from the log-likelihood (L and L’), the number of degrees of freedom (DOF and DDL’) of the full model and nested models (Table 7). The ratio presented in the last column of the table makes it possible to understand the extent of the contribution of each variable to the quality of the fit. The six models are:
1. Full model

2. Model without sex

3. Model without income

4. Model without occupation (PCS)

5. Model without stratum

6. Model without age

Table 7

Fit of regression models

L L’ DDL DDL’ L’–L DDL–DDL’ (L’–L)/ (DDL–DDL’) Model 1 7,739 66 Model 2 7,817 63 78 3 26 Model 3 7,766 54 27 12 2.25 Model 4 7,800 33 61 33 1.84848485 Model 5 7,761 51 22 15 1.46666667 Model 6 7,783 63 44 3 14.6666667 L L’ DDL DDL’ L’–L DDL–DDL’ (L’–L)/ (DDL–DDL’) Model 1 11,381 66 Model 2 11,401 63 20 3 6.66666667 Model 3 11,435 54 54 12 4.5 Model 4 11,530 33 149 33 4.51515152 Model 5 11,408 51 27 15 1.8 Model 6 11,472 63 91 3 30.3333333

Fit of regression models

Note: Above, the regression models have as their field 25% of the least intensive drivers; below, 25% of the most intensive drivers.

52This is used to understand the specific effects of various socio-demographic characteristics of drivers on the choice of car model, in terms of energy performance and dangerousness. Five variables are introduced here; with regard to the owner, the variables of sex, age and socio-occupational group in twelve indices. At the household level, place of residence and income level per household consumption unit are used. To understand how such determinants connect with the intensity of use of the vehicle, we carried out regression models in two separate fields, one consisting of the 25% of vehicles being used the least and the other 25% of those being used the most.

53The explanatory variables use the baseline of a man, a skilled worker in the first quintile of income distribution living in an urban unit of more than 20,000 inhabitants, outside the Paris area. Regarding the dependent variable, possession of a vehicle of type 1, 3 or 5, i.e. the least dangerous and least polluting models, is taken as a baseline. The estimated model thus leads to an understanding of the distancing factors to owning a “town” car.

54We compare the closeness of fit of different models compared to the full model using the following approach. From the full model, five different models are created, from each of which a different single explanatory variable is eliminated. This is done so that we can rank the relative sizes of the specific effects of different variables. Choosing a dangerous and polluting car seems initially to be driven by income differentiation. The deterioration of the quality of fit between models 2 and 3, for the two regression models, shows the importance of this effect in the ownership of a car belonging to one of the three classes of the most dangerous and polluting vehicles (2, 4 and 6). Owning a model belonging to one of these three classes rather than a vehicle of Classes 1, 3 and 5, is in fact largely related to the level of household income, and membership in the bottom quintile of income distribution has a strong effect on the probability of owning a vehicle of one of these three classes; the specific effect of income appears particularly significant for the more powerful engined cars in cluster 4. The price elasticity of demand for fuel is indeed much lower for the most affluent households than for others (Calvet and Marical 2011); in addition to the distinctive rationale of features for their own sake, there is a distinctive rationale of intensive consumption. This dimension is related to the classic analyses of Theodore Veblen on conspicuous consumption ([1899] 1979), demonstrating that consumption is all about how to consume. [34]

55Comparing the two models also serves to unravel what relates to either the intensity of use (which can also be linked to strong geographical constraints), or to a symbolic rationale of distinction. When one looks more precisely at the effects of socio-occupational group, we find that the self-employed (and those in private sector management), whether they belong to the group of the most intensive drivers, or instead the group of the less frequent drivers, will more often choose vehicles in cluster 6 (for the self-employed) or in cluster 4 (for private sector management). In other words, two rationales are emerging: owning a vehicle that is amongst the most dangerous and most polluting can be understood, firstly, as being constrained by the type of use (those who drive the most have the most expensive vehicles, environmentally and in human terms).

56Secondly, among those who drive the least, all other things being equal, only the self-employed (excluding farmers) and private sector managers choose, despite this modest use, vehicles that produce the most externalities. Such distinctive rationales therefore have quite material effects.

57Furthermore, the regression models also show the importance of specifically social factors: firstly age and secondly sex. The age parameter is always positive and would seem to indicate that ownership of a more dangerous and polluting vehicle will increase with age, all things being equal. This effect, often less significant than income or sex, is difficult to interpret for two reasons; firstly, because it could conceal a generational effect; secondly, because age may not have a linear effect, as is claimed for other dimensions of car consumption, such as the type of engine (Demoli 2013b), the intensity of use (and Druhle Pervanchon 2004), or the use of car-sharing (Vincent 2008).

58The negative value of the sex parameter shows that, all other things being equal, women are less likely to own a dangerous and polluting vehicle, rather than a vehicle of type 1, 3 or 5. The high value of such parameters, on the one hand, and the significant deterioration in the quality of the fit between the full model and model 2, on the other hand, are good evidence of such a relationship. All things being equal, women are less likely than men to own a vehicle belonging to the lowest performing and most dangerous classes, and particularly vehicles of classes 4 and 6. The only exception is that women seem to be more likely to be found driving vehicles of class 2: whether they belong to the class of most intense drivers, or, conversely, the least regular, they more often own vehicles of this class. Such vehicles however are nonetheless the least energy-intensive and the least dangerous of the three classes in question. Furthermore, although they are poor in terms of fuel economy this is essentially because of their age (these vehicles are likely to have been purchased used) and the very low representation of German marques suggests they are rarely higher category vehicles. Such models, often old family vehicles, seem to be polluting and dangerous cars not so much because of any symbolic rationale as because of material constraint. It is interesting that a reading of La Distinction (Bourdieu 1979) in terms of material goods shows marked differences between men and women.

59* * *

60The choice of a car, as we have shown, is not a trivial matter, as it involves both driver safety and that of other road users, and produces a collective cost to the environment. This choice, which at first sight seems to be the consumer’s private concern, in fact involves social groups in ways that are not randomly located in social space and are informed, on the contrary, by systematic and divided preferences.

61Such rationales of differentiation of social groups are not merely symbolic and have completely material impulses and effects. A thorough study of the dangerousness of different car models shows that the danger is not randomly located; the rationales at work are not just symbolic, as Boltanski argued, when road hazards appear to be systematically asymmetrical. Similarly, the polluting capacity of vehicles is not homogeneously distributed in social space. Choosing an ecologically inexpensive car is not necessarily only to be found amongst those social groups traditionally sensitive to environmental values. This leads to the conclusion that the car does not seem to be one of the objects upon which conversion to environmentally friendly practices is practised.

62The most polluting models, which are the most protective but also the most dangerous to others, are owned by the social categories with the highest volume of capital. Conversely, the vehicles of the working classes and of the fractions most endowed with cultural capital tend to display a certain asceticism on the part of their owner, the details of which are yet to be explored. For some, choosing a relatively safe and environmentally efficient car could belong to a rationale of necessity and, for others, to a taste for asceticism. The car thus demonstrates differentiated relationships to risks, the social rationales for which deviate from the traditional analysis in the literature on the subject. Generally low contributors to pollution, the lower social categories, however, are widely subject to road risk, in particular because of the low protective power of their vehicles. Conversely, the most advantaged social categories contribute significantly to environmental risk, even though they own cars that are safe and of high quality. Other rationales however tend to blur such a schema. Firstly, some lower social class households that are particularly dependent on the car, especially in rural areas, make intensive use of diesel cars; secondly, new mobility practices have found an important echo among the most affluent social classes, living in locations where alternative forms of mobility are possible (Vincent 2008).

63However, the indirect costs of the car—pollution, road safety—induced and sustained by different social groups recall overall the unequal distribution of the direct costs that the car involves, a phenomenon highlighted by the work of Julie Froud and her colleagues (2005) for Great Britain. Assessing the budgetary costs devoted to the automobile, the authors have shown that there are three rationales for automotive expenses, closely linked to the characteristics of the car fleet concerned: a choice rationale (characterized by quality equipment and low budget constraint) a constraint rationale (with relatively old equipment and an average budget constraint) and a risk rationale (characterized by an aging fleet, subject to costly repairs, making automotive expenses uncertain and particularly high). The car therefore loads unequal direct and indirect costs on to the different social groups, reminding us how car mobility is defined as a particularly powerful and systematic constraint (Urry 2004). Thus, far from being a symbolic benefit of mass consumption, which should have been able to “erode social, sexual, geographic or age barriers” (Yonnet 1984, p. 148), the car seems far more to crystallize several highly similar types of inequality.

64This analysis would benefit from further study in two respects. First, an analysis of the rationales for the choice of vehicles for other survey dates would allow us to understand how polluting and protective characteristics change over time and, in particular, according to the many public policies that aim to improve the safety and energy performance of vehicles. More specifically, such studies would help us to understand how different social groups react to tax incentive policies for moderation in consumption and regulatory policies concerning road safety. Then, in the same way that recent studies have shown that the intensity of car use embraces very differentiated rationales—Vincent Kaufmann, for example, shows that the “wide” forms of mobility, quite specific to the most wealthy classes are often constrained (Kaufmann, et al 2014)—we need to better understand the social rationales behind car use, captured here by a rather crude criterion, the overall volume of use. Such an analysis would then allow us to relate the contributions of different social groups to automotive safety and pollution more precisely to the rationales for the use of cars.