Does Environmental Awareness Fuel The Electric Vehicle Market A Twitter Keyword Analysis

Energy Economics 101 (2021) 105337
Contents lists available at ScienceDirect
Energy Economics
journal homepage: www.elsevier.com/locate/eneeco
Does environmental awareness fuel the electric vehicle market? A Twitter

keyword analysis
Leonhard M. Austmann, Samuel A. Vigne *
Trinity Business School, The University of Dublin, Luce Hall, Pearse St, Dublin 2 D02, H308, Ireland
A R T I C L E I N F O A B S T R A C T
JEL codes: Zero emission battery electric vehicles and plug-in hybrid electric vehicles have the ability to reduce CO2
C12 emissions and hence mitigate climate change. Consequently, understanding the macroeconomic drivers of sales
C23 has been of increasing interest in the past years. One factor that has frequently been investigated in relation to an
C58
individual's electric vehicle purchase intentions is his or her level of environmental awareness (EA). Previous
D91
E01
research analysing this question on a macroeconomic level is, however, very limited. Applying a novel method in
E61 this field in form of a Twitter keyword analysis, this study aims to analyse the effect of EA on the electric vehicle
H21 market in 27 member states of the European Union (EU) as well as two European Free Trade Association (EFTA)
L62 states. The results of the investigation show that EA does not significantly influence the electric vehicle market in
L94 the investigated time period, and instead indicate why other factors might currently play a more significant role.
O33 We contribute to the existing literature by introducing a new measurement method of EA to this research field
O52 and by contributing to the understanding of the impact of EA on electric vehicle registrations.
Q55
Keywords:
Electric vehicles
Energy
Green
Environmental awareness
Twitter
1. Introduction for almost 27% of the overall emissions in the EU. Diving further into it,
the largest share of CO2 emissions (44.3%) in the transportation sector
Climate change is today's biggest threat to the planet (IPCC, 2014). are emitted by passenger cars (European Environment Agency, 2019). In
Natural disasters, expected to continue to increase in frequency, are order to meet the climate goals set in the Paris agreement and to address
often assumed to be a direct result of climate change and not only claim the need for a global reduction of CO2 emissions, a focus on the private
lives but also cause severe economic damages (Coronese et al., 2019). As transportation sector becomes evident. To achieve the recently set target
a consequence, the international community has agreed upon several in the European Green Deal of becoming a CO2 neutral continent by
treaties and pledged to engage in environmental protection measures. 2050 (European Commission, 2019), innovative solutions in this sector
One of the most prominent treaties, established by the United Nations, are required.
which is ratified by 188 states and the EU is the Paris agreement (United One promising alternative to normal internal combustion cars often
Nations Climate Change, 2020). A main aspect of this treaty is to keep discussed in academic literature as well as in politics are electric vehicles
the increase in average temperature below 2C∘ compared to pre- (EVs). Incorporating the ability to either significantly reduce (plug-in
industrial levels. This can only be achieved by drastically reducing hybrid electric vehicle (PHEV)) or completely eliminate (battery electric
CO2 emissions. In the European Union the transportation sector is the vehicle (BEV)) the emission of green-house gases (GHG), these types of
second highest contributor to CO2 emissions, responsible for more than cars provide a potential solution to mitigate climate change. As a result,
1250 million tonnes of greenhouse gases in 2017 and therefore accounts countries worldwide are subsidising the purchase of EVs not only
* Corresponding author.
E-mail addresses: [email protected] (L.M. Austmann), [email protected] (S.A. Vigne).
https://doi.org/10.1016/j.eneco.2021.105337
Received 27 January 2021; Received in revised form 25 April 2021; Accepted 23 May 2021
Available online 26 May 2021
0140-9883/© 2021 Published by Elsevier B.V.
L.M. Austmann and S.A. Vigne Energy Economics 101 (2021) 105337
through monetary incentives but also through non-monetary incentives

such as access to bus lanes. While the effects of subsidies have been
investigated by different scholars and shown to be effective (e.g. Sier
zchula et al., 2014; Mersky et al., 2016; Münzel et al., 2019), they do not
offer a sustainable solution for the turnaround from internal combustion
cars to EVs. Paradoxically, several countries decreased the amount of
subsidies in recent years (Münzel et al., 2019); therefore calling on the
individual to accelerate the development of the EV market. As proposed
by the theory of planned behavior (TPB) (Ajzen, 1991), which is
frequently used as a framework to explain antecedents of purchasing
behavior, an individual's attitude can be a major driver in the decision
whether or not to purchase an EV. One attitudinal dimension that comes
to mind in relation to EVs is environmental awareness (EA). As
mentioned earlier, EVs are by many politicians and scholars seen as an
important step toward a more environment friendly transportation
sector. In addition, many car manufacturers such as Mercedes Benz
(Daimler, 2020) or Volkswagen (Volkswagen, 2020) are using sustain
ability and eco-friendliness as a key component of the brand they are
building around their EV lines. As branding literature suggests, con
sumers are more likely to purchase from a brand that reflects their
personal values and beliefs (Mitchell and Olson, 1981). This in turn
indicates that whether or not an individual is concerned about climate
change and sustainability, i.e. is environmentally aware, might play a
role in his or her purchasing decisions regarding EVs. The underlying
assumption would be that an increase in EA leads to more EV sales and Fig. 1. Drivers of the EV market.
hence more EV registrations. Note: Categorization based on Austmann (2020).
This paper aims to investigate whether EA has a significant influence
on the registrations of EVs to gain a better understanding of the sales − demographic factors, infrastructure & geography, energy prices, the
drivers in this market. Thereby, a novel method to measure EA in form of technological development of EVs, as well as psychological factors.
a Twitter keyword analysis is applied. This method distinguishes itself The first group of our categorization includes determinants dealing
from the predominantly used surveys in this field of research, as the with the automobile sector. Considering the local automobile sector in
results are not exposed to potential biases and exhibit macroeconomic the investigation is especially important when doing cross country
characteristics rather than individual characteristics. Furthermore, it studies to account for potential country differences. Examples for de
allows to address the gap in research that focuses on actual EV adoption. terminants included in the automobile sector category are vehicle fleet
While most studies dealing with EA focus on one country, the properties size or the country a global EV producer is located in (Münzel et al.,
of our method allow us to investigate the effects for almost the entire EU 2019; Sierzchula et al., 2014). The impact of the second category,
and EFTA from 2010 to 2017. Another advantage of the variable is that namely incentives, has been quite thoroughly investigated by several
it provides the ability to determine the marginal effects of EA on the studies, also on a macroeconomic level. Thereby, scholars usually
registration rate of EVs. distinguish between financial and non-financial incentives. While
The structure of the remaining paper is as follows. The second financial incentives are for example grants or tax reductions, non-
chapter of this study discusses the relevant literature. In the third financial incentives can for example be in form of special permissions
chapter we will describe our data collection and the methodological such as the usage of bus lanes (e.g. Mersky et al., 2016; Münzel et al.,
approach used to investigate the underlying research question. With the 2019). While the impact of socio-demographic determinants such as age
fourth chapter presenting the results of the statistical analysis and and gender are predominantly considered when data is collected by
robustness tests, a discussion of the results is provided in the fifth means of a survey (e.g. de Rubens, 2019; Ziegler, 2012), socio-economic
chapter. Chapter 5 also includes recommendations for future research as factors like average household disposable income are also more
well as the limitations of this study. The last chapter summarises our commonly analysed on a national economic level (e.g. Sierzchula et al.,
main findings and contributions. 2014; Münzel et al., 2019; Hirte and Tscharaktschiew, 2020). De
terminants from this category are often considered to account for
2. Literature review differing living conditions. One of the most widely applied determinants
in EV studies are charging stations and fuel prices, which can be
This chapter is dedicated to reviewing the relevant existing litera attributed to the categories infrastructure and energy prices, respec
ture. It is divided into three key pillars. First, we examine potential tively (e.g. Li et al., 2017; Sierzchula et al., 2014; Baur and Todorova,
determinants identified by previous literature. Thereafter, we critically 2018). Determinants focusing on specific characteristics of EVs such as
discuss the common measurement of EA in the EV market. The last driving range, maximum speed or price (Nie et al., 2018) are summar
section of this chapter reviews existing research that has used Twitter as ised under the category EV development. The last category determined
a measurement tool to reflect the public opinion, allowing us to based on previous studies deals with the beliefs and personality traits of
conclude with the hypothesis investigated in this study. people regarding EV market influencing aspects and are summarised as
psychological determinants. This category of variables is rarely inves
2.1. Drivers of the electric vehicle market tigated on macroeconomic level (Austmann, 2020) and will be discussed
in more detail in the subsequent chapter.
Previous literature has looked at a variety of determinants that could
potentially explain movements in the EV market. These determinants
can broadly be categorised into seven different groups, which are 2.2. EA and the electric vehicle market
depicted in Fig. 1. In particular, these determinants relate to the auto
mobile sector, incentives granted by politics, socio-economic/ As mentioned above, one category of determinants that researchers
2
assume to have an impact on the EV market are related to an individual's over time. As most studies only consider data that was collected at one
psychological attributes such as the person's lifestyle or level of inno point in time, insufficient conclusions can be derived about the devel
vativeness (Chu et al., 2019). The most commonly researched determi opment of the influence of EA on the motivation to purchase an EV.
nant in this area is, however, EA (Cohen et al., 2019). The impact of EA is The last seven articles in Table 1 are, to the best of our knowledge,
frequently explained with the help of the Theory of Planned Behavior the only studies that have applied some kind of EA measure and inves
(TPB) established by Ajzen (1991) (e.g. Tu and Yang, 2019; Wang and tigated the impact on actual EV market data. While these approaches are
Dong, 2016; Zhang et al., 2018). This theory hypothesises that an in good initial indicators, they also exhibit room for further improvement.
dividual's behavior is largely determined by its behavior intention which For example, it might be useful to establish a measure that captures EA
is in turn influenced by its attitude toward the behavior, perceived behav more continuously throughout a year and focuses on the perception of
ioral control and subjective norms (Ajzen, 1991). The latter can be un the population. Furthermore, it would be beneficial to apply a measure
derstood as the social pressure the individual expects to face when that captures both, the frequency in which the society is confronted with
engaging in the respective behavior. Attitudes toward the behavior, a topic and in which the topic is discussed by individuals in the society.
perceived behavioral control and subjective norms also exhibit inter These limitations suggest that it could be beneficial to investigate the
action effects.1 A comprehensive list of TPB related EV studies can be relationship between EA and EV sales at a macroeconomic level by
found in Zhang et al. (2018). To better illustrate some characteristics applying a more objective measure of EA that allows the investigation
that previous studies investigating the effect of EA on the EV market over a longer period of time and addresses the required improvements
have in common, we put together a table (1) of some of the core studies mentioned above. It should also be considered that EA is a psychological
in this research field. This table provides an overview of the findings and factor and therefore no direct measurement exists. Surrogate variables
approaches applied in this field. Furthermore, it helps determine for EA can however deviate in their level of representativeness and ac
possible limitations that this study tries to encounter. curacy. This can partially be caused by different definitions of EA or data
Firstly, researchers have found, that existing research analysing the availability. Different approaches can therefore capture different aspects
determining factors of the EV market, in general focus on intention to of EA. Consequently, it makes sense to elaborate different and innova
adopt EV rather than actual adoption behavior (Kumar and Alok, 2020). tive measures of EA to obtain a more diversified view on the influence of
One possible reason to survey people whether they intend to purchase an EA on the EV market. In the following section we will present an
EV instead of surveying people that actually own an EV is because of the alternative measurement approach for EA which is largely able to
small ownership ratio of this vehicle type and therefore the difficulty to overcome the described limitations.
obtain a sufficient sample size. Based on the intention to purchase an EV,
the researchers then draw conclusions on the EV market. This can 2.3. Twitter as an awareness barometer
however compromise the reliability of the results as Lane and Potter
(2007) for example uncovered the existence of an attitude-action gap in Different ways of measuring public awareness or opinion toward a
the EV market. Consequently, the intention of an individual does not certain topic have been applied in academic literature as well as prac
necessarily correspond with his or her actual behavior. The approach of tice. As previously discussed, the most frequently used method to asses
measuring intention has also been used in previous research when an individual's awareness are surveys. However, surveys are not the only
analysing the influence of EA on the EV market (Ziegler, 2012; Zhang way to capture the public's opinion. Another method to measure
et al., 2018). Thus, it is important to conduct studies that focus on EV awareness is to investigate the sources from which the general public
adoption behavior to ensure the reliability of the findings. obtains information. As the key function of mass media is to report to
Furthermore, as can be seen in Table 1, surveys are the most people an indirect awareness assessment is possible by looking at the
frequently used method for data collection. Thereby, the predominant publication frequency of the investigated topic (Olteanu et al., 2015).
survey methods researchers are using are either directly asking the re This inference can be made as scholars discovered a significant corre
spondents about their level of EA or whether EA is a decisive factor for lation between public's opinion and mass media coverage (Besova and
the purchase decision of an EV (e.g. Trommer et al., 2015) or indirectly Cooley, 2009). Consequently, newspaper coverage has been used either
deduce the level of awareness from questions regarding representative supplementary to surveys or as a sole indicator for awareness. Thereby,
behaviors, such as whether the individual actively recycles (e.g. Chu this method can be applied to a wide range of topics. The employment of
et al., 2019). newspapers as an opinion barometer for example ranges from topics
However, surveys are exposed to certain limitations. One such lim such as the interest in the OPEC (Plante, 2019) to economic uncertainty
itation is biases. Biases such as leading or social desirability bias can (Baker et al., 2016). Scholars, especially found interest in using news
significantly influence the answers of respondents and therefore papers as measures for EA (Sampei and Aoyagi-Usui, 2009; Olteanu
potentially compromise the reliability of a study (Smith et al., 2017). et al., 2015). The advantages of applying an analysis based on news
When investigating social desirability in regard to environmental topics, papers instead of surveys are the better access to old reports due to
studies have found a strong tendency of survey respondents to over- databases and the simpler analysis by means of analytical software
report environmentally responsible actions and their willingness to packages. However, this method also entails several downsides. As the
pay (Krumpal, 2013; Kovalsky and Lusk, 2013). This might raise con investigation of newspaper is a rather one sided measurement of
cerns about the significance of the impact of EA on the EV market and awareness and does not capture the conversation the society at large
highlights the necessity to investigate actual sales data (Moser, 2015). engages in, it might not accurately reflect the public's opinion. This
Furthermore, surveys depend on a high response rate, wherefore it is becomes even clearer when considering that journalists have a deonto
either difficult to generate a sufficient sample size that is large enough to logical obligation to determine whether they can report about a topic
be representative or often only possible with extensive costs and effort. (Olteanu et al., 2015).
This is especially the case when investigating a large geographic region. A relatively new medium used to investigate public opinion that has
This effect can also be observed when looking at previous studies. Most received increasing academic interest over the last years is social media
studies either focus on a single country or relatively small regions. channels and in particular Twitter. According to the company's annual
Another observation which can be traced back to the data collection report, the “[...] primary product, Twitter, is a global platform for public
method of surveys is the lack of studies investigating the development self-expression and conversation in real time” (Twitter, 2019, p. 7). Hence,
the main purpose of Twitter is to share opinions. Furthermore, the social
media platform has witnessed a worldwide growth in user numbers over
1
For a more detailed description of the theory please see: Ajzen (1988, the last years, allowing the conclusion of a global reach. This assumption
1991). can be further underscored as even politicians and supranational
3
Table 1
Selected studies investigating the impact of EA.
Author Observations N Method Significance of EA Distinctive features
Schlüter and Germany 124 Survey Significant (+) Study focus on EV acceptance by means of
Weyer (2019) car sharing
Trommer et al. Germany 3111 Online Survey 84% concerned about environmental Respondents were EV owners; Direct
(2015) impact question
Tu and Yang China 114 Survey Significant (+) Influence of EA indirectly deduced from TPB
(2019) questions
Wang et al. (2018) Shanghai 458 Survey Significant (+) Only 18.1% of respondents were willing to
purchase an EV
Zhang et al. (2018) China 264 Survey Paper Significant (+) Respondents were MBA students; Analysis of
questionnaires perceived environmental benefit
Zhuge and Shao China 550 Survey Discrete Four of six investigated factors are more People with higher income tend to have
(2019) choice model important to respondent than EA higher EA
Ziegler (2012) Germany 598 Survey Discrete Significant (+) Indirect assessment of EA via representing
choice model questions regarding environmental behavior
Wang and Dong China 240 Discrete choice Significant (+) EA represented by subjective norms based on
(2016) survey the TPB
Okada et al. (2019) Japan 246,642 (785 EV Online survey EA has a direct effect on purchase intention Investigation of purchase intentions and
owners) and indirect effect on post-purchase post-purchase satisfaction
satisfaction
Chu et al. (2019) China, Korea 204 (China) 177 Online Survey Environmental concern is important for Respondents categorised in early majority
(Korea) early majority (China) (China) and early adopters (Korea)
Nie et al. (2018) Shanghai 760 Stated choice Significant (+) Indirect measurement of EA
Survey
Plötz et al. (2014) Germany 210 Survey Significant (+) Directly asked about importance of a car's
environmental performance
Sierzchula et al. 30 countries 30 (one year) OLS Not significant Measures countries' environmentalism:
(2014) ranks environmental regulation and
performance
Wang et al. (2019) 30 countries 30 (one year) OLS Not significant Measures countries' environmentalism:
ranks environmental regulation and
performance
Vergis and Chen US states 50 (one year) OLS Not significant Measures states' environmentalism: ranks
(2015) environmental regulation and performance
Clinton and US states 3864 FE panel data Significant (+) Votes for members of congress used as
Steinberg (2019) (2010–2014) regression surrogate for EA
Zambrano- US states 300 (2010–2016) FE panel data Mixed results Democrats in legislature & green party
Gutiérrez et al. regression victories used as surrogate for EA
(2018)
Egnér and Trosvik Sweden 2030 FE panel data Significant (+) Green party votes used as surrogate for EA
(2018) (2010–2016) regression
Sheldon and California 7772 (December Generalized Not significant Votes of a suspension of a global warming
DeShazo (2017) census tract 2013) propensity score act used as surrogate for EA
Note: This table depicts a list of studies in which EA has been used as an explanatory factor for purchase intention or adoption of EVs.
institutions use this medium to communicate and share information. Samantray and Pin, 2019; Littman and Wrubel, 2019). Consequently,
Moreover, Shapiro and Hemphill (2017) found a significant influence of Twitter could be a promising method to measure EA in this field of
tweets from politicians on newspaper coverage. Their findings indicate research.
that Twitter posts in some areas are replacing what formerly would have Because most of the previous studies have applied surveys as a data
been communicated through press releases. Therefore, one can assume collection method, only few studies have investigated the potential
that topics discussed on Twitter will also reach a wider population, impact of EA on the EV market as a whole. Therefore, they only allow
outside of the active Twitter users. limited conclusions about the influence of EA on a macroeconomic level.
Twitter as an opinion barometer provides several advantages Hence, we see a research gap which is worth investigating as the results
compared to the previously mentioned methods. First, data software of such a study can not only supplement previous research but also
particularly focusing on Twitter has been developed to analyse tweets, provide new valuable insights.
which enables more advanced analyses. Furthermore, Twitter provides Our study contributes to existing literature in several ways. First, it
access to a large sample that would not be as easily accessible otherwise. contributes to the limited number of research that has investigated the
Based on the characteristics of a Twitter analysis discussed above, we actual impact of EA on electric vehicle registrations. Consequently, we
can conclude that this measurement method can be a useful supplement are able to potentially determine the marginal effects of EA. Further
to combat the drawbacks when using surveys as a barometer for public more, by applying Twitter as an awareness barometer, we are able to
opinion. overcome the limitations of surveys mentioned in the previous section.
Twitter as a tool to measure public opinion and sentiment is also Consequently, our analysis is not exposed to potential over-reporting of
becoming increasingly popular in sustainability related research fields. EA, an analysis of a larger region, namely the EU and EFTA, can be
For example Reboredo and Ugolini (2018) executed a sentiment anal conducted and the development over time of EA can be considered in
ysis, which examined the influence of investor sentiment on renewable our investigation. Based on the TPB and previous findings that EA has a
energy stocks, where Twitter was used as an investor sentiment proxy. significant positive impact on an individual's purchasing decision, we
There are also first studies which suggest that Twitter can be a useful hypothesise that EA significantly increases the sales of EVs.
tool in measuring EA (An et al., 2014; Cody et al., 2015). In addition,
more and more databases with tweet IDs are being published that can be
used to investigate public opinion regarding climate change (e.g.
4
3. Data and methodology accurate picture of the public's interest in this topic at the respective
point in time, as fewer tweets have been deleted. We obtained access to a
3.1. Data database of tweet IDs that include all tweets with the terms ‘climate
change’ and ‘global warming’ published from 2007 to 2017, collected by
To investigate the hypothesis established in the previous section, a Samantray and Pin (2019), from the Harvard Dataverse. Consequently,
data set in form of a panel was compiled. This data set is mostly in line all tweets that mention these word pairs in text or in form of a hashtag
with the study of Münzel et al. (2019) and captures branch specific as were considered in this study. The applied search terms are also in line
well as macroeconomic data but also extends it by our main independent with previous studies employing Twitter or other social media platforms
variable EA. As this analysis focuses on the European market, annual to analyse environmental issues (e.g. Kirilenko and Stepchenkova, 2014;
data on national-level was collected. More specifically, this investigation Kirilenko et al., 2015; Veltri and Atanasova, 2017; Santi, 2021). While
comprises 29 countries covering a period of 8 years between 2010 and other terms may be used in tweets relating to environmental topics, we
2017. Thereby, 27 countries of the panel are EU member states and 2 are believe that adding further terms would not significantly impact the
part of the EFTA.2 The time horizon of eight years is determined by the findings as it can be expected that the general trend would remain the
relatively short existence of these vehicles in the mainstream market. We same. A distinction between positive and negative statements was not
gathered several data from brochures and other online information made, since we assume that negative statements can also be seen to
material published on the homepages of institutions. As some in indicate the relevance of this topic for people as they might trigger
stitutions have updated their homepages, older information was no educational conversations. In total, Samantray and Pin (2019) collected
longer accessible. To encounter this problem we utilized the internet 14,353,859 tweets between 2007 and 2017. 13,671,284 of those fell into
archive as suggested by Münzel et al. (2019). A more detailed discussion the investigated time horizon of this study.
of the variables takes place in the following subsections. Due to data process restrictions put in place by Twitter, we have
decided to neglect the usage of a Twitter API on a data analytic software
3.1.1. Dependent variable and to count the number of tweets within a year by interpolation. This is
The dependent variable in this study is the ratio of EV registration possible as tweet IDs contain a timestamp component. More recent tweet
figures to the total number of the new registrations in the automobile IDs have a higher value than older tweet IDs. By organising tweet IDs in
sector. Registration figures for BEVs and PHEVs of the M1 car category ascending order, we were able to determine the first and last tweet
(passenger cars) were collected from the European Alternative Fuel within a year. Subsequently, all tweets between the first and last tweet of
Observatory (EAFO) (European Commission, 2021). EAFO is a highly a year can be counted into the corresponding year. The used data does
reliable source as it receives data directly from the ministries of the not violate any ethical standards, as no tweet content is being published
respective countries. The decision to select registration figures instead of or evaluated. The final measurement of EA was in form of a ratio,
actual sales figures is twofold. First, there are more reliable sources for whereby the amount of tweets using these search terms in one year was
the registration of EVs available as these values are mostly measured by put into relation with the number of active Twitter users in the corre
state agencies. The second and more important reason behind the choice sponding year.
is that this figure is more likely to represent where cars are actually used,
as vehicles can potentially have been bought in foreign countries. Values 3.1.3. Control variables
for BEVs and PHEVs were collected separately and then aggregated, We follow previous studies for the selection of our control variables.
representing the numerator of the ratio. These figures are then put in Most of our selected control variables thereby, capture the average value
relation to the total number of passenger car registrations in the corre of an indicator for the corresponding country and year. The sources
sponding year and country. where we obtained the values for the control variables are listed in the
As the introduction of EVs to the market is relatively new, there are Appendix (Table A.1). First, we account for different living conditions by
several countries in the data set that neither exhibit registration values controlling for the average disposable income of a household (Eurostat,
for BEVs nor PHEVs in the first years, leading to zero values for the 2021). As EVs are more expensive than comparable internal combustion
ratios. While there were no registrations in 17 countries in 2010, only cars (ICC), we assume a positive correlation between disposable income
after 2014 each country registered at least one EV. As recommended by and the registration number of EVs. As discussed in the literature review,
Münzel et al. (2019) we used the natural logarithm of the ratio which energy prices were determined to be another significant driver of the EV
means that ratios of 0 become missing observations. market. Thereby, we control for both fuel as well as electricity prices. We
gathered values for fuel prices from the EU energy statistical pocketbook
3.1.2. Environmental awareness (European Commission, 2010, 2013, 2014, 2015, 2018) and for elec
The measurement of EA applied in this investigation significantly tricity prices we obtained data from Eurostat's Energy statistics (Euro
differs from previous studies in this field of research discussed in chapter stat, 2020a). Thereby, we followed the approach of Münzel et al. (2019)
2.2 of the literature review. While the usual technique to determine an and calculated the average of semi-annual prices for medium-sized
individual's level of EA, was by carrying out surveys a more unbiased households. We also controlled for charging points per 100 km high
approach in form of a Twitter keyword analysis has been applied in this way and the number of fast chargers per 100 km highway, which we
research. A keyword analysis entails the use of certain words that can retrieved from EAFO, as well as the total number of cars in a country
function as search terms surrogating for the analysed topic. The choice (Eurostat, 2020b) and available EV models in Europe (Tsakalidis and
of our keywords was restricted to two word pairs, namely ‘climate Thiel, 2018). The reason for considering a determinant that reflects the
change’ and ‘global warming’, as we have decided to use an existing number of available EV models in Europe is that it captures the variety of
database rather than collecting the tweets ourselves. The decision was products between which a consumer can choose. This can especially be
based on the fact that when tweets are deleted they are no longer important in the early years as only a severely limited number of EV
accessible for analysis. Using a database established closer to the ana models was available. Furthermore, we acknowledge country specific
lysed time frame therefore provides the possibility to obtain a more differences in subsidies for EVs. While previous literature (Münzel et al.,
2019; Sierzchula et al., 2014) analysing the effect of subsidies on EVs
often distinguished between financial and non-financial incentives, we
2
Austria, Belgium, Bulgaria, Cyprus, Czech Republic, Denmark, Estonia, only consider financial incentives in our analysis. This is caused by the
Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, fact that non-financial incentives are predominantly granted on local
Luxembourg, Malta, Netherlands, Norway, Poland, Portugal, Romania, level, thus making them inapplicable for consideration in a study on a
Slovakia, Slovenia, Spain, Sweden, Switzerland, United Kingdom national level. More precisely, we only include incentives mentioned in
5
the ACEA tax report for the respective countries and years (ACEA, to 2017,3 the coefficient of our main independent variable EA indicates a
2017). This, however, should provide a sufficient database to capture negative influence of the EA level on the sales of EVs. This contradicts
the development of the incentives. For further explanation on the the theory that EA has a positive influence on the number of sales. When
determination of the taxes we recommend reading Münzel et al. (2019). investigating the data set to find a possible explanation for this phe
Thus, we include variables for all categories depicted in Fig. 1. nomenon, it becomes clear that this can most likely be attributed to the
EA outlier in 2010. 2010 exhibits by far the highest value for EA
(3.17%), while simultaneously exhibiting the lowest number of new EV
3.2. Methodology registrations. A potential explanation for the high EA value in 2010
could be the high number of global disasters and resulting fatalities in
Given the longitudinal structure of our data we apply panel data this year. Several scholars have identified a correlation between the
regression to determine the significance of EA on the shares of new EV number of global natural disasters and tweets about environmental
registrations. A panel data regression is favorable for our investigation topics (Molodtsova et al., 2013; Cody et al., 2015) which confirms our
as it cannot only capture the development of the influence of EA on the observation. Furthermore, we analysed the correlation between our EA
EV market for one but rather for several countries. This makes our model determinant and the number of global disasters (Statista, 2020) and the
more meaningful because it increases the representativeness for a larger amount of fatalities (Ritchie and Roser, 2020) as a result of natural di
region and improves the accuracy of the model parameters (Hsiao, sasters. Although there are only a few observations, the correlation test
2007). (Table A.3 in the Appendix) indicates an almost perfect correlation be
To determine whether a fixed effects (FE) or random effects (RE) tween EA and fatalities. Therefore, we assume that EA increases in times
model should be applied we looked at related studies and executed of increased natural disasters as climate change will come more into
statistical tests. In line with the study of Münzel et al. (2019), we applied focus due to media reports. When investigating the EV market and the
a fixed effects model in our analysis. This decision is further supported corresponding infrastructure, which existing research has shown to have
by an auxiliary regression that exhibits results in favour of a fixed effects a significant influence on EV sales, it becomes clear that in 2010 the
model. As suggested by Wooldridge (2010), we employed an auxiliary infrastructure e.g. charging station network was not very well developed
regression instead of a Hausmann test as a specification test due to the and partially even non-existent (only Norway and the Netherlands had
heteroskedasticity in our model. The final model is defined as follows: charging points). In addition, PHEVs which do not only rely on elec
( )
EP tricity by also incorporating an internal combustion motor were no real
ln(EV)it = β1 EAit + β2 CPit + β3 FCit + β4 DI it + β5 ln
DP it alternative as there was only one model available on the European
(1) market. Due to these facts and the on average significantly higher prices
+β6 CF it + β7 EVM it + β8 Sit + αi + εit of EVs compared to ICCs, we can assume that even for environmentally
aware people, EVs represented no real substitute for ICCs. Consequently,
we ran a second regression omitting the year 2010, while keeping
where the subscripts i and t denote the country and year, respectively. everything else the same. The results of the FE and FD regressions are
The variable names are abbreviated in the equation and can be inter illustrated in Table 2.
preted as follows: EV (EV registration share), EA (environmental
awareness), CP (charging points total per 100 km highway), FC (fast
chargers per 100 km highway), DI (disposable income), EP (electricity Table 2
Regression results.*
price), DP (diesel price), CF (car fleet), EVM (EV models), S (subsidies).
Based on the econometric model of Münzel et al. (2019), we used the Variable Fixed effects First differences
natural logarithm of EV sales shares and electricity to diesel price. To Marginal Robust Std. Marginal Robust Std.
control for the potential influence of energy prices we also solely Effects Err. Effects Err.
considered diesel prices to prevent potential multicollinearity and used Environmental 26,217,203.4% 14.54289 6354% 15.37895
the ratio of electricity to diesel prices to save one degree of freedom. αi awareness
and εit denote the country specific effects and the error term respec Charging points − 0.0803%*** 0.0002818 − 0.1244% 0.0009503
total
tively. To account for serial correlation and heteroskedasticity in our
Fast chargers 0.1732%** 0.0006783 0.5566% 0.0044542
data set, we employed robust standard errors clustered at country level. Disposable 0.0091% 0.0001353 0.0310%*** 0.00009
This approach is suggested by literature to counteract these issues when income
having a large n and small t panel (Arellano, 1987; White, 1980). All Number EV 7.1596%*** 0.00842 6.5699%*** 0.0091951
tests and statistical analyses were conducted with the statistical software models
LN (electricity to − 1.776%** 0.8111937 − 2.281%*** 0.7068673
Stata.
diesel price)
Car fleet 0.00% 1.21E-07 0.00% 1.13E-07
4. Results Subsidies 0.0112%*** 2.21E-05 0.0076%*** 2.67E-05
Observations 193 163
F-score 98.91*** 74.56***
We performed a panel data regression as described in the method Within R2 0.7372
ology section and tested our model for specification accuracy. Thereby, Between R2 0.0877
we detected that our model violates the assumptions of homo Overall R2 0.2657 0.4087
skedasticity and non-serial correlation. While we identified the presence Note: Country Fixed Effects model with robust standard errors clustered at
of heteroskedasticity by employing a likelihood ratio test as well as a country level. The marginal effects are calculated using the following formula:
visual analysis, we were able to determine the existence of serial cor 100 * (eβ − 1).
relation by following the approach of Wooldridge (2010). To counteract *
p < 0.1.
**
the potential influences of heteroskedasticity and serial correlation, we p < 0.05.
***
employed robust standard errors clustered at country level. The pres p < 0.01.
ence of serial correlation in the idiosyncratic errors and the relatively
small sample of our analysis suggest a higher validity of estimates
stemming from an FE than those stemming from a first difference (FD)
regression (Wooldridge, 2016; Münzel et al., 2019).
After running the regression with the aspired time horizon from 2010 3
The results of this regression can be found in Table A.2 in the Appendix
6
An F-score of 98.91 and a within R2 result of 0.7372 indicate a high becomes evident that the main independent variable is still insignificant.
explanatory power of our model. When analysing the significance of the
variables included in the model, it can be seen that charging points total 4.1.3. Charging infrastructure
per 100 km highway, the number of EV models and subsidies are sig Several researchers have raised a general concern of including
nificant at a 1% level. At a 5% level the number of fast chargers per 100 charging infrastructure in an analysis on a national level, as the number
km highway and the ratio of electricity to diesel price are significant. and distribution of charging stations can significantly differ between
Most importantly, our main independent variable EA exhibits no sig regions within a country (Coffman et al., 2017; Münzel et al., 2019).
nificant influence on the new registrations of EVs. Moreover, some scholars pointed out that there might be a reversed
causality problem between charging infrastructure and the sales of EVs,
4.1. Robustness tests as it could be possible that governments react to increased EV pur
chasing behavior by providing new charging stations (Coffman et al.,
Before a discussion of the results will take place, the robustness of the 2017). Since according to Münzel et al. (2019), the allocation of
outcomes will be tested. For the robustness tests, we orientate on the charging stations is, however, considered to be based on subsidies and
work of Münzel et al. (2019) and apply some of their robustness tests. In regulations and not sales volume of EVs, we would also assess this
particular, we also test for a trend in the EV market and remove small problem as not significant for this study. Nevertheless, we want to ac
vehicle markets from the regression. Furthermore, we run three re count for the small probability of endogeneity in the model by running a
gressions that should check the robustness of our results in case of po regression that omits the charging infrastructure determinants.
tential endogenity or multicollinearity between our charging Furthermore, when looking at the selected variables of our regres
infrastructure variables. In addition, we repeatedly exclude one country sion model, it can be seen that we have included two variables that
from the regression to assure that the results are not driven by one represent the entire and fast charging infrastructure of a country.
particular country. Another robustness test that was carried out is a Although, the absence of multicollinearity was checked by employing a
random effects model to account for a potential misspecification of the Variance Inflation Factor test, we want to make sure that no undetected
regression. Lastly, we applied an alternative measurement of EA in form multicollinearity affects the outcome of our results. Therefore, two more
of a Google Trends analysis. robustness tests are carried out that omit CP and FC, respectively. The
insignificance of the regression results of our main independent vari
4.1.1. Trend variable able, do not change for either of these variations and thus support our
Pointed out by Münzel et al. (2019) as an important determinant in analysis (Table A.6 in the Appendix).
the regression, a trend variable is added into the model as a robustness
test. The authors justify the inclusion of a trend variable to capture 4.1.4. Stepwise country exclusion
general trends in the EV market such as increasing EV model availability A stepwise exclusion of each country is executed to eliminate the
or decreasing battery prices. Also accounting for these factors we rerun possibility that for example outliers in a single country too strongly in
the regression by including a trend variable into the model. The trend fluence the results of the regression. Based on the results of the re
variable is estimated by using the natural logarithm of the years and gressions, which can be seen in Table A.7 in the Appendix, no strong
therefore corresponds with the approach of Münzel et al. (2019). The deviations from the baseline results can be noticed. EA as the focus
results of the regression can be seen in the Table A.4 in the Appendix. variable in this investigation remains insignificant for all alternatives.
The variable EA still neither exhibits significant results in the FE
model nor in the FD model. Another interesting finding is that in 4.1.5. Random effects model
contrast to the results of Münzel et al. (2019) the trend variable in this Although, the application of an FE model was not only proven by
investigation is not significant. This can potentially be traced backed to diagnostic tests but also more importantly with the help of past litera
the inclusion of a variable reflecting the available EV models in the EU, ture and the scope of analysis, an RE model will be executed as a
in this research. In contrast, in the study of Münzel et al. (2019), this robustness test to account for the possibility of a misspecification of the
determinant should have been captured by the trend variable. This model. As it can be seen in Table A.8 in the Appendix, an RE model does
assumption can be supported when looking at the significance level of not yield a different result for the main independent variable. Conse
the variable Number EV Models. While this variable is significant at a 1% quently, even though there might be a small probability that the model
level in the baseline model, the variable is only significant at a 5% level could be misspecified, the results of this regression support the findings
when including a trend variable in the model. A possible explanation for of the baseline regression.
this could be that the variable Number EV Models captures large parts of
the explanatory power of the trend variable. 4.1.6. Google trends as an EA measurement
It is difficult, if not impossible to develop an unbiased variable that
4.1.2. Small vehicle markets can perfectly reflect a psychological factor such as EA. This is because
As in Münzel et al. (2019), we also control for the influence of small psychological factors are more difficult to observe and consequently
car markets. Since the registration of an EV has a stronger influence on more challenging to measure. In contrast to for example the registration
the registration rate in a small car market than in a large car market, rate of cars, EA levels are not directly measurable and must be derived
anomalies can severely bias the results. This is amplified, as in an FE by means of appropriate surrogates. Although, Twitter is as pointed out
model each country has an equal weighting in the estimation, wherefore in the literature review an appropriate and innovative source of infor
the country's influence on the value of the coefficient is independent of mation to measure the awareness of a society regarding a certain topic, it
its car market. Consequently, when considering the EV market in a Eu is not free of limitations. Therefore, in this robustness test a different
ropean context, small car markets could have a leveraging effect on the measure of EA will be applied to check for the validity of the results. In
outcome. To control for this effect small car markets will be excluded as particular, Google Trends (GT) will be utilized to measure a country's
a robustness test. Small car markets in this research are defined as level of EA. GT is not an uncommon measure for awareness in this strand
markets with less than 30,000 new car registrations per year. Thus, a of literature. Vergis and Chen (2015) for example used GT to measure a
country that reports less than 30,000 car registrations in one of the societies awareness of EVs. The applied search term to measure EA in GT
considered years will be excluded from the regression. Only 4 countries, is climate change in the corresponding languages of the countries. In
exhibit registration values below 30,000. These are: Estonia, Latvia, some cases, when the search term climate change was not used frequently
Lithuania and Malta. Comparing the results of the regression, which are enough in one language, the translation of global warming in the do
illustrated in Table A.5 in the Appendix with the baseline results, it mestic language was used instead. Table A.9 in the Appendix presents
7
the applied search term for each country and the corresponding English high purchase prices of EVs compared to ICCs. Therefore, this type of car
translation. The results of the FE regression which can be seen in can, in the early stages, be seen to be primarily attractive to status
Table A.10 in the Appendix exhibit insignificant results for EA and seekers which usually have a high disposable income. As EV prices are
therefore, support the findings of the analysis. assumed to continuously drop, it can be expected that this factor will
In the subsequent chapter a discussion of the results will take place. become less relevant over time (de Rubens, 2019). Similarly to price,
Thereby, different explanations for the insignificant outcome of EA and ICCs currently still have the advantage of a farther range than EVs which
its implication for the EV market will be presented. Furthermore, we will could further reduce the attractiveness of the latter.
outline the limitations of this study and provide suggestions for further A further hypothesis might be that very environmentally aware
research. people might not yet see EVs as sustainable enough and therefore prefer
other modes of transportation such as public transport. Concerns that
5. Discussion of results have been expressed regarding the sustainability of EVs include the issue
of non renewable energy sources used to power EVs, as well as the
Contradictory to our hypothesis, the results of our investigation comparatively higher CO2 emissions caused by the production, main
exhibit no significant influence of the level of EA on the registration rate tenance and disposal of EVs (Federal Ministry for the Environment and
of new EVs. Based on the findings it cannot be ruled out that EA to some Safety, 2019). Similar to other factors, the advancement of technology in
degree influences the EV market, however, the results suggest that other this sector is expected to reduce this issue.
factors might have played a more important role. The outcome, when It becomes clear that during the investigated time frame, customers
taking on a macro-economic perspective and looking at actual sales faced several barriers that might have prevented them from purchasing
data, does not confirm the findings of previous literature that predom EVs. As the market evolves, these barriers can be expected to become
inantly investigated the determinants of EV purchase decisions at an less significant compared to the barriers of purchasing an ICC. There
individual level. fore, we hypothesise that once the barriers have been reduced to a level
The TPB provides a possible explanation for these findings through where EVs become a feasible alternative to ICCs, EA will play a more
the influence of the perceived behavioral control dimension. Perceived significant role in the development of EV sales. Consequently, we sug
behavioral control, which can be understood as external factors that gest to repeat this investigation at a later point in time. This recom
hinder the engagement in a certain behavior, is the only factor that has a mendation can be further underscored when looking at the protests for
direct influence on both intention and adoption behavior. Hence, it EA (e.g. Fridays for Future) that took place in the last years. Studies have
could be assumed that while EA has a significant positive impact on the shown that demonstrations can significantly change peoples awareness
intention to purchase an EV, the influence of factors such as high prices regarding a topic (Olteanu et al., 2015), wherefore peoples' EA poten
or limited charging infrastructure mitigates the positive influence of EA tially has significantly increased especially in the years 2018 and 2019
on actual EV adoption. We hereby provide further support for the which are not included in our study.
attitude-action gap in this research field. As discussed in the introduction, EA is also an important topic for car
Our results further correspond with the findings of de Rubens (2019) manufacturers when advertising their products. However, based on our
who carried out a comprehensive survey in five different Nordic regions results it might be favorable for car manufacturers to alter the focus and
with the objective of determining different customer groups and inves highlight other advantages of EVs to illustrate that some perceived
tigating customers' interest in purchasing EVs. After clustering the re barriers might have already been overcome. For example, several
spondents into six different customer groups, the researcher found that studies that have focused on total cost of ownership, found that the
although the group with the highest EA level has the largest interest in initial higher purchasing prices of EVs are to some extent compensated
EVs, this group exhibits the second smallest ownership ratio of EVs. Even by lower operational costs (e.g. Wu et al., 2015; Lévay et al., 2017).
the customer group this study refers to as 'Sceptics' (a cluster of people Furthermore, the compensation effectiveness increases with the mileage
that scored low on EA) demonstrates a more than four times higher EV of the car (Wu et al., 2015). Another potential characteristic to highlight
ownership ratio. Subsequently, these findings support our analysis that in marketing, could be the technological features such as the accelera
EA does not significantly impact the sales ratio of EVs and that this can tion of EVs to better attract the technological affine consumer group (de
potentially be attributed to the fact that other variables have a stronger Rubens, 2019). Martin and Väistö (2016) for example found that Tesla
influence on the EV purchasing decisions in the investigated time period. which is seen to be one of the leading manufacturers of EVs primarily
For example, control variables shown to have a significant influence on appeal to the hedonic motivation of consumers with their marketing
the EV market in our study are the number of charging points and fast strategy. This leads us to another recommendation for future research.
chargers per 100 km highway, the number of EV models available, the We propose to further investigate what kind of marketing messages are
electricity price in relation to the diesel price as well as subsidies for EVs. especially appealing to EV customers.
Based on these results and findings of previous literature, we have Another hypothesis we would recommend investigating in future
established multiple hypotheses that could explain the insignificant research involves the indirect influence of EA on the EV market via
impact of EA on the EV market. Most of the hypotheses focus on po politics. Our study supports prior findings that show, that subsidies have
tential barriers which need to be overcome before EA can develop into a a significant influence on EV sales. A potentially interesting research
decisive factor. First of all, EVs are a relatively new technology. To make question could be, whether EA has an influence on political decisions by
this technology competitive to normal ICCs it requires a certain infra increasing political actions such as subsidies or investments into sus
structure. This infrastructure is mainly reflected in public charging tainability research. Lastly, our study indicates, there are potentially
points, which has been determined to be an important requirement (e.g. significant differences in findings on a macroeconomic level as opposed
Sierzchula et al., 2014; Mersky et al., 2016). Thereby, not only the to the individual consumer level. We therefore recommend conducting
number of public charging points is relevant but also the standardization more research on a macroeconomic level to gain a better understanding
of the charging process (Van Den Bossche et al., 2016; Hall and Lutsey, of the actual effects on the market.
2017). While the number of fast chargers per 100 km highway has a Like other studies, this study is of course not free of limitations.
significant positive impact on EV sales in our investigation, paradoxi Although our methodology addresses several issues of other measure
cally, the total number of charging points per 100 km highway exhibits a ment approaches for EA, it also entails some drawbacks which are worth
significant negative relationship to the EV ratio. This unexpected addressing. First of all, Twitter is at first sight limited by the users of this
outcome is similar to that of Münzel et al. (2019) but will not be further platform and thus does not represent the entire population. The users of
discussed as it is not the focus of the present research. this platform, however, are seemingly well diversified regarding their
Another barrier that hinders the expansion of the EV market are the demographic characteristics. In addition, when comparing the
8
demographics of the Twitter population and the individuals who are 6. Conclusion
interested in purchasing EVs, according to previous literature, it can be
seen that those characteristics correspond very closely. Similar to the This study has investigated the impact of EA on the EV market. While
population of Twitter users, the typical EV owners and potential cus previous studies in this field of research have focused on whether EA has
tomers of EVs tend to be younger, male, well educated and generally a significant influence on an individual's purchase intention, this study
have above average income (Tien, 2018). Furthermore, as discussed in takes on a macroeconomic perspective by examining the impact on
the literature review, this social media platform potentially has a reach actual EV registration rate. By applying a novel approach to measuring
that exceeds its user base as it has been shown that Twitter activity in EA in this strand of literature in form of a Twitter keyword analysis, we
fluences media coverage. were able to contribute to existing research by gaining insights into the
Another limitation of this research is that it solely relies on tweets influence of EA at a macroeconomic level. Conducting a panel regression
written in the English language and only relies on two word pairs as on 29 EU and EFTA countries, we demonstrated, that contrary to what
search terms. This is the case because we decided to use an existing previous findings might suggest EA does not significantly influence the
database of tweet IDs due to the fact that older tweets could have been EV market. Based on the findings it cannot be ruled out that EA to some
deleted and therefore might not be accessible anymore. This in turn degree influences the EV market, however, the results suggest that other
could distort the results. For future research it could be beneficial to factors might have played a more important role. We hypothesise that
collect data in the native languages of the investigated countries as well, EA will become more relevant once the market has evolved and the
and add more keywords to the search, to get a more accurate measure of factors that currently might still hinder EV purchasing decisions are
EA. It should, however, be noted that English is the most used language minimised. Consequently, it could be of academic interest to repeat this
on Twitter and that Twitter's reach is not restricted by geographical study at a later point.
barriers. Users therefore are able to follow and engage with users from
various countries. For these reasons, and also because climate change is Author contribution
a global problem, we assume that the English tweets relatively closely
capture EA also for non English speaking countries. In addition, as Research design: Austmann and Vigne.
pointed out earlier, Twitter users are generally well educated and Data Collection: Austmann.
therefore it is likely that most users understand and speak English in Analysis: Austmann.
addition to their native language. The results of the GT analysis further Supervision: Vigne.
support this assumption as they led to the same results as the Twitter
analysis. Declaration of Competing Interest
None.
Appendix A
Table A.1
Data sources.
Variable Country (if divergent) Source
EV registrations European Commission (2021)

Total new registrations Cyprus, Malta ACEA (2018)
Eurostat (2019)
Environmental awareness Samantray and Pin (2019)
Charging points total European Commission (2021)
Fast chargers European Commission (2021)
Disposable income Eurostat (2021)
Malta CEIC (2020a)
CEIC (2020b)
Number of EV models Tsakalidis and Thiel (2018)
Diesel price European Commission (2010, 2013, 2014, 2015, 2018)
Switzerland Bundesamt für Statistik (2020)
Norway Statistics Norway (2018)

Electricity Price Switzerland Eurostat (2020a)
Bundesamt für Statistik (2020)
Car Fleet Eurostat (2020b)
Global Disasters Statista (2020)

Subsidies ACEA (2010, 2011, 2012, 2013, 2014, 2015, 2016, 2017)
Note: When single observations for countries were missing, we have applied the Microsoft Excel trend function to achieve a complete data
set. This approach is in line with previous literature (Münzel et al., 2019).
Table A.2
Regression results 2010–2017.
Variable Fixed effects First differences
Marginal Effects Robust Std. Err. Marginal Effects Robust Std. Err.
Environmental awareness − 100%*** 7.47428 − 100% 10.88801

Charging points total − 0.02858% 2.03E-04 − 0.0052%* 2.84-E05
(continued on next page)
9
Table A.2 (continued )

Fast chargers − 0.0215% 0.0010322 0.0470%* 0.0002699

Disposable income 0.0103% 0.0001148 0.0306%*** 0.0001014
Number EV models 7.5085%*** 0.0066825 7.9900%*** 0.0096365
LN (electricity to diesel price) − 1.117% 0.8107383 − 2.344%*** 0.7519457
Car fleet 0.00% 1.21e-07 0.00% 1.28E-07
Subsidies 0.0097%*** 2.05E-05 0.0102%** 4.00E-05
F-score 118.53*** 73.86***
Within R2 0.8296
Between R2 0.0131
Overall R2 0.2674 0.5369
Note: Country Fixed Effects model with robust standard errors clustered at country level. The marginal effects are calculated using the following formula: 100 * (eβ
− 1).
*
p < 0.1.
**
p < 0.05.
***
p < 0.01.
Table A.3
Correlation EA & disasters.
Environmental awareness Global disasters Fatalities
Environmental awareness 1.0000

Global disasters 0.7734 1.0000
Fatalities 0.9528 0.7151 1.0000
Table A.4
Regression results with trend variable.*
Environmental awareness 20,288,199.1% 14.2629 3,121,142.5% 15.31784

Charging points total − 0.0885%*** 0.000308 − 0.1518% 0.0010396
Fast chargers 0.1750%** 0.0006408 0.5352% 0.0044312
Number EV models 5.4261%** 0.0222916 3.5196% 0.0267733
LN(electricity to diesel price) − 1.828%** 0.7888554 − 2.373%*** 0.6883063
Car fleet 0.00% 1.26E-07 0.00% 1.30E-07
Subsidies 0.0108%*** 2.29E-05 0.0070%*** 2.54E-05
Trend 1087.65% 335.2596 13,947.2% 414.3284
F-score 85.19*** 70.65***
Within R2 0.7381
Between R2 0.0824
Overall R2 0.2880 0.4143
Note: Country Fixed Effects model with standard errors clustered at country level. The marginal effects of the coefficients are calculated using the following
formula: 100 * (eβ − 1).
*
p < 0.1.
**
p < 0.05.
***
p < 0.01.
Table A.5
Regression results excluding small markets.*
Environmental awareness − 91.07% 16.14028 − 75% 19.67168

Fast chargers 0.2909%*** 0.0007999 0.4052% 0.0027536
Disposable income 0.0192%** 8.58E-05 0.0371%*** 0.0000914
Number EV models 6.7996%*** 0.00749 91.9239%*** 0.0103508
LN (electricity to diesel price) − 0.519% 0.586119 − 1.449%** 0.7230032
Car fleet 0.00% 9.22E-08 0.00% 1.17E-07
Subsidies 0.0102%*** 2.18E-05 0.0082%** 2.95E-05
10

F-score 79.09*** 115.30***

Within R2 0.8182
Between R2 0.6365
Overall R2 0.7141 0.4617
Note: Country Fixed Effects model with standard errors clustered at country level. The marginal effects of the coefficients are calculated using the following
formula: 100 * (eβ − 1).
*
p < 0.1.
**
p < 0.05.
***
p < 0.01.
Table A.6
Regression results charging infrastructure.*
Variable w/o charger w/o charger total w/o charger fast
Marginal Effects Robust Std. Err. Marginal Effects Robust Std. Err. Marginal Effects Robust Std. Err.
Environmental awareness 4,594,444.53% 14.71364 3,889,162.8% 14.61002 10,842,899.1% 14.32455

Charging points total − 0.0511% 0.0003591
Fast chargers − 0.0736% 0.0007757
Disposable income 0.0100% 0.0001329 0.0099% 0.0001328 0.0093% 0.0001337
Number EV models 6.7704%*** 0.0073009 6.8317%*** 0.0075869 7.1098%*** 0.008396
LN (electricity to diesel price) − 1.650%** 0.7866404 − 1.656%** 0.7809143 − 1.740%** 0.791798
Car fleet 0.00% 1.18E-07 0.00% 1.19E-07 0.00% 1.20E-07
Subsidies 0.0112%*** 2.27E-05 0.0113%*** 2.28E-05 0.0113%*** 2.22E-05
Observations 193 193 193
F-score 126.78*** 107.76*** 103.67***
Within R2 0.7343 0.7345 0.7365
Between R2 0.1381 0.1324 0.0978
Overall R2 0.2985 0.2940 0.2714
Note: Country Fixed Effects model with standard errors clustered at country level. The marginal effects of the coefficients are calculated using the following formula:
100 * (eβ − 1). Only Fixed Effects results are reported.
*
p < 0.1.
**
p < 0.05.
***
p < 0.01.
Table A.7
Regression results stepwise country exclusion.
Variable Austria Belgium Bulgaria Cyprus Czechia Denmark Estonia Finland France Germany
Environmental 7.70E+06% 4.36E+09% 2.23E+10% 3.08E+05% 5.76E+06% 1.74E+06% 1.58E+04% 4.85E+08% 1.44E+07% 2.48E+08%
awareness
Charging points − 0.0815%*** − 0.0804%*** − 0.0839%*** − 0.0764%*** − 0.0887%*** − 0.0797%** − 0.0821%*** − 0.0736%** − 0.0779%** − 0.0769%**
total
Fast chargers 0.1703%** 0.1870%*** 0.1824%*** 0.1670%** 0.1835%** 0.1749%** 0.2399%*** 0.1614%** 0.1621%** 0.1611%**
Disposable income 0.0094% 0.0080% 0.0086% 0.0066% 0.0104% 0.0111% 0.0198%** 0.0091% 0.0089% 0.0096%
Number EV models 7.2745%*** 7.0144%*** 7.1972%*** 7.2079%*** 7.2811%*** 7.3692%*** 6.9151%*** 6.9556%*** 7.2096%*** 7.0696%***
LN (electricity to − 1.827%** − 1.966%** − 2.030%** − 1.599%* − 1.815%** − 1.935%** − 1.129%* − 1.734%** − 1.753%** − 1.749%**
diesel price)
Car Fleet 0.0000% 0.0000% 0.0000% 0.0000% 0.0000% 0.0000% 0.0000% 0.0000% 0.0000% 0.0000%
Subsidies 0.0144%*** 0.0122%*** 0.0111%*** 0.0113%*** 0.0109%*** 0.0103%*** 0.0103%*** 0.0112%*** 0.0114%*** 0.0115%***
Variable Greece Hungary Ireland Italy Latvia Lithuania Luxembourg Malta Netherlands Norway
Environmental 6.88E+08% 5.73E+06% 8.10E+06% 5.46E+07% 1.49E+07% 2.35E+06% 4.02E+08% 8.69E+07% 4.28E+07% 2.77E+07%
awareness
Charging points − 0.0657%** − 0.0840%** − 0.0832%*** − 0.0781%** − 0.0770%*** − 0.0799%*** − 0.0809%*** − 0.0877%*** − 0.0612% − 0.0917%***
total
Fast chargers 0.1589%** 0.1825%** 0.1692%** 0.1718%** 0.1641%** 0.1756%** 0.1827%*** 0.2008%*** 0.01296% − 0.2676%**
Disposable income 0.0142% 0.0085% 0.0055% 0.0107% 0.0095% 0.0080% 0.0090% 0.0047% 0.0099% 0.0090%
Number EV models 6.5813%*** 7.3338%*** 7.4299%*** 6.9919%*** 7.0480%*** 7.2182%*** 7.1038%*** 7.3201%*** 6.9672%*** 7.3363%***
LN (electricity to − 1.983%** − 1.941%** − 1.653%* − 1.744%** − 1.549%* − 1.831%** − 1.773%** − 1.653%* − 1.771%** − 1.743%**
diesel price)
Car fleet 0.0000% 0.0000% 0.0000% 0.0000% 0.0000% 0.0000% 0.0000% 0.0000% 0.0000% 0.0000%
Subsidies 0.0121%*** 0.0105%*** 0.0120%*** 0.0114%*** 0.0113%*** 0.0118%*** 0.0109%*** 0.0113%*** 0.0108%*** 0.0112%***
Variable Poland Portugal Romania Slovakia Slovenia Spain Sweden Switzerland UK

Environmental 3.69E+06% 9.60E+06% 3.58E+09% 8.45E+05% 5.64E+06% 3.86E+07% 1.64E+07% 2.09E+07% 5.31E+07%
awareness
Charging points − 0.0829%*** − 0.0816%*** − 0.0785%*** − 0.0849%*** − 0.0806%** − 0.0803%** − 0.0726%** − 0.0898%*** − 0.0862%***
total
Fast chargers 0.1731%** 0.1685%** 0.1857%*** 0.1772%** 0.1745%** 0.1736%** 0.1446%* 0.1957%*** 0.1706%**
11

Variable Austria Belgium Bulgaria Cyprus Czechia Denmark Estonia Finland France Germany
Disposable income 0.0086% − 0.0084% − 0.0053% − 0.0068% − 0.0081% − 0.0088% 0.0095% − 0.0090% − 0.0096%
Number EV models 7.1990%*** 7.2213%*** 7.3153%*** 7.4557%*** 7.2307%*** 7.1810%*** 6.9867%*** 7.1841%*** 7.1788%***
LN (electricity to − 1.761%** − 1.684%* − 1.933%** − 1.842%* − 1.790%** − 1.787%** − 1.722%** − 1.865%** − 1.835%**
diesel price)
Car fleet 0.0000% 0.0000% 0.0000% 0.0000% 0.0000% 0.0000% 0.0000% 0.0000% 0.0000%
Subsidies 0.0111%*** 0.0115%*** 0.0106%*** 0.0114%*** 0.0107%*** 0.0113%*** 0.0110%*** 0.0112%*** 0.0113%***
Note: Country Fixed Effects model with standard errors clustered at country level. The marginal effects of the coefficients are calculated using the following formula:
100 * (eβ − 1). The standard errors are available upon request.
*
p < 0.1.
**
p < 0.05.
***
p < 0.01.
Table A.8
Regression results random effects model.*
Variable Random effects First differences
Environmental awareness 40,199.66% 13.27839 6354% 15.37895

Charging points total 0.0375% 0.0006048 − 0.1244% 0.0009503
Fast chargers 0.0720% 0.0026896 0.5566% 0.0044542
Disposable income 0.0168%*** 2.92E-05 0.0310%*** 0.00009
Number EV models 6.2544%*** 0.004977 6.5699%*** 0.0091951
LN (electricity to diesel price) − 1.436%** 0.709784 − 2.281%*** 0.7068673
Car fleet 0.00% 9.31E-09 0.00% 1.13E-07
Subsidies 0.0103%*** 2.02E-05 0.0076%*** 2.67E-05
Wald χ 2 758.20*** 74.56*** (F-score)
Within R2 0.7263
Between R2 0.6783
Overall R2 0.7019 0.4087
Note: Country Random Effects model with standard errors clustered at country level. The marginal effects of the coefficients are calculated using the following formula:
100 * (eβ − 1).
*
p < 0.1.
**
p < 0.05.
***
p < 0.01.
Table A.9
Search terms google trends.
Country Search term English translation
Austria Klimawandel Climate change

Belgium Klimawandel & Réchauffement Climatique Climate change
Bulgaria глобално затопляне Global warming
Cyprus κλιματική αλλαγή Climate change
Czech Republic změna klimatu Climate change
Denmark global opvarmning Global warming
Estonia globaalne soojenemine Global warming
Finland ilmastonmuutos Climate change
France Réchauffement Climatique Climate change
Germany Klimawandel Climate change
Greece κλιματική αλλαγή Climate change
Hungary éghajlatváltozás Climate change
Ireland Climate Change Climate change
Italy cambiamento climatico Climate change
Latvia globālā sasilšana Global warming
Lithuania klimato kaita Climate change
Luxembourg Réchauffement Climatique Climate change
Malta Climate Change Climate change
Netherlands klimaatverandering Climate change
Norway global oppvarming Global warming
Poland zmiana klimatu Climate change
Portugal das Alterações Climáticas Climate change
Romania schimbarea climei Climate change
Slovakia globálne otepľovanie Global warming
Slovenia globalno segrevanje Global warming
Spain cambio climático Climate change
Sweden klimatförändring Climate change
Switzerland Klimawandel & Réchauffement Climatique Climate change
United Kingdom Climate Change Climate change
Note: Google Trends measures the search frequencies of the search terms based on an index. The highest value of
this index is 100 and represents the month with the highest popularity of the search term. The annual value has
12
been calculated using the average of the 12 months values. Two different languages were used for Belgium and
Switzerland as there is no single main language spoken in these countries. The Italian language has not been
considered for Switzerland in this investigation, as the term cambiamento climatico was not used frequently as a
search term in this country.
Table A.10
Regression results Google trends 2011–2017.*
Environmental awareness 1.0212% 0.0089458 − 0.4886% 0.0100661

Fast chargers 0.2207%*** 0.0007049 0.5549% 0.004336
Number EV models 7.2244%*** 0.0078434 6.4776%*** 0.0093
LN (electricity to diesel price) − 1.631%** 0.7198995 − 2.198%*** 0.6166989
Car fleet 0.00% 1.26E-07 0.00% 1.12E-07
Subsidies 0.0109%*** 2.04E-05 0.0077%*** 2.65E-05
F-score 68.27*** 63.37***
Within R2 0.7391
Between R2 0.3147
Overall R2 0.5014 0.4094
Note: Country Fixed Effects model with standard errors clustered at country level. The marginal effects are calculated using the following formula: 100 * (eβ − 1).
*
p < 0.1.
**
p < 0.05.
***
p < 0.01.
Table A.11
Descriptive statistics.
Variable Mean Std. Dev. Min Max Obervations
EV registration share Overall 0.0115543 0.0396366 0 0.3918689 N = 203

Between 0.0304471 0.0006055 0.164367 n = 29
Within 0.0259146 − 0.1382766 0.2390562 T=7
Environmental awareness Overall 0.0064279 0.0029467 0.0033392 0.111372 N = 203
Between 8.83e-19 0.0064279 0.0064279 n = 29
Within 0.0029467 0.0033392 0.0111372 T=7
Charging points total Overall 102.6278 245.9346 0 1966.539 N = 203
Between 217.0903 0.6558189 1125.867 n = 29
Within 121.472 − 426.1074 943.2998 T=7
Fast chargers Overall 11.60849 36.40812 0 381.0707 N = 203
Between 26.14267 0 115.0505 n = 29
Within 25.73737 − 100.0004 277.6287 T=7
Disposable income Overall 19,582.88 5771.139 8678 33,089 N = 203
Between 5779.306 9746.429 31,694.86 n = 29
Within 947.5025 17,314.59 22,413.59 T=7
Number EV models Overall 43.42857 13.92016 22 61 N = 203
Between 0 43.42857 43.42857 n = 29
Within 13.92016 22 61 T=7
Electricity to diesel price Overall 0.098341 0.0236435 0.0551923 0.1739604 N = 203
Between 0.0212856 0.0662782 0.1461551 n = 29
Within 0.0109268 0.0737239 0.1261463 T=7
Car fleet Overall 8,762,393 1.22e+07 247,174 4.65e+07 N = 203
Between 1.24e+07 266,971 4.46e+07 n = 29
Within 450,154.6 6,721,008 1.11e+07 T=7
Subsidies Overall 3210.19 3645.31 0 15,123.85 N = 203
Between 3348.52 0 13,720.56 n = 29
Within 1552.01 − 2389.63 8602.80 T=7
Global disasters Overall 348.7143 14.74997 331 378 N = 203
Between 0 348.7143 348.7143 n = 29
Within 14.74997 331 378 T=7
Note: The descriptive statistics for the variable Electricity to Diesel Price are presented without ln transformation.
Appendix B. Supplementary data
Supplementary data to this article can be found online at https://doi.org/10.1016/j.eneco.2021.105337.
References ACEA, 2011. Tax Guide 2011 (European Automobile Manufacturers Association).
ACEA, 2012. Tax Guide 2012 (European Automobile Manufacturers Association).
13
ACEA, 2014. Tax Guide 2014. European Automobile Manufacturers Association. tors/transport-emissions-of-greenhouse-gases/transport-emissions-of-greenhouse-ga
Available online at. http://www.oyder-tr.org/Content/document/duyuru/ACEA ses-12 (checked on 04/14/2020).
_Tax_%0AGuide_2014.pdf (checked on 12/28/2019). Eurostat, 2019. New Registrations of Passenger Cars by Unloaded Weight. Available
ACEA, 2015. Tax Guide 2015. European Automobile Manufacturers Association. online at. http://appsso.eurostat.ec.europa.eu/nui/show.do?dataset=road_eqr
Available online at. https://www.vda.de/dam/vda/publications/2015/ACE _unlweig&lang=en (checked on 12/28/2019).
A-TAX-GUIDE_%0A2015/ACEATAXGUIDE_2015.pdf (checked on 12/28/2019). Eurostat, 2020a. Electricity Prices for Household Consumers - Bi-Annual Data (From
ACEA, 2016. Tax Guide 2016. European Automobile Manufacturers Association. 2007 Onwards). Available online at. http://appsso.eurostat.ec.europa.eu/nui/show.
Available online at. https://www.acea.be/uploads/news_documents/ACEA_TA do?dataset=nrg_pc_204&lang=en (checked on 06/08/2020).
X_GUIDE_2016.pdf (checked on 12/28/2019). Eurostat, 2020b. Passenger Cars. Available online at. https://appsso.eurostat.ec.europa.
ACEA, 2017. Tax Guide2017. European Automobile Manufacturers Association. eu/nui/show.do?dataset=road_eqs_carage&lang=en (checked on 04/14/2020).
Available online at. https://www.acea.be/uploads/news_documents/ACEA_Tax_Gu Eurostat, 2021. Adjusted Gross Disposable Income of Household Per Capita. Available
ide_2017.pdf (checked on 12/28/2019). online at. https://ec.europa.eu/eurostat/tgm/refreshTableAction.do?tab=table
ACEA, 2018. Consolidated Registrations - By Country. New registrations in European &pcode=tec00113&language=en (checked on 02/11/2021).
Union and EFTA. European Automobile Manufacturers Association. Available online Federal Ministry for the Environment, N. C. and N. Safety, 2019. How Eco-friendly are
at. https://www.acea.be/statistics/article/consolidated-figures-by-country (checked Electric Cars? A Holistic View. Available online at. https://www.bmu.de/en/service
on 12/28/2019). /publications/ (checked on 04/05/2020).
Ajzen, I., 1988. Attitudes, Personality, and Behavior. Dorsey Press. Hall, D., Lutsey, N., 2017. Emerging Best Practices for Electric Vehicle Charging
Ajzen, I., 1991. The theory of planned behavior. Organ. Behav. Hum. Decis. Process. 50 Infrastructure. The International Council on Clean Transportation (ICCT),
(2), 179–211. Washington, DC, p. 54.
An, X., Ganguly, A.R., Fang, Y., Scyphers, S.B., Hunter, A.M., Dy, J.G., 2014. Tracking Hirte, G., Tscharaktschiew, S., 2020. The optimal subsidy on electric vehicles in German
climate change opinions from twitter data. In: Workshop on Data Science for Social metropolitan areas : a spatial general equilibrium analysis. Energy Econ. 40 (2013),
Good, pp. 1–6. 515–528.
Arellano, M., 1987. PRACTITIONERS’ CORNER: computing robust standard errors for Hsiao, C., 2007. Panel data analysis—advantages and challenges. Test 16 (1), 1–22.
within-groups estimators. Oxf. Bull. Econ. Stat. 49 (4), 431–434. IPCC, 2014. Climate Change 2014: Synthesis Report. Contribution of Working Groups I,
Austmann, L.M., 2020. Drivers of the electric vehicle market: a systematic literature II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate
review of empirical studies. Financ. Res. Lett., 101846 https://doi.org/10.1016/j. Change. Technical report. IPCC, Geneva.
frl.2020.101846. Kirilenko, A.P., Stepchenkova, S.O., 2014. Public microblogging on climate change: one
Baker, S.R., Bloom, N., Davis, S.J., 2016. Measuring economic policy uncertainty*. Q. J. year of Twitter worldwide. Glob. Environ. Chang. 26 (1), 171–182.
Econ. 131 (4), 1593–1636. Kirilenko, A.P., Molodtsova, T., Stepchenkova, S.O., 2015. People as sensors: mass media
Baur, D.G., Todorova, N., 2018. Automobile manufacturers , electric vehicles and the and local temperature influence climate change discussion on Twitter. Glob.
price of oil. Energy Econ. 74, 252–262. Environ. Chang. 30, 92–100.
Besova, A.A., Cooley, S.C., 2009. Foreign news and public opinion: attribute agenda- Kovalsky, K.L., Lusk, J.L., 2013. Do consumers really know how much they are willing to
setting theory revisited. Ecquid. Novi 30 (2), 219–242. pay? J. Consum. Aff. 47 (1), 98–127.
Bundesamt für Statistik, 2020. LIK, Durchschnittspreise für Energie und Treibstoffe, Krumpal, I., 2013. Determinants of social desirability bias in sensitive surveys: a
Monatswerte (ab 1993) und Jahresdurchschnitte (ab 1966). Available online at. htt literature review. Qual. Quant. 47 (4), 2025–2047.
ps://www.bfs.admin.ch/bfs/de/home/statistiken/preise/landesindex-konsumenten Kumar, R.R., Alok, K., 2020. Adoption of electric vehicle: a literature review and
preise/lik-resultate.assetdetail.11887635.html (checked on 03/05/2020). prospects for sustainability. J. Clean. Prod. 253.
CEIC, 2020a. Croatia Disposable Hpusehold Income: Avg. Available online at. https Lane, B., Potter, S., 2007. The adoption of cleaner vehicles in the UK: exploring the
://www.ceicdata.com/en/croatia/household-income/disposable-household consumer attitude-action gap. J. Clean. Prod. 15 (11− 12), 1085–1092.
-income-avg (checked on 03/05/2020). Lévay, P.Z., Drossinos, Y., Thiel, C., 2017. The effect of fiscal incentives on market
CEIC, 2020b. Malta Household Income per Capita. Available online at. https://www.cei penetration of electric vehicles: a pairwise comparison of total cost of ownership.
cdata.com/en/indicator/malta/annual-household-income-per-capita (checked on Energy Policy 105, 524–533.
03/05/2020). Li, S., Tong, L., Xing, J., Zhou, Y., 2017. The market for electric vehicles: indirect
Chu, W., Im, M., Song, M.R., Park, J., 2019. Psychological and behavioral factors network effects and policy design. J. Assoc. Environ. Resour. Econ. 4 (1), 89–133.
affecting electric vehicle adoption and satisfaction: a comparative study of early Littman, J., Wrubel, L., 2019. Climate Change Tweets Ids. Available online at. https
adopters in China and Korea. Transp. Res. Part D: Transp. Environ. 76 (September), ://doi.org/10.7910/DVN/5QCCUU (checked on 03/16/2020).
1–18. Martin, D.M., Väistö, T., 2016. Reducing the attitude-behavior gap in sustainable
Clinton, B.C., Steinberg, D.C., 2019. Providing the Spark: impact of financial incentives consumption: a theoretical proposition and the American electric vehicle market.
on battery electric vehicle adoption. J. Environ. Econ. Manag. 98, 1–18. Rev. Market. Res. 13, 193–213.
Cody, E.M., Reagan, A.J., Mitchell, L., Dodds, P.S., Danforth, C.M., 2015. Climate change Mersky, A.C., Sprei, F., Samaras, C., Qian, Z.S., 2016. Effectiveness of incentives on
sentiment on Twitter: an unsolicited public opinion poll. PLoS One 10 (8). electric vehicle adoption in Norway. Transp. Res. Part D: Transp. Environ. 46, 56–68.
Coffman, M., Bernstein, P., Wee, S., 2017. Electric vehicles revisited: a review of factors Mitchell, A.A., Olson, J.C., 1981. Are product attribute beliefs the only mediator of
that affect adoption. Transp. Rev. 37 (1), 79–93. advertising effects on brand attitude? J. Mark. Res. 18 (3), 318–332.
Cohen, J., Azarova, V., Kollmann, A., Reichl, J., 2019. Q-complementarity in household Molodtsova, T., Kirilenko, A., Stepchenkova, S., 2013. Utilizing the social media data to
adoption of photovoltaics and electricity-intensive goods: the case of electric validate ‘climate change’ indices. In: AGU Fall Meeting Abstracts, 2013 (pp.
vehicles. Energy Econ. 83, 567–577. GC11D–1039).
Coronese, M., Lamperti, F., Keller, K., Chiaromonte, F., Roventini, A., 2019. Evidence for Moser, A.K., 2015. Thinking green, buying green? Drivers of pro-environmental
sharp increase in the economic damages of extreme natural disasters. Proc. Natl. purchasing behavior. J. Consum. Market. 32 (3), 167–175.
Acad. Sci. U. S. A. 116 (43), 21450–21455. Münzel, C., Plötz, P., Sprei, F., Gnann, T., 2019. How large is the effect of financial
Daimler, 2020. Mercedes-Benz Models With Environmental Certificates. Available online incentives on electric vehicle sales?–a global review and european analysis. Energy
at. https://www.daimler.com/sustainability/environmental-certificates/ (checked Econ. 104493.
on 05/06/2020). Nie, Y., Wang, E., Guo, Q., Shen, J., 2018. Examining Shanghai consumer preferences for
Egnér, F., Trosvik, L., 2018. Electric vehicle adoption in Sweden and the impact of local electric vehicles and their attributes. Sustainability (Switzerland) 10 (6), 1–16.
policy instruments. Energy Policy 121 (November 2017), 584–596. Okada, T., Tamaki, T., Managi, S., 2019. Effect of environmental awareness on purchase
European Commission, 2010. EU Energy in Figures. Statistical Pocketbook 2010. intention and satisfaction pertaining to electric vehicles in Japan. Transp. Res. Part
Available online at. https://ec.europa.eu/energy/data-analysis/energy-statistical-po D: Transp. Environ. 67 (February), 503–513.
cketbook_en (checked on 02/15/2020). Olteanu, A., Castillo, C., Diakopoulos, N., Aberer, K., 2015. Comparing events coverage
European Commission, 2013. EU Energy in Figures. Statistical Pocketbook 2013. in online news and social media: The case of climate change. In: Proceedings of the
Available online at. https://ec.europa.eu/energy/data-analysis/energy-statistical-po 9th International Conference on Web and Social Media, ICWSM 2015 (May),
cketbook_en (checked on 02/15/2020). pp. 288–297.
European Commission, 2014. EU Energy in Figures. Statistical Pocketbook 2014. Plante, M., 2019. OPEC in the news. Energy Econ. 80, 163–172.
Available online at. https://ec.europa.eu/energy/data-analysis/energy-statistical-po Plötz, P., Schneider, U., Globisch, J., Dütschke, E., 2014. Who will buy electric vehicles?
cketbook_en (checked on 02/15/2020). Identifying early adopters in Germany. Transp. Res. A Policy Pract. 67, 96–109.
European Commission, 2015. EU Energy in Figures. Statistical Pocketbook 2015. Reboredo, J.C., Ugolini, A., 2018. The impact of Twitter sentiment on renewable energy
Available online at. https://ec.europa.eu/energy/data-analysis/energy-statistical-po stocks. Energy Econ. 76, 153–169.
cketbook_en (checked on 02/15/2020). Ritchie, H., Roser, M., 2020. Natural Disasters. Available online at. https://ourworl
European Commission, 2018. EU Energy in Figures. Statistical Pocketbook 2018. dindata.org/natural-disasters (checked on 02/20/2020).
Available online at. https://ec.europa.eu/energy/data-analysis/energy-statistical-po Samantray, A., Pin, P., 2019. Data and Code For: Credibility of Climate Change Denial in
cketbook_en (checked on 02/15/2020). Social Media. Available online at. https://doi.org/10.7910/DVN/LNNPVD (checked
European Commission, 2019. The European Green Deal. Available online at. https://ec. on 01/15/2020).
europa.eu/info/sites/info/files/european-green-deal-communication_en.pdf Sampei, Y., Aoyagi-Usui, M., 2009. Mass-media coverage, its influence on public
(checked on 03/27/2020). awareness of climate-change issues, and implications for Japan’s national campaign
European Commission, 2021. European Alternative Fuel Observatory (EAFO). Available to reduce greenhouse gas emissions. Glob. Environ. Chang. 19 (2), 203–212.
online at. https://www.eafo.eu/ (checked on 02/11/2021). Santi, C., 2021. Investors’ climate sentiment and financial markets. https://doi.org/
European Environment Agency, 2019. Greenhouse Gas Emissions From Transport in 10.2139/ssrn.3697581. Available at.
Europe. Available online at. https://www.eea.europa.eu/data-and-maps/indica
14
Schlüter, J., Weyer, J., 2019. Car sharing as a means to raise acceptance of electric Veltri, G.A., Atanasova, D., 2017. Climate change on Twitter: content, media ecology and
vehicles: An empirical study on regime change in automobility. Transport. Res. F 60, information sharing behaviour. Public Underst. Sci. 26 (6), 721–737.
185–201. Vergis, S., Chen, B., 2015. Comparison of plug-in electric vehicle adoption in the United
Shapiro, M.A., Hemphill, L., 2017. Politicians and the policy agenda: does use of twitter States: a state by state approach. Res. Transp. Econ. 52 (December 2010), 56–64.
by the U.S. Congress direct New York Times content? Policy Internet 9 (1), 109–132. Volkswagen, 2020. AdBlue: Selective Catalytic Reduction (SCR). Available online at.
Sheldon, T.L., DeShazo, J.R., 2017. How does the presence of HOV lanes affect plug-in https://www.volkswagen.co.uk/technology/engines/adblue (checked on 05/06/
electric vehicle adoption in California? A generalized propensity score approach. 2020).
J. Environ. Econ. Manag. 85, 146–170. Wang, Z., Dong, X., 2016. Determinants and policy implications of residents’ new energy
Sierzchula, W., Bakker, S., Maat, K., Van Wee, B., 2014. The influence of financial vehicle purchases: the evidence from China. Nat. Hazards 82 (1), 155–173.
incentives and other socio-economic factors on electric vehicle adoption. Energy Wang, N., Tang, L., Pan, H., 2018. Analysis of public acceptance of electric vehicles: an
Policy 68, 183–194. empirical study in Shanghai. Technol. Forecast. Soc. Chang. 126 (October 2017),
Smith, B., Olaru, D., Jabeen, F., Greaves, S., 2017. Electric vehicles adoption: 284–291.
environmental enthusiast bias in discrete choice models. Transp. Res. Part D: Transp. Wang, N., Tang, L., Pan, H., 2019. A global comparison and assessment of incentive
Environ. 51, 290–303. policy on electric vehicle promotion. Sustain. Cities Soc. 44 (January 2019),
Statista, 2020. Annual Number of Natural Disaster Events Globally From 2000 to 2019. 597–603.
Available online at. https://www.statista.com/statistics/510959/number-of-natura White, H., 1980. A heteroskedasticity-consistent covariance matrix estimator and a direct
l-disasters-events-globally/ (checked on 04/14/2020). test for heteroskedasticity. Econometrica 48 (4), 817–838.
Statistics Norway, 2018. Prices on Engine Fuel (NOK Per Litres), by Petroleum Products, Wooldridge, J.M., 2010. Econometric Analysis of Cross Section and Panel Data. MIT
Contents and Month. Available online at. http://www.ssb.no/en/statbank/sq/%0A Press.
10005655/ (checked on 03/05/2020). Wooldridge, J.M., 2016. Introductory Econometrics: A Modern Approach. Nelson
Tien, S., 2018. Top Twitter Demographics That Matter to Social Media Marketers. Education.
Available online at. https://blog.hootsuite.com/twitter-demographics/ (checked on Wu, G., Inderbitzin, A., Bening, C., 2015. Total cost of ownership of electric vehicles
03/10/2020). compared to conventional vehicles: a probabilistic analysis and projection across
Trommer, S., Jarass, J., Kolarova, V., 2015. Early adopters of electric vehicles in market segments. Energy Policy 80, 196–214.
Germany unveiled. World Electr. Vehicle J. 7 (4), 722–732. Zambrano-Gutiérrez, J.C., Nicholson-Crotty, S., Carley, S., Siddiki, S., 2018. The role of
Tsakalidis, A., Thiel, C., 2018. Electric Vehicles in Europe from 2010 to 2017: Is Full- public policy in technology diffusion: the case of plug-in electric vehicles. Environ.
scale Commercialisation Beginning. Publications Office of the European Union, Sci. Technol. 52 (19), 10914–10922.
Luxembourg. de Rubens, Z.G., 2019. Who will buy electric vehicles after early adopters? Using
Tu, J.C., Yang, C., 2019. Key factors influencing consumers’ purchase of electric vehicles. machine learning to identify the electric vehicle mainstream market. Energy 172,
Sustainability (Switzerland) 11 (14). 243–254.
Twitter, 2019. Annual Report 2019. Available online at. https://investor.twitterinc.co Zhang, X., Bai, X., Shang, J., 2018. Is subsidized electric vehicles adoption sustainable:
m/financial-information/annual-reports/default.aspx (checked on 05/20/2020). consumers’ perceptions and motivation toward incentive policies, environmental
United Nations Climate Change, 2020. Paris Agreement - Status of Ratification. Available benefits, and risks. J. Clean. Prod. 192, 71–79.
online at. https://unfccc.int/process/the-paris-agreement/status-of-ratification Zhuge, C., Shao, C., 2019. Investigating the factors influencing the uptake of electric
(checked on 02/02/2020). vehicles in Beijing, China: statistical and spatial perspectives. J. Clean. Prod. 213
Van Den Bossche, P., Turcksin, T., Omar, N., Van Mierlo, J., 2016. Developments and (February 2018), 199–216.
challenges for EV charging infrastructure standardization. World Electr. Vehicle J. 8 Ziegler, A., 2012. Individual characteristics and stated preferences for alternative energy
(2), 557–563. sources and propulsion technologies in vehicles: a discrete choice analysis for
Germany. Transp. Res. A Policy Pract. 46 (8), 1372–1385.
15

Does Environmental Awareness Fuel The Electric Vehicle Market A Twitter Keyword Analysis

Uploaded by

Copyright:

Available Formats

Does Environmental Awareness Fuel The Electric Vehicle Market A Twitter Keyword Analysis

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Does Environmental Awareness Fuel The Electric Vehicle Market A Twitter Keyword Analysis

Uploaded by

Copyright:

Available Formats

Energy Economics 101 (2021) 105337

Contents lists available at ScienceDirect

Does environmental awareness fuel the electric vehicle market? A Twitter

through monetary incentives but also through non-monetary incentives

Variable Country (if divergent) Source

EV registrations European Commission (2021)

Norway Statistics Norway (2018)

Global Disasters Statista (2020)

Variable Fixed effects First differences

Environmental awareness − 100%*** 7.47428 − 100% 10.88801

Table A.2 (continued )

Fast chargers − 0.0215% 0.0010322 0.0470%* 0.0002699

Environmental awareness Global disasters Fatalities

Environmental awareness 1.0000

Variable Fixed effects First differences

Environmental awareness 20,288,199.1% 14.2629 3,121,142.5% 15.31784

Variable Fixed effects First differences

Environmental awareness − 91.07% 16.14028 − 75% 19.67168

Table A.5 (continued )

F-score 79.09*** 115.30***

Variable w/o charger w/o charger total w/o charger fast

Environmental awareness 4,594,444.53% 14.71364 3,889,162.8% 14.61002 10,842,899.1% 14.32455

Variable Poland Portugal Romania Slovakia Slovenia Spain Sweden Switzerland UK

Table A.7 (continued )

Variable Random effects First differences

Environmental awareness 40,199.66% 13.27839 6354% 15.37895

Country Search term English translation

Austria Klimawandel Climate change

Variable Fixed effects First differences

Environmental awareness 1.0212% 0.0089458 − 0.4886% 0.0100661

Variable Mean Std. Dev. Min Max Obervations

EV registration share Overall 0.0115543 0.0396366 0 0.3918689 N = 203

Appendix B. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.eneco.2021.105337.

You might also like

F-score 79.09* 115.30*