1. Introduction
In 2016, transport generated ca. 8 gigatonnes of CO
2 globally, with road transport accounting for 74% of these emissions, followed by air- (12%) and waterborne transport (11%) [
1]. To be on a sustainable path, the transport system needs to reduce substantially the level of air pollution and greenhouse gas (GHG) emissions it annually emits. The deployment of zero-emission (tailpipe) vehicle (ZEV) technology (battery electric vehicles—BEVs) and fuel cell electric vehicles (FCEVs)) and low-emission vehicles (plug-in hybrid electric vehicle—PHEVs)) is a step towards achieving this goal. The number of electric vehicles (EVs; BEVs plus PHEVs) in use worldwide exceeded five million in 2018 [
2]. The focus of this study is on the European Union (EU) road transport, particularly, on passenger cars. Road transport remains a major source of air pollution in Europe [
3]. Cars account for almost 44% of transport GHG emissions in the EU [
4]. In this context, an investigation into the drivers which lead to greater electric and fuel cell car sales remains an important topic of research.
The objective of our work is to provide evidence on and quantify the factors that influence the European market for zero- and low-emission car technologies. This paper builds upon Gómez Vilchez et al. [
5] and Rohr et al. [
6]. Whereas [
5] focused on describing the survey (see
Section 3.1) and illustrating the representativeness of the sample; [
6] introduced the statistical model estimated and willingness-to-pay values. Thus, the present article complements these papers and deviates from them by showing remaining results that were not reported previously. Specifically, the main aim of this paper is to highlight statistical differences based on socio-economic characteristics as well as to present country-specific results on the most important factors in car choices, with a focus on government incentives and the role of payment options and depreciation.
Due to the variability of consumers’ socio-economic characteristics in each of the surveyed countries, the effectiveness of financial incentives may differ across Europe, as highlighted in the conclusions.
The structure of the paper is as follows: the existing literature is briefly reflected upon in
Section 2;
Section 3 describes the survey and the type of choice analysis undertaken; in
Section 4, the results are reported and discussed; and finally, conclusions are drawn in
Section 5.
2. Literature
The existing literature on vehicle type choice analysis is vast and continues to expand. The factors influencing vehicle choice in general and EVs in particular can be studied using empirical and theoretical models. An example of the latter is provided by Tu and Yang [
7], who attempted to combine three pieces of theory: theory of planned behavior, technology acceptance model and innovation diffusion theory.
Table 1 shows the list of such factors identified in three studies: Mueller and de Haan [
8], Jensen et al. [
9] and Struben and Sterman [
10]. The interest is in the vehicle attributes, not in the demographic and socio-economic characteristics of consumers (see
Section 4). As can be seen, in all the three studies the purchase price, operating cost and emissions impact are included.
In terms of methods, Mueller and de Haan [
8] applied agent-based simulation (ABS), Jensen et al. [
9] used a discrete choice model (DCM), and Struben and Sterman [
10] relied on system dynamics (SD). The focus of each method is different: ABS captures agent heterogeneity (Hamill and Gilbert [
11]), DCM stresses the observation of individual behaviour by means of discrete variables (Ben-Akiva and Lerman [
12]; for more recent treatments of DCMs, see e.g. Hensher et al. [
13] or Train [
14]) and SD highlights the role of feedback structure and dynamic behavior (Forrester [
15] and Sterman [
16]). In the context of modelling consumer preferences for alternative vehicle technologies, each method has a particular strength: ABS can account for consumer interactions, DCM can model innovations at early market stages and SD can represent system-wide effects (Jochem et al. [
17]). In Al-Alawi and Bradley [
18], discrete choice analysis was also identified as a main method to model vehicle technology (mainly powertrain) choice. In their review of SD models, Gómez Vilchez and Jochem [
19] found that the embedment of DCMs within such models is common practice. The use of disaggregate (multinomial (MNL) and nested MNL (NMNL)) models in EV market diffusion studies was identified by Gnann et al. [
20]. In their review of eleven models, Lopez-Arboleda et al. [
21] found that DCMs, especially MNL, were generally used to represent consumer purchases of electric passenger vehicles. According to these authors, Greene et al. [
22] and Kieckhäfer et al. [
23] applied NMNL. Whereas the former combined NMNL with computable general equilibrium modelling, the latter opted for a combination with ABS. Pure DCMs accounting for the determinants of vehicle choice vary widely but they tend to be based on regional (e.g. Bunch et al. [
24] for California) or national surveys (e.g. Hackbarth and Madlener [
25] for Germany and Batley et al. [
26] for the UK).
In contrast, the use of cross-national surveys to investigate the factors that influence EV choice is less commonly reported in the scientific literature. An early example is the survey by Thiel et al. [
27], which was carried out in 2012 (see
Section 4). However, the results by [
27] did not translate into a DCM.
More recently, Christidis and Focas [
28] conducted a cross-national survey to examine the propensity of EU respondents to purchase a hybrid or BEV, finding that propensity was higher for Spain and Italy and rather low for Poland and France. Instead of developing DCMs, they applied machine learning techniques. This method has also been applied in a few travel studies. In the context of travel mode choice studies, Hagenauer and Helbich [
29] found that machine learning classifiers, though not widely used in such studies, outperformed MNL. For a hybrid model that combines neural networks and DCM while being based on a SP survey on travel mode choice, see Sifringer et al. [
30]. More recently, van Cranenburgh and Alwosheel [
31] asserted that neural networks help investigate heterogeneity in traveller’s decision rules. Christidis and Focas [
28] concluded that between 2014 and 2018 propensity increased across all socio-economic groups and highlighted the relevance of policy measures at the local level.
The survey reported in the present paper is, to a certain extent, a follow-up of the survey by Thiel et al. [
27]. The paper contributes to the existing literature by reporting the salient results of a cross-national stated preference survey, as opposed to regional or national ones, and by quantifying through a DCM the factors that determine consumers’ choice of electric cars in Europe, with an emphasis on the purchase price.
3. Methodological Approach
The methodology applied in this work is illustrated in a stylized manner in
Figure 1. First, car owners in six European countries (France (FR), Germany (DE), Italy (IT), Poland (PL), Spain (ES) and the United Kingdom (UK)) were contacted. 1,248 of them completed the computer-based web interviewing exercise in mid-2017. The questionnaire to this survey comprised two stated choice experiments, alongside other question. The attributes and levels considered in these experiments as well as a sample of the choice set-up can be found in Rohr et al. [
6]. The stated choice experiments provided information from which DCMs were estimated using ALOGIT
® (see [
32]).
3.1. Stated Preference Survey
Income and access to recharging infrastructure were found by Mersky et al. [
33] to have the largest power to predict EV registrations.
Figure 2 and
Table 2 show the distribution of survey respondents by income level and other socio-economic characteristics such as location.
Respondents provided an estimate of their household’s combined yearly income (before tax), in the currency of their country (see
Figure 2). The sample was divided into eight income range levels: ≤ €11,999, €12,000–€17,999, €18,000–€29,999, €30,000–€39,999, €40,000–€59,999, €60,000–€89,000, ≥ €90,000. Respondents were also given the option not to declare their income. This option was chosen by a sizeable proportion of Polish (27%) and German (19%) respondents. The German and UK sample illustrated a larger proportion of their respective samples with higher incomes. In contrast, Poland exhibited a tendency for income levels towards the lower end of the spectrum. Almost one-fifth of the sample, excluding Poland, reported an income level of €30,000–€39,999. Over 27% of the Italian and French respondents indicated a level of income in the €18,000–€29,999 category. Around 40% of the German and UK respondents declared an income level of €30,000–€59,999. Clusters of high-income urban dwellers and low-income rural dwellers could not be identified clearly in the samples.
The distribution of respondents by gender was evenly split across the countries (see
Table 2). The age profiles of respondents by country were constrained by quotas. About a quarter of the sample is between 18 and 34 years of age, although this is slightly higher for the sample from Poland and the UK. Respondents who declared not to own a car were excluded from the survey. Thus the survey is not representative of the whole population. Instead, it exhibits a reasonable degree of representativeness of car-owning households (see Section 2.2 in [
5] for a discussion on sample representativeness, where data of the countries’ populations and survey samples are compared).
In terms of education, the sample from France, Poland and Spain contained higher proportions of respondents who only had primary school education (or less). Nearly half of the sample from Italy had some secondary school education only, compared to between 10 and 20% for the other countries. The sample from Italy also had much lower proportions who completed secondary school or were university or college graduates.
Overall, nearly 40% of the respondents lived in urban areas with a population greater than 200,000. 11.6% lived in large urban areas with a population of more than 1 million. The sample from Spain shows a higher proportion of respondents living in urban areas, compared to the other countries. Almost one third of all respondents lived in rural areas, with higher proportions of respondents in France and Italy.
Furthermore, the questionnaire asked how the respondent would most likely purchase the car. Three payment options were provided: cash or personal loan, hire purchase (HP) and personal contract purchase (PCP) (for the questionnaire template, which contains the actual questions and descriptions of these options, see Annex 1 in [
5]).
3.2. Statistical Tests on the Effect of Socio-Economic Characteristics
A variety of statistical tests and indicators were employed to test for significant differences between selected socio-economic (gender, age, education and income level) groupings within each of the countries surveyed. A selection of statistical test results are summarized in
Table 3. Full detailed statistical tests are available upon request from the authors.
Car size may play a role in powertrain choice. For instance, fuel cell cars powered by hydrogen are currently available in the European market only in the large-size segment. As can be seen, the effect of income and gender on car size turned out to be statistically significant for Germany, Italy and Spain (in the latter, the Kruskal-Wallis test also confirmed statistically significant results for age, as in the French and Polish samples). With regards to the effect of respondents’ socio-economic characteristics on the payment option, no statistically significant impact was found for Germany and Poland. In contrast, this effect was found for gender and age for France and Spain (in the latter, income also turned out to be statistically significant). The effect of education was found to be statistically significant for France (on payment option) and Poland (on car size). No statistically significant results were obtained in most of the UK tests.
3.3. The Estimated Model
The model was specified with a ‘low-emissions’ nest, which contains hybrid and zero-emission cars. This assumption was empirically tested and a nesting parameter (
θlow-emissions) equal to 0.613 was obtained. To estimate the utility coefficients, the data from the two choice experiments were pooled. For robustness, a bootstrap technique [
34] was applied. The bootstrap was used to address the possible correlation issue across the multiple observations from the choice experiments. The preferred model (NMNL) is shown in
Table 4.
As can be seen, the estimated utility coefficients for purchase price vary by car size. Sensitivity to purchase price is greater for consumers intending to pay via HP or PCP. Furthermore, a key inter-country difference is for the variable ‘operating costs’, which differ among France, Italy and the rest of the surveyed countries. These two vehicle attributes, together with refueling time, have a negative impact on purchase behaviour, as expected. In contrast, driving range positively influences purchase behaviour. For a quantification of relative values, see Rohr et al. [
6].
5. Conclusions and Further Research
The main conclusion of this study is that, while the purchase price remains the crucial factor, country-specific socio-economic characteristics of consumers influence their intention to purchase an electric car in Europe. The evidence presented in this paper reaffirms the argument that incentives, in particular government financial interventions, are likely to be critical in accelerating widespread uptake of ZEVs, as electric car prices are still perceived by European consumers to be too high. This is also consistent with experience from outside the EU (for instance, incentives have been in place since 1990 in Norway [
44], the world’s leading market in electric car sales [
2]). Government incentives to buy an electric car were generally seen positively by the surveyed car owners. However, the number of respondents stating that such incentives were bad for the market increased from 2012 to 2017. However, the effectiveness of government incentives may vary across European countries as a result of variations in the socio-economic characteristics of consumers and consumers’ level of awareness of payment options in the different countries.
We further conclude that stated preference surveys, in combination with discrete choice analysis, remains a useful source of information to investigate the market for electric cars. The usefulness of this type of survey for the EV market, however, may be declining over time as diffusion for this technology is proceeding. Hence, greater reliance on revealed preference surveys (with ideally a larger sample) will become necessary.
The main limitation of this study is that it addresses consumer behavior, which is only one side of the market. As highlighted by Figenbaum [
44], the effect of incentives was felt in Norway when manufacturers launched (lithium-ion) BEVs. On the supply side, factors that encourage electric car model offering by the manufacturers and technology-push policies are also important (see e.g. the evidence provided by Zarazua de Rubens et al. [
45]). Some of these factors and policies could be explored with other tools, such as the Powertrain Technology Transition Market Agent Model (PTTMAM; available at:
https://ec.europa.eu/jrc/en/pttmam) grounded on the SD method (for a recent application of this model to simulate the effect of purchase incentives, see [
46]; for the model documentation, see Harrison et al. [
47]).
Further work is required to embed the quantitative results of the survey and corresponding model within a wider modelling framework that takes into account the drivers of change for vehicle attributes. An initial step in this direction, using the PTTMAM as such of a framework, was reported in [
48]. The difficulty of this approach, however, lies in the fact that the PTTMAM does not explicitly consider consumers’ socio-economic characteristics and their heterogeneity across populations. This is also the case for three of the SD models reviewed by Lopez-Arboleda et al. [
21] (see their
Table 4). Thus, there is an opportunity for further SD modelling work on detailed consumer characterization and disaggregation. Thanks to the open accessibility of PTTMAM through EU Public Licence and the evidence on consumer behaviour gathered in this study, the research community has two powerful tools to pursue this direction.