Renewable and Sustainable Energy Reviews 142 (2021) 110874

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews

journal homepage: http://www.elsevier.com/locate/rser

Electric vehicles and consumer choices

F. Mandys
School of Economics, University of Surrey, United Kingdom

A R T I C L E I N F O A B S T R A C T

Keywords: Electric vehicles (EVs) have many attractive features compared to conventional vehicles (CVs). Their main
Alternative fuel vehicle advantage lies in their significant economic benefits due to the much higher fuel efficiency, and substantial
Consumer behavior environmental benefit of lower greenhouse gas emissions. However, the market share of EVs in the UK remains
Discrete choice model
low and their benefits will not be realized unless the government and the manufacturers can gain crucial in­
Lasso
Socio-technical barrier
formation, necessary to effectively support and speed up the adoption of EVs. Such contemporary information for
Early adopter the UK is missing in the literature, and this study aims to fill the gap. Therefore, a stated preferences UK dataset is
used, and a discrete choice mode is applied, using an adaptive Lasso methodology, binomial logit and ordered
logit regressions. The main goals include: finding the characteristics of potential early adopters of EVs in the UK,
the vehicle attributes that they consider important for their buying decisions, and the key barriers that slow EV
adoption. The results suggest that the propensity of being a potential EV early adopter increases with youth,
education, being a student, living in the more southern parts of UK, being married and, to a lesser extent, income.
Additionally, purchase cost, performance, maximum range and environmental friendliness are found to be
important vehicle attributes for the potential buyers. Furthermore, two key barriers to wide EV adoption are
identified – high purchase cost and low maximum range of the vehicle.

and environmental benefits, caused by their cheaper fuel and lower

emissions [3]. Even the emissions of HEVs are about half of standard CVs
1. Introduction [4]. Therefore, a large-scale adoption of EVs can reduce transport
emissions, related local health risks, slow down global warming and
The very first attempts at producing an electric vehicle (EV) came as promote the use of renewable energy [5,6], and should thus be pro­
early as the 1830s, with the first practical car built in London by Thomas moted by the world governments.
Parker in 1884. However, EVs were very soon driven out of the market Nevertheless, both manufacturers and governments have only
by vehicles running on fossil fuels. Recently, commercial EVs are making limited amount of critical information necessary to effectively support a
a comeback (due to advances in battery and hybrid technologies, as well large-scale and successful EV diffusion. This includes, for example, the
as environmental concerns [1]), with their bright future being predicted knowledge of consumers’ wants, what will make them consider a switch
by the manufacturers, governments and many researchers. For example, from CVs, or the most serious barriers that slow the diffusion of EVs. This
BloombergNEF [2] predicts that the new car sales of EVs will grow from kind of information is crucial to manufacturers, not only for optimal
2.7% in 2020, to 58% by 2040, France wants to end the sale of all design purposes, but also to optimize their marketing and advertising
fossil-fuel based vehicles by 2040, and UK considers banning the sale of strategies for the correct consumer groups. Governments also value this
all cars that are unable to travel at least 50 miles on battery power by information highly for policy purposes, educational purposes, or, for
2040, and reach net-zero emissions by 2050. example, for support programs targeted at a certain typology of con­
General alternative fuel vehicles (AFVs) can be separated into several sumers. The described contemporary information for the UK is currently
different categories, but primarily these are hybrid electric vehicles missing in the literature, and this paper aims to fill this important gap.
(HEVs) and battery electric vehicles (BEVs). Hybrids run on a combi­ Therefore, vehicle aspects, vehicle usage, and consumer profiles in the
nation of gasoline and electric engines and are therefore cleaner than UK are analyzed, using consumer stated preferences survey data and a
conventional vehicles (CVs), while BEVs are fully clean, as they do not discrete choice model. The main goals of the study include: firstly
possess any form of combustion engine. Due to lack of data, HEV’s and finding the most important characteristics of potential EV consumers.
BEV’s are treated as one category of EVs in this paper. Attractiveness of Secondly, finding which vehicle characteristics chiefly affect the
EVs for governments and consumers lies in their significant economic

E-mail addresses: f.mandys@surrey.ac.uk, f.mandys@surrey.ac.uk.

https://doi.org/10.1016/j.rser.2021.110874
Received 3 July 2020; Received in revised form 9 February 2021; Accepted 22 February 2021
Available online 3 March 2021
1364-0321/© 2021 Elsevier Ltd. All rights reserved.
F. Mandys Renewable and Sustainable Energy Reviews 142 (2021) 110874

buying EVs in the next few years. This first large group of consumers is
Abbreviations usually considered highly influential by researchers in deciding the
success of a new technology, according to Plötz et al. [20]. Their results
AFV alternative fuel vehicle show that the most likely early adopters in Germany are men of middle
AIC Akaike information criterion age, in technical professions, and living in rural or suburban households.
BEV battery electric vehicle These people travel many kilometers annually, benefit from the lower
BIC Bayesian information criterion EV operational costs, and have the funds to purchase an expensive
CV conventional vehicle vehicle. Hackbarth and Madlener [21] similarly focus on German con­
EV electric vehicle sumers, but rather find that potential EV buyers are younger, highly
GCSE General Certificate of Secondary Education educated and environmentally conscious individuals, as in Hidrue et al.
HEV hybrid electric vehicle [12], He et al. [22], or more recently in Higgins et al. [23], using their
IIA independence from irrelevant alternatives multivariate analysis of variance. Similarly, Lane et al. [24] find that EVs
Lasso least absolute shrinkage and selection operator appeal significantly more to individuals who think about them in terms
OLS ordinary least squares of their technical and environmental components, compared to the more
ONC Ordinary National Certificate traditional households.
PHEV plug-in hybrid electric vehicle The socio-technical1 barriers to an extensive EV diffusion were
R&D research and development explored by Egbue and Long [5]. They find that the major obstacles
include high battery costs contributing to the already high price of EVs,
and the lacking charging infrastructure, similarly to Axsen et al. [25]
and Li et al. [26]. Globisch et al. [27] discovered that improvements to
consumers’ desire to buy EVs. Thirdly, analyzing how these aspects the public charging infrastructure could attract more “mainstream”
change with different consumer characteristics. And fourthly, identi­ buyers, not just early adopters. Further point of Egbue and Long [5] is
fying the key barriers to EV adoption for potential EV consumers. that consumers generally tend to resist new, unproven technologies.
The rest of this paper is organized as follows. Section 2 reviews the Therefore, incentives such as subsidies or fuel taxes will have limited
main strands of relevant literature. Section 3 presents the dataset used. effect and governments should rather focus on education of consumers
Section 4 describes the econometric methods. Section 5 presents the of e.g. the non-financial EV benefits (such as environmental). Similarly,
results and discusses the findings. Section 6 concludes the paper. Franke and Krems [28] discover evidence that consumers generally
demand a significantly larger range potential in their vehicles, than they
2. Literature review typically use in the end. Li et al. [26] find that even with the introduction
of newer EVs, low maximum range generally remains a major obstacle
Evaluation of consumer choices of EVs represents a dynamic and for many consumers. However, hands-on experience with EVs and
contemporary area of literature. While the topic is relatively new, and government programs, such as gasoline taxes, tend to have a signifi­
there are still many areas for potential research, there exists a rich cantly positive effect on EV uptake [3,16].
literature, some of which was well evaluated by e.g. Coffman et al. [7], Typically, the most widely used methodology of the discussed studies
Liao et al. [8], or Hardman et al. [9], in their literature reviews. One of is the application of a discrete choice model, where the decision of
the early works is the paper by Beggs et al. [10]. They attempted to buying/switching to an EV is described as a choice between two or more
assess which vehicle aspects are the most important for the demand for vehicle alternatives [8]. In most case, stated preference data is used,
EVs. Their main finding was that consumers put a very high negative mainly due to limited presence of EVs in many markets, causing lack of
valuation on limited vehicle range, low performance and long refueling market data. This means that consumer actions may be only hypothet­
period, and surprisingly low importance on operating costs. Ewing and ical, and the questionnaires may provide large number of explanatory
Sarigöllü [11] revisited this topic, exploring how the preferences variables. The typical identification problems of overfitting and multi­
respond to improved performance and government regulation. Simi­ collinearity are then solved by either using a less detailed, more specific
larly, they discover that consumers are most responsive towards per­ survey, or using a method, such as principal component analysis [22], in
formance attributes, in particular range, acceleration and refueling time. order to extract only the important variables. Modeling-wise, the current
Government regulations, incentives, and environmental benefits were standard in the literature is to use some form of either a logit or a probit
found to be less important, as also concluded by e.g. Graham-Rowe et al. model [23]. This often includes a multinomial logit [29] which assumes
[6]. Results of Hidrue et al. [12], support the conclusions of Ewing and independence from irrelevant alternatives2 (IIA), or a nested logit which
Sarigöllü [11], stressing that higher fuel costs and operational savings relaxes this assumption and allows for the correlation between various
are mostly ignored by consumers, who rather focus on purchase price, as alternatives [14,30,31]. Accounting for taste parameters using a mixed
in Larson et al. [13]. Similarly to Huang and Qian [14] in China, they logit [17,32], or a latent factor identifying hybrid choice model of
found that high price, particularly due to expensive batteries, is one of Ben-Akiva [33], is also popular [19,34].
the main concerns of potential EV consumers. An alternative result was
reached by Krupa et al. [15] using a modified agent-based model of 3. Data
Eppstein et al. [16], finding that 86% of consumers consider potential
fuel savings as a crucial EV attribute. Similar conclusions are reached in This research uses a stated preferences data set created by combining
a recent study by Li et al. [17], finding that any incentives that reduce and quantifying the answers to a UK survey run in 2014 and again
operation costs should in turn increase the adoption rates. Nevertheless,
in line with Larson et al. [13], they found that consumers are not willing
to pay a significant premium for EVs. The analysis of Larson et al. [13]
1
also revealed that consumers that were exposed to EV’s are less sensitive The term socio-technical includes social, political, economic, cultural and
to previously found critical attributes, such as range [18]. Liao et al. [19] technological barriers.
2
furthermore discovered that the type of EV business model affects the An axiom stating that if there exists some alternative x that is chosen from a
set S, and the alternative x is also an element in the subset A of S, then the
attractiveness of EV adoption, with vehicle leasing being the most
alternative x must be the one chosen from the subset A. This implies that
attractive.
consumers are homogeneous.
One of the primary groups of consumers that policymakers and
manufacturers are interested in are the early adopters – the group

2
F. Mandys Renewable and Sustainable Energy Reviews 142 (2021) 110874

in20153 (with different individuals compared to 2014). Both of the insensitive to spacing … except in the most extreme cases” [37]. Thus,
survey years come from the UK Data Service catalogue, specifically the even variables without equal spacing are generally constructed as
Electric Vehicles Module of the Opinions and Lifestyle Survey. continuous. Thirdly, the survey answers provide each respondent’s
Combining and then quantifying the survey answers yields a pooled ranking of several vehicle attributes (e.g. performance or range) and key
cross-section dataset with two time periods of 2014 and 2015. The barriers, according to their perceived importance. Each of these vehicle
Opinions and Lifestyle Survey is a multipurpose social survey, carried attributes is thus constructed as a numerical variable where, for
out by the Office for National Statistics, while the Electric Vehicle example, if there are 7 vehicle attributes in a particular question, the
Module is a more topic-specific set of questions from the Department for most important attribute chosen would be labelled with 7, the second
Transport, that are included into the core questions which cover mainly one would be labelled with 6, and so on for all of the attributes. It is
demographic topics. The module contains information about the re­ assumed that if a person did not choose an attribute, that attribute is
spondents’ vehicle buying priorities and their considerations of possibly unimportant to them, and thus is labelled with a 0. After performing the
purchasing an EV vs. a CV. The module surveyed and interviewed UK described operations, the constructed dataset contains 65 variables,
adults aged 16 years or older and living in private households, using a where the descriptive statistics can be seen in Table 1 below.
multi-stage stratified random sampling procedure.4 The results from
2014 to 2015 contain 962 and 1034 respondents respectively, yielding a
total of 1996 observations; however, due to missing answers, this de­
creases to 1347 observations for the final dataset. The survey contains
76 questions about consumer attributes (e.g. age or income), and vehicle Table 1
attributes (e.g. ratings of performance or maximum range). In the final Descriptive statistics of chosen important variables from the constructed dataset.
dataset, 28 unique5 questions are used for variable creation. Variable Description Mean Standard Minimum Maximum
Dev.
3.1. Variable construction Respondent thought about EV 0,25 0,43 0 1
purchase
The main dependent variable is constructed from the survey ques­ Discouraged potential EV 0,21 0,40 0 1
adopter
tion: “Which statement best describes your attitude towards buying an
Surveyed in 2015 0,52 0,50 0 1
electric car”. The answers are separated into two groups. Firstly, in­ Respondent Demographics
dividuals who have thought or are thinking of buying an EV and thus can Income 20 15 169 0 54 600
be considered more likely to buy an EV in the future, and secondly, those 460
who either do not want to buy an EV at the moment or who have not Household size 2,44 1,26 1 9
Male 0,51 0,50 0 1
thought of buying it, and thus are unlikely to purchase an EV in the near Age 49,94 17,39 16 93
future. The main dependent variable is a binary variable, where the Married 0,54 0,50 0 1
individuals who thought about buying an EV are labelled as 1, and those No. of vehicles in the 1,38 0,78 0 3
who have not are labelled as 0. The secondary dependent variable, used household
Health 4,10 0,88 1 5
to identify the key barriers to greater EV diffusion, is constructed in a
Region of Habitude
similar way as a binary variable. Specifically it identifies the discour­ North of England 0,26 0,44 0 1
aged potential early adopters, by labeling individuals who thought of an Middle of England 0,23 0,42 0 1
EV purchase but for some reason decided not to at that point in time (i.e. South of England 0,34 0,47 0 1
encountered a barrier to adoption) as 1, and the remaining individuals Scotland 0,07 0,25 0 1
Education Level Attained
as 0. Higher education 0,39 0,49 0 1
Most variables for the dataset are created as dummy variables from A-Levels equivalent 0,15 0,36 0 1
the survey answers, however, three variable types were constructed in a Below A-Levels 0,22 0,41 0 1
more specific manner. Firstly, the survey answers about income came in Other qualification 0,11 0,31 0 1
Current Job
the form of income bands, and to construct a single continuous variable,
Full-time 0,56 0,50 0 1
the middle point of the band was assigned to each individual. Since the Managerial and 0,35 0,48 0 1
number of income bands is very large, and the size of the intervals is very professional
low, this method of variable creation stays as true to the data as possible. Intermediate 0,11 0,31 0 1
Secondly, answers about the health of respondents came in the form of a Small employers 0,08 0,28 0 1
Lower supervisory and 0,06 0,24 0 1
Likert scale,6 and since the answer categories were equally spaced, this Technical
ordinal variable can be treated as continuous [35], as in Agresti [36]. Routine and semi-routine 0,16 0,37 0 1
Nevertheless, even in the case where the spacing is not equal across Student 0,05 0,21 0 1
categories, many researchers make a strong case on treating ordinal Ratings of Vehicle Attributes and Barriers
Comfort 3,20 3,33 0 7
variables as continuous [35,37], since “the results are remarkably
Positive environmental 2,19 2,68 0 7
impact
Vehicle being fully electric 0,23 1,04 0 7
3 Style 1,68 2,42 0 7
The survey has been run only in the two years – 2014 and 2015. As of 2021,
no further survey years have been published by the ONS. Interior space 1,71 2,34 0 7
4 Reliability 3,48 2,42 0 7
A sampling procedure where members of the population are divided into
Safety 2,28 2,25 0 7
homogeneous subgroups before sampling, where each subgroup is mutually
Performance 0,29 1,06 0 7
exclusive. Samples are then taken in stages using increasingly smaller sampling Price 1,28 2,09 0 5
units. Recharging cost 0,54 1,42 0 5
5
The survey contains 76 questions, but many are inquiring about the same Maintenance cost 0,57 1,45 0 5
subject with different groups of possible answers. E.g., 2 questions asking about Lack of choice 0,45 1,40 0 5
age, with different age categories as possible answers. As such, there were in the Lack of knowledge about 0,67 1,65 0 5
end 28 questions about unique, non-repeating subjects. technology
6
A type of symmetric scale typically used in questionnaires. It most Maximum range 1,61 2,17 0 5
Recharging convenience 1,70 2,14 0 5
frequently provides answers to a question in the format “strongly disagree,
Technology proven 0,30 1,10 0 5
disagree, neutral, agree, strongly agree”.

3
F. Mandys Renewable and Sustainable Energy Reviews 142 (2021) 110874

4. Methodology (AIC), the Bayesian information criterion10 (BIC), or cross-validation.

Cross-validation can be seen as a middle ground between AIC and BIC,
4.1. Characteristics of early EV adopters and importance of vehicle and thus 10 rounds of cross-validation are used to determine the tuning
attributes parameter. However, although that the Lasso has many desirable
properties, it suffers from a handful of issues, such as lacking the oracle
The first and second goal of this research is to find the most impor­ properties11 [38]. The method is also sensitive to the selection of the
tant characteristics of the potential UK EV consumers, and to find out tuning parameter λn . Therefore, a refined version of the Lasso is used
which vehicle attributes are the most important to them. As the main instead – the adaptive Lasso.
dependent variable is a binary variable, a binomial logit regression of
this dependent variable is applied, on all the consumer and vehicle 4.1.2. The adaptive lasso
variables that were constructed in section 3. The logit is a well- The adaptive Lasso method does have the oracle properties for a
established method, and provides more intuitive inference on the co­ suitable choice of the tuning parameter, and thus performs correct
efficients than, for example, a probit regression.7 However, there is an variable selection. The method is also much less sensitive to the choice of
issue that the number of attributes is very large, which causes problems the tuning parameter and estimates the non-zero coefficients with the
of overfitting, multicollinearity, and can lead to parameter estimation same efficiency as the least-squares estimator. The procedure of getting
issues and subsequent problems in interpretation [22]. To improve the the adaptive Lasso estimator is very similar to the ordinary Lasso:
estimation, an important step is to build a model that only includes p
∑
n
( )2 ∑
important attributes. Researchers, such as e.g. He et al. [22], used
AL
(3)
′
̂
β n = argmin yj − u xj + λn λn,i |ui |
principal component analysis to alleviate these issues. However, the
u
j=1 i=1

final extracted principal components are not the same as the original
variables, and thus interpreting the results of the analysis of the prin­ where λn,i = ( ⃒⃒ 1 ⃒⃒)γ is the adaptive weights vector, ̂
β n,i is an initial es­
̂β n,i
cipal components and drawing conclusions from them can often be
problematic. Therefore, this paper applies the least absolute shrinkage timate of the coefficients, and γ is a positive constant for adjustment of
and selection operator (Lasso) to avoid these issues. the adaptive weights vector, set between 13 and 10 3.
The main difference between the standard Lasso and the adaptive
4.1.1. The Lasso Lasso comes from the adaptive weights vector λn,i , which applies
The Lasso is an econometric regularization8 method, that is able to different regularization for each coefficient. This means that the penal­
combine parameter estimation and variable selection into a single step. ization is adjusted for each coefficient and provides a stronger penali­
The method shrinks the coefficients of those variables that are irrelevant zation to those coefficients that are smaller. The initial estimate of the
for the research to exactly zero, thus performing a selection of attributes. coefficients for the adaptive weights vector is acquired using a standard
Those variables that are not shrunk to zero by Lasso can be considered ordinary least squares (OLS) regression. These coefficients are then
important for the research and can be used further in the subsequent applied to the data points of each consumer profile variable. Therefore,
binomial logit regression. Considering the linear regression model for while in the standard Lasso, the relationship and the matrix of inde­
this study: pendent variables looks the following:
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
yj = β’ xj + εj , j = 1, 2, …, n (1) y1 x1,1 x1,2 … x1,p ε1
⎢ y2 ⎥ ⎢ x2,1 x2,2 … x2,p ⎥ ⎢ ε2 ⎥
⎢ ⎥=⎢ ⎥+⎢ ⎥ (4)
where yj is the dependent variable, xj = (xj,1 , xj,2 , …, xj,p ) is the vector of ⎣⋮ ⎦ ⎣⋮ ⋱ ⋮ ⎦ ⎣⋮ ⎦
′
⋮
independent variables, β = (β1 , β2 , …, βp ) are the unknown coefficients
′ yn xn,1 xn,2 … xn,p εn
of the independent variables, εj is the error term, n is the no. of obser­
in the adaptive Lasso, the matrix is adjusted by the coefficients from the
vations, and p is the number of consumer and vehicle variables. initial OLS regression:
L
then the Lasso estimator ̂
β n is defined as: ⎡⃒ ⃒ ⃒ ⃒ ⃒ ⃒ ⎤
⎡ ⎤ ⃒̂β 1 ⃒x1,1 ⃒̂
β 2 ⃒x1,2 … β p ⃒x1,p
⃒̂ ⎡ ⎤
L ∑
n
( )2 p
∑ y1 ⎢ ⃒ ⃒ ⃒ ⃒ ⃒ ⃒ ⎥ ε1
(2)
′
̂
β n = argmin yj − u xj + λn |ui | ⎢ y2 ⎥ ⎢ ⃒̂ ⃒ ⃒̂
β 2 ⃒x2,2 β p ⃒x2,p ⎥
⃒̂ ⎢ ε2 ⎥
u ⎢ ⎥ = ⎢ β 1 x2,1 …
⎥+⎢ ⎥ (5)
j=1 i=1 ⎣⋮ ⎦ ⎣ ⎦ ⎣⋮ ⎦
yn ⃒⋮ ⃒ ⋮
⃒ ⃒ ⋱ ⋮
⃒ ⃒ εn
β 1 ⃒xn,1
⃒̂ ⃒̂
β 2 ⃒xn,2 … β p ⃒xn,p
⃒̂
where λn > 0 is a tuning parameter, and ui are the estimated coefficients.
The most important parameter of equation (2) is the tuning param­
eter λn , which determines the level of penalization that is applied, i.e. After applying 10-rounds of cross validation to determine the
what kind of relevance a variable has to have in order for its coefficient optimal tuning parameter, and checking for multicollinearity issues, 26
to not be shrunk to zero. If the tuning parameter is set too high, most final variables are found to be important and thus are used in the sub­
coefficients will be eliminated and thus the analysis of the effects on the sequent regressions.
dependent variable can be limited. If the tuning parameter is set too low,
almost no coefficients will be eliminated, hindering what was desired in 4.1.3. Binomial logistic regression
the first place – elimination of coefficients. The tuning parameter can be In order to find the characteristics of potential EV adopters and the
determined using three main methods: the Akaike information criterion9 importance of various vehicle attributes, the relevant variables from the
adaptive Lasso are used in a binomial logit regression:
∑
L ∑
K ∑
R
yj = β 0 + βl zj,l + βk wj,k + βr cj,r + εj , j = 1, 2, …, n (6)
7
The logit regression coefficients (log odds) can be transformed into odds l=1 k=1 r=1
ratios. The odds ratios allow for direct intuitive inference of the results, as
opposed to the coefficients of the probit regression.
8
The process of regularization introduces additional information, in order to
10
prevent the problem of overfitting. Closely related to the AIC. The BIC penalizes the number of parameters
9
An estimator that estimates the quality of a model, relative to each of the more than the AIC.
11
other models, and thus provides means for model selection. The model with the Oracle properties state that a procedure has to identify the right subset of
minimum AIC value, relative to other models, is selected. true variables, with optimal estimation rate.

4
F. Mandys Renewable and Sustainable Energy Reviews 142 (2021) 110874

where yj is the dependent variable equal to 1 if a respondent is a po­ consumer attributes. As in the previous parts, the adaptive Lasso is used
tential EV buyers and 0 otherwise, zj,l are consumer variables, wj,k are before each regression:
vehicle variables, cj,r are control variables, βl are coefficients of con­ ∑
B ∑
R

sumer variables, βk are coefficients of vehicle variables, βr are co­ qd = β0 + βb kd,b + βr cd,r + εd , d = 1, 2, …, D (8)
efficients of control variables, εj is the error term, n is the number of
b=1 r=1

observations, L is the number of consumer variables, K is the number of where qd is the discouraged adopter dependent variable, kd,b are the
vehicle variables, and R is the number of control variables. barrier variables, cd,r are the consumer attribute control variables, βb are
The binomial logit regression provides coefficients βl of the con­ coefficients of the barrier variables, βr are coefficients of the control
sumer variables, allowing for interpretation of their effect on the EV variables, εd is the error term, D is the number of observations, B is the
dependent variable, as well as coefficients of the vehicle attributes βk , number of barrier variables, and R is the number of control variables.
providing means of inference of each variable’s importance for potential The regression above provides coefficients of the barrier variables,
EV consumers. Positive and significant variables (i.e. variables with and consequently their effect on the discouraged potential EV adopters.
coefficients significant at least at a 90% level) will increase the chances Thus, if a barrier variable has a significant (coefficient significant at least
of the individual to have thought about EV purchase, and thus, the at a 90% level) and positive effect, an individual that ranked this vari­
characteristics of the likely early EV adopters and key vehicle attributes able highly will have higher probability to be a discouraged potential EV
are revealed. adopter.

4.2. Effect of consumer attributes on important vehicle attributes 5. Results

The third goal of this paper is to explore how the importance of 5.1. Characteristics of potential EV adopters and important vehicle
different vehicle attributes varies with consumer attributes. In section attributes
4.1, the most important vehicle attributes for potential EV buyers were
identified; however, this level of importance will vary among the buyers. 5.1.1. Adaptive lasso
For example, assume that comfort has been found as a very important The first and second aim of the paper is finding the characteristics of
vehicle attribute. This attribute’s importance is likely to vary with consumers that thought of buying an EV, and the most important vehicle
changes in consumer attributes, e.g. age. Perhaps older potential EV attributes. Firstly, let’s consider the results of the preliminary adaptive
adopters find comfort much more important compared to younger Lasso estimation, namely the consumer and vehicle variables that are
adopters. To investigate this, each of the significant vehicle attributes discovered to be unimportant. These include the year dummy, nation­
that were found using equation (6) is regressed on the set of consumer ality, some regions of habitude, type of accommodation ownership,
attributes. Since the interest lies in what the potential EV adopters found number of children, presence of disability, full-time vs. part-time job,
important, only the data of these individuals is considered in these re­ and some occupation types. In particular, as expected, there is no dif­
gressions. As before, the adaptive Lasso method is used before every ference between 2014 and 2015 in the popularity of EVs versus CVs, as
regression, to use only those consumer attributes that are relevant. the surveys are only a year apart. Furthermore, no UK nationality is
The dependent variables in these regressions are the most important found to be more prone to considering an EV purchase.
vehicle variables found using equation (6). As discussed in section 3, Almost half of the vehicle attributes were found to be insignificant
vehicle attribute variables are constructed as importance ratings, and using the preliminary adaptive Lasso estimation, meaning that potential
thus binomial logit regression cannot be used. Therefore, ordered logit EV adopters do not value these any more or any less compared to CV
regressions are applied instead: consumers. These attributes are: comfort, interior size, the width of
vehicle choice, reliability, technology establishment, maintenance costs,
∑
M ∑
R
we = β 0 + βm ve,m + βr ce,r + εe , e = 1, 2, …, E (7) vehicle taxes, resale value, and insurance costs. This means that extra
m=1 r=1 potential charges (such as taxes and insurance costs) are found to not be
of primary concern to potential buyers, as opposed to more direct pay­
where we is one of the important vehicle attributes, ve,m are the consumer ments, such as purchase cost, or recharging costs.12 It is unexpected that
variables, ce,r are the control variables, βm are coefficients of the con­ the technology establishment is found irrelevant, as it was expected that
sumer variables, βr are coefficients of the control variables, εe is the error non-adopters would feel that the EVs are generally not proven. A
term, E is the number of observations of potential EV adopters, M is the possible explanation is that most respondents considered hybrids rather
number of consumer variables, and R is the number of control variables. than BEVs, which may consequently not feel unproven. The remaining
The ordered logit regressions provide a coefficient βm for each con­ consumer and vehicle variables are analyzed using a binomial logit
sumer attribute, and thus their effect on each of the key vehicle attri­ regression.
butes. From these results, it is therefore possible to conclude what type
of individuals put the greatest importance on each of the key vehicle 5.1.2. Characteristics of potential EV adopters
attributes. Let’s first consider the characteristics of the potential UK EV early
adopters (Table 2), focusing on the found key vehicle attributes in the
4.3. Key adoption barriers for potential EV buyers next section. Both the consumer characteristic and the vehicle attribute
results come from the same binomial logit regression. This means that
The fourth goal of the paper is to identify the key barriers to EV when finding key consumer characteristics, all vehicle attributes (such
adoption in the UK. The interest in this case lies in those individuals who as purchase cost or charging cost) have been fully controlled for. The
thought about an EV purchase, but for some reasons decided to not propensity of being a potential UK EV adopter is found to increase with:
proceed with it. These respondents can be labelled as temporarily education, being a student, living in the south/middle of the UK, mar­
discouraged potential EV adopters, and the reasons why they decided to riage, lower age, and to small extent, income.
not purchase an EV are of main interest for this part of the work.
Therefore, binomial logit is used to regress the dependent variable (a
dummy variable labelling the discouraged potential adopters as 1, and 12
Evidence of possible greater importance of tax incentives for the UK con­
everyone else as 0) on a set of adoption barriers stated by the re­ sumers was recently suggested by Santos and Davies [39] using a survey of
spondents of the survey, and a set of control variables in the form of stakeholders in the UK, albeit for a small study sample.

5
F. Mandys Renewable and Sustainable Energy Reviews 142 (2021) 110874

Table 2 effective in convincing a large group of consumers to adopt EVs. To

Results of the binomial logit regression for consumer profile attributes. examine the robustness of the results, further binomial regressions were
Thought of Buying an EV Odds Ratio Standard Error P-value run with interactions, and combining variables into categories. This
includes numerous regressions with several interaction variables (e.g.
Frequency of Public Transport Use 1.010 (0.035) 0.774
Frequency of Car Use 0.981 (0.051) 0.707 age interacted with regions of habitude, income, etc.), and combining
West Midlands 1.668** (0.402) 0.034 variables (e.g. regions of habitude into only 3 – north, mid, and south).
East of England 1.457† (0.361) 0.128 However, all these variable changes had very little effect on the final
South-East 1.203 (0.236) 0.347 results, and the interaction variables’ coefficients were generally found
South-West 1.716** (0.430) 0.031
Household Size 0.924 (0.066) 0.270
either insignificant, or irrelevant from the adaptive Lasso, and thus were
Male 1.193 (0.186) 0.258 not reported.
Age (years) 0.988** (0.006) 0.036
Married 1.393** (0.231) 0.046 5.1.3. Most important vehicle attributes for potential UK EV adopters
Number of Vehicles 1.041 (0.112) 0.709
Now let’s examine the most important vehicle attributes for the
Degree Level 2.470*** (0.726) 0.002
Higher Education 2.166** (0.710) 0.018 potential UK EV early adopters. As seen from Table 3, all attributes
A-Levels 2.541*** (0.845) 0.005 except for one are significant at the 10% level and most are at the 5%
ONC 1.526 (0.636) 0.310 level. The key attributes found are: purchase cost, performance/power,
GCSE A-C 1.344 (0.413) 0.336 maximum range, positive environmental effect, and the vehicle being
GCSE D-G 2.054* (0.891) 0.097
Other/Foreign Qualification 1.949** (0.620) 0.036
fully electric.
Health (Likert scale) 0.875† (0.074) 0.115 A typical key vehicle attribute for potential EV buyers, found in all
Lower Managerial/Professional 0.630** (0.119) 0.014 previous studies [13,14], is purchase cost. A vehicle purchase represents
Intermediate 0.516** (0.137) 0.013 a major expenditure for any individual, and since the price of EVs is still
Lower Supervisory/Technical 0.551* (0.180) 0.069
higher than CVs, this is even more true for potential EV buyers. A similar
Semi-routine Occupation 0.543** (0.158) 0.036
Routine Occupation 0.571* (0.191) 0.094 magnitude of importance is found for performance and vehicle range.
Full-time Student 1.909* (0.686) 0.072 Especially range was found to be of major importance in most similar
Yearly income (ten-thousands of £) 1.082† (0.056) 0.143 studies [26,28]. EVs in general have lower maximum range compared to
Notes: ***p < 0.01, **p < 0.05, *p < 0.1, †p < 0.15. CVs, and potential buyers are worried that this range will not be enough
to accommodate their needs, despite being enough to cover the vast
majority of each individual’s trips. Therefore, it is advisable for research
Education has by far the largest effect and is the most significant.
and development (R&D) to focus on improving batteries to lower their
Every level of education has a positive effect on individuals’ consider­
cost and make them last longer, as both these attributes are of major
ation of buying an EV, and this effect is increasing in magnitude for each
importance to potential buyers. The importance of performance may be
education level. This strong impact was expected and found in many
explained by the fact that EVs are in general much quieter than CVs, and
previous studies [12,23,40,41]. Since individuals with higher education
the lack of engine noise may be seen by some as lack of performance.
have generally greater knowledge of the EV positives and CV negatives,
Furthermore, expected small and insignificant result can be seen for
they can be expected to be favorable to a potential EV purchase. In terms
safety, as both CV and EV buyers can be expected to have a similar
of occupation, the greatest positive effect is observed with individuals
average desire for vehicle safety.
who are in full time education, as expected. Students are in general
The more unexpected results involve the EV environmental benefits,
people with more up-to-date knowledge of technology and world issues
and recharging factors. Potential EV adopters are found to put signifi­
in general, as well as a modern outlook on the world. It is thus expected
cant importance on their vehicle having a positive impact on the envi­
that students will be generally positive and open towards modern, clean
ronment. However, past research [6] found rather mixed results, with
technology. This is also related to individuals’ age, where the expecta­
potential adopters usually not putting major value on this aspect.
tion would be that younger people are less conservative, more open to
Therefore, these results make the UK more similar to Germany [21],
new, modern technology, and thus more in favor of a potential EV
rather than USA or Canada, as environmental concerns are found to be
purchase, which the regression results confirm. Location-wise, many
significant among the potential buyers. Furthermore, the discovered low
regions of the UK were found as insignificant, but a positive impact was
importance of convenient recharging and recharging costs lends support
found for West-Midlands, East-Anglia and South-West. People living in
the south/middle of the UK could thus be expected to be more open to
adopting an EV, possibly caused by these regions being richer, having Table 3
greater awareness of environmental issues, and greater exposure to EVs. Results of the binomial logit regression for vehicle attributes.
Furthermore, an expected positive impact is found for marriage status, Thought of Buying an EV Odds Ratio Standard Error P-value
as discovered in most previous studies [26].
Positive Effect on the Environment 1.163*** (0.032) 0.000
An unexpected result is found for income, with an unexpectedly
Fully Electric 1.293*** (0.080) 0.000
small coefficient of low significance. This could arguably point to the Style 0.932** (0.030) 0.030
falling price of EVs (due to better technology and cheaper components, Safety 0.951† (0.032) 0.132
such as lithium batteries) and thus the gradual predicted blurring of the Performance/Power 1.128* (0.071) 0.058
Purchase Cost 1.102*** (0.040) 0.008
difference between EV and CV price tag. With individuals expecting the
Recharging Cost 0.905* (0.050) 0.069
price of EVs to fall in the future due to new EVs(such as Škoda Citigo), Lack of Knowledge 0.796*** (0.045) 0.000
high income may no longer be a significantly important condition for Range 1.085** (0.035) 0.012
future EV purchase considerations. Convenience of Recharging 0.941* (0.032) 0.075
The discussed results are consistent with the German findings of Notes: ***p < 0.01, **p < 0.05, *p < 0.1, †p < 0.15.
Hackbarth and Madlener [21], US findings of Hidrue et al. [12], He et al.
[22], and Higgins et al. [23]. This suggests that the characteristics of
to those studies which conclude that fuel savings are not fully
potential EV early adopters are similar across countries, and that the
differences come rather from varying importance of different vehicle
attributes. Therefore, UK policymakers and manufacturers should focus
their efforts on the stated consumer groups, if they want to be the most

6
F. Mandys Renewable and Sustainable Energy Reviews 142 (2021) 110874

appreciated by consumers and are not discounted in the same way as by strong negative effect of frequent car users may point to their worries
the experts.13 In particular, the convenience of recharging serves as a about BEV range or charging convenience. Additionally, those married
proxy variable to the number of charging points available. As such, the are more likely to be interested in buying a BEV, possibly meaning that
results suggest that these recharging aspects are not of significant they are less worried about a major expenditure due to their partner’s
concern to the potential EV buyers in the UK, but could serve as possible income cushion and support.
barriers to the individuals not interested in an EV purchase. Additional
vehicle attributes (such as vehicles taxes or insurance costs) have been 5.3. Key barriers for the potential EV adopters
part of the preliminary adaptive Lasso estimation, but were removed as
they were found to be insignificant. Identifying the most important barriers that hamper a large-scale EV
diffusion in the UK is the last goal of this paper. The key barriers iden­
5.2. Varying importance of the key vehicle attributes tified are high purchase cost and low maximum range (see Table 5).
Several barrier variables that were removed after the adaptive Lasso
Since 5 vehicle attributes were found to be key for the potential analysis closely relate to the unimportant variables found in sections
adopters, the third goal of this paper will now be covered – exploring 5.1.1 and 5.1.3 (e.g. safety, car tax, or insurance cost), showing
how the importance of key vehicle attributes vary with changes in robustness of the found results.
consumer attributes (Table 4 below). To recapitulate, these are: cost, In line with previous results found in this paper, high cost and low
performance, range, environmental benefit, and the vehicle being fully vehicle range are found as two key attributes. Even though the signifi­
electric. Those vehicle attributes that show a low number of significant cance of these barrier variables is on the lower side, when also consid­
independent variables can be concluded to not vary too much with ering the similar results of sections 5.1 and 5.2, it is safe to conclude that
consumer variables, and vice versa. these results are valid. Since price and range are also found as the most
The importance of cost varies the most among potential EV buyers, important vehicle attributes for EV buyers (section 5.1.3), unsatisfactory
showing the highest number of significant consumer variables out of the price or range will understandably form major barriers to overall EV
5 key vehicle attributes. The results show that those living in the east of diffusion [28]. These findings support the strategy of focusing R&D on
UK put lower importance on EV cost, compared to other regions. Simi­ battery improvements, as these would both reduce purchases cost and
larly, younger, more educated males, who own their accommodation increase maximum vehicle range.
tend to find the EV cost more important. This is understandable, as The remaining barrier variables confirm the results of the previous
younger people do not usually have the savings for a significant expense sections of this paper. Notably, lack of knowledge about EV technology
such as an EV purchase. Similarly, people who spent their savings on an is found with a significant negative effect, meaning that respondents
accommodation purchase may be reluctant for another large purchase in considering this to be a barrier are unlikely to be the discouraged po­
foreseeable future. tential buyers. This means that lack of knowledge is a barrier to those
The importance of performance is less varied among the potential that never considered an EV purchase, but is not for those who already
adopters, with significant attributes including public transport use, considered it. Thus, if the government seeks to persuade the discouraged
gender, age and profession. Performance is found to be less important to potential adopters, EV education would be ineffective. However,
potential adopters that frequently use public transport. Since these in­ educating those that never considered an EV purchase represents a
dividuals use cars infrequently, they are unlikely to be car enthusiasts, promising government policy. Similarly, the costs of recharging display
and thus not extensively interested in performance. On the other hand, a negative coefficient, significant at a 5% level, suggesting that charging
younger males give higher importance to performance when choosing a costs and possibly poor charging infrastructure may represent a barrier
vehicle, likely caused by males being on average interested in cars, their to those that never considered an EV purchase. Nevertheless, it is not
speed and engine power, and younger generations being naturally more found to be of a significant concern for the potential EV adopters.
thrill-seeking.
The importance of range for the potential adopters varies mainly 6. Conclusions
over region of habitude and occupation type. Individuals from southern
UK are significantly more likely to rate vehicle range as important. This The aim of this paper was to investigate the interests and motivations
difference likely stems from driving routines and commuting patterns of UK EV early adopters in detail, while providing information for
between northerners and southerners. People from managerial and government and manufacturer policies. More specifically, this consisted
professional jobs likewise rate maximum range as more important. of identifying the key characteristics of UK early adopters, finding their
Additionally, while the effect of income is positive and significant at the important vehicle attributes, exploring how the importance varies
10% level, the magnitude of the effect is very low, even for an increase of among the potential buyers, and pin-pointing the key barriers to EV
£10 000 in yearly income. Thus, any inference on the effect of income on diffusion. To achieve this, a state preferences UK survey from 2014 to
the importance of vehicle range would be dubious. 2015 was used, and the adaptive Lasso technique for variable selection
The importance of the vehicle environmental benefit varies the least was applied before implementing binomial and ordered logit
among the potential buyers, with no variable significant at 5% or higher regressions.
level. The only variables with some significance, even at the generous The research results offer valuable information to manufacturers,
15% level, are region of habitude, some occupations, and marriage advertisers, marketers and the UK government, for increasing the
status, however, these have generally a small effect even if significant. adoption rates of EVs. The key characteristics identified show that the
Therefore, environmental benefit is an important attribute to the po­ probability of being a UK EV early adopter increases with education,
tential EV buyers in general, with this importance not particularly strong youth, being a student, living in the middle/south, marriage and, to a
or weak for any specific consumer group. low extent, income. Policymakers and marketers should take these
The last attribute is the importance of the vehicle being fully electric, findings into account, with younger and educated groups being a
which varies among potential buyers mainly by frequency of public promising target. Consistently with previous studies, key vehicle attri­
transport and car use, region of habitude, and marriage status. The butes found include cost (particularly for younger, more educated
males), performance (particularly for younger males), and range
(particularly for higher managerial jobs and south of UK). Therefore,
13
Nevertheless, a recent study by DeShazo et al. [42] for California found that R&D should focus on improvements such as batteries, to both lower
redesigned EV rebate programs may achieve greater adoption rates, while also purchase price, and increase maximum range. Additionally, positive
reducing program costs. environmental effect of EVs is important to the UK potential adopters,

7
F. Mandys Renewable and Sustainable Energy Reviews 142 (2021) 110874

Table 4
Results of the 5 ordered logit regressions for varying importance of key vehicle attributes.
Dependent Variable (1) (2) (3) (4) (5)

Purchase Cost Performance Maximum Range Environmental Impact Fully Electric

Surveyed in 2015 (vs. 2014) − 0.03 (0.25) − 0.82** (0.38) − 0.25 (0.23) 0.12 (0.22) − 0.42 (0.38)
Freq. of Public Transport Use 0.03 (0.06) − 0.16* (0.09) 0.07 (0.06) 0.02 (0.05) 0.17* (0.09)
Freq. of Car Use – – 0.04 (0.12) – – – – − 0.31*** (0.11)
North-East − 1.61* (0.84) – – 0.76 (0.61) 0.85† (0.54) – –
North-West − 0.36 (0.42) – – 0.48 (0.44) – – 1.13* (0.60)
Yorkshire − 1.07* (0.55) – – 1.01** (0.50) – – – –
East Midlands − 1.04** (0.53) − 0.95 (0.86) – – − 0.30 (0.42) – –
West Midlands – – – – 0.55 (0.44) – – 0.62 (0.63)
East of England − 0.78* (0.47) − 0.92 (0.80) 0.99** (0.44) – – – –
South-East − 0.17 (0.34) 0.65† (0.44) 1.05*** (0.36) 0.50* (0.28) 1.11** (0.46)
South-West – – – – 1.05** (0.44) 0.28 (0.37) 0.55 (0.72)
Wales – – – – 2.17*** (0.59) 0.37 (0.46) – –
Accommodation Owned 0.78** (0.38) – – 0.32 (0.35) – – – –
Accommodation for Mortgage 0.47 (0.35) – – − 0.06 (0.33) – – – –
Household Size 0.04 (0.13) 0.06 (0.17) 0.14 (0.13) − 0.12 (0.12) − 0.21 (0.20)
Male 0.67** (0.28) 1.01** (0.44) 0.32 (0.26) − 0.03 (0.24) − 0.10 (0.39)
Age (years) − 0.02* (0.01) − 0.03* (0.02) 0.01 (0.01) 0.01 (0.01) − 0.01 (0.01)
Married − 0.13 (0.30) − 0.56 (0.47) − 0.04 (0.29) 0.39† (0.26) 1.29*** (0.49)
Degree Level 0.76** (0.34) 0.77† (0.50) 0.14 (0.34) 0.29 (0.36) − 0.10 (0.73)
Higher Education 0.60 (0.45) 0.90 (0.66) 0.51 (0.41) 0.03 (0.43) 1.24† (0.76
A Levels 0.87** (0.44) 0.87 (0.61) 0.64† (0.43) − 0.32 (0.46) 0.23 (0.90)
GCSE – – – – 0.70* (0.42) 0.04 (0.43) 0.96 (0.78)
Other/Foreign Qualification 0.46 (0.47) – – – – 0.28 (0.43) 1.28† (0.78)
Health (Likert scale) 0.07 (0.15) − 0.05 (0.22) 0.02 (0.14) 0.02 (0.12) − 0.23 (0.22)
Higher Managerial/Prof. − 0.20 (0.36) − 0.98* (0.55) 0.97*** (0.35) 0.52* (0.31) 0.52 (0.51)
Lower Managerial/Prof. – – − 1.00* (0.55) 0.21 (0.33) – – – –
Intermediate − 0.14 (0.52) – – 0.81* (0.45) 0.12 (0.42) – –
Small Employers − 0.08 (0.44) – – – – 0.34 (0.37) 0.56 (0.63)
Lower Supervisory/Technical 1.25** (0.53) − 14.6 (746) – – 0.64 (0.49) – –
Routine Occupation – – – – – – 0.72† (0.50) 0.82 (0.81)
Full-time Student − 0.82 (0.58) 0.02 (0.70) − 0.71 (0.58) 0.55 (0.51) 0.80 (0.78)
Yearly income (10 000s of £) − 0.10 (0.10) 0.06 (0.15) 0.17* (0.09) − 0.03 (0.08) 0.11 (0.13)

Notes: Notes: ***p < 0.01, **p < 0.05, *p < 0.1, †p < 0.15.

represent possible barriers to individuals that haven’t considered an EV

Table 5
purchase. Lastly, high cost and low range were also found as the key
Results of the binomial logit regression for the key barriers to EV adoption.
barriers to wide-scale diffusion of EVs in the UK. This adds robustness to
Notes: Notes: ***p < 0.01, **p < 0.05, *p < 0.1, †p < 0.15.
the previous results, and further strengthens the proposed focus of R&D
Discouraged Potential Early Adopter Odds Ratio Standard Error P-value on batteries, to both reduce the cost and increase range at the same time.
Vehicle Range 1.05† (0.04) 0.117 As expected, lack of knowledge about EV technology was not found to be
Lack of Knowledge 0.81*** (0.05) 0.001 a barrier for EV adopters, but rather for people who never considered an
Lack of Choice 0.96 (0.05) 0.408
EV purchase. Therefore, the government should target its educational
Purchase Cost 1.07* (0.04) 0.081
Recharging Costs 0.87** (0.05) 0.015 efforts on these individuals, if it aims to expand the base of the potential
Surveyed in 2015 0.98 (0.15) 0.907 EV buyers.
Freq. of Car Use 0.93 (0.05) 0.168 A potential addition to the paper could be a wider data set with
Southern UK 1.22 (0.29) 0.398 different types of alternative fuel vehicles (AFVs). However, as noted
Middle UK 1.27 (0.32) 0.332
Northern UK 0.76 (0.20) 0.303
before, getting relevant data is quite difficult, due to the low penetration
Accommodation Owned 1.32 (0.26) 0.159 of the UK market. As this paper is based on stated preferences, an
Household Size 0.99 (0.08) 0.866 interesting area for further study would be to use revealed preferences,
Male 1.17 (0.19) 0.341 and thus actual market actions of the UK consumers. Additionally,
Age (years) 0.99 (0.01) 0.294
looking at the EV problematic from the supply rather than demand side,
Married 1.14 (0.21) 0.479
No. of Vehicles 0.88 (0.10) 0.262 could bring interesting insight and further useful information for man­
Higher Education 2.39*** (0.74) 0.005 ufacturers, marketers and the UK government.
A-Levels Equivalent 1.78* (0.61) 0.090
Below A-Levels 1.46 (0.48) 0.255 Author contribution
Other/Foreign Qualification 1.78† (0.63) 0.103
Health (Likert scale) 0.83* (0.08) 0.065
Long-run Illness/Disability 1.04 (0.20) 0.849 Mandys, F: topic, Investigation, Conceptualization, data construc­
Routine/Semi-routine Jobs 0.63* (0.17) 0.081 tion, Methodology, Software, Formal analysis, Visualization, Writing –
Small Employers 1.55† (0.42) 0.105 original draft, review, editing.
Full-time Student 2.23** (0.84) 0.033
Yearly income (ten-thousands of £) 1.05† (0.06) 0.367
Data availability

something which was not frequently found in other countries. Surpris­ The survey dataset related to this paper can be found at http://doi.
ingly, recharging factors and fuel savings were found unimportant to the org/10.5255/UKDA-SN-7571-2, at the UK Data Service [43].
potential buyers, signifying that larger fuel savings do not offset the
higher cost in the consumers’ eyes. Nevertheless, recharging costs may

8
F. Mandys Renewable and Sustainable Energy Reviews 142 (2021) 110874

Declaration of competing interest [18] Franke T, Cocron P, Bühler F, Neumann I, Krems JF. Adapting to the range of an
electric vehicle – the relation of experience to subjectively available mobility
resources. In: Proceedings of the European conference on human centered design
The authors declare that they have no known competing financial for intelligent transport systems, valencia, Spain; 2012.
interests or personal relationships that could have appeared to influence [19] Liao F, Molin E, Timmermans H, van Wee B. Consumer preferences for business
the work reported in this paper. models in electric vehicle adoption. Transport Pol 2019;73:12–24.
[20] Plötz P, Schneider U, Globisch J, Dütschke E. Who will buy electric vehicles?
Identifying early adopters in Germany. Transport Res Pol Pract 2014;67:96–109.
Acknowledgements [21] Hackbarth A, Madlener R. Willingness-to-pay for alternative fuel vehicle
characteristics: a stated choice study for Germany. Transport Res Pol Pract 2016;
85:89–111.
I would like to specially thank Shivani Taneja for her help and [22] He L, Chen W, Conzelmann G. Impact of vehicle usage on consumer choice of
comments during the research of this paper. Furthermore, I would like to hybrid electric vehicles. Transport Res Transport Environ 2012;17(3):208–14.
thank my colleagues from the 16th IAEE European Conference in [23] Higgins C, Mohamed M, Ferguson M. Size matters: how vehicle body type affects
consumer preferences for electric vehicles. Transport Res Pol Pract 2017;100:
Ljubljana, for their helpful comments and discussion. 182–201.
This research did not receive any specific grant from funding [24] Lane B, Dumortier J, Carley S, Siddiki S, Clark-Sutton K, Graham J. All plug-in
agencies in the public, commercial, or not-for-profit sectors. electric vehicles are not the same: predictors of preference for a plug-in hybrid
versus a battery-electric vehicle. Transport Res Transport Environ 2018;65:1–13.
[25] Axsen J, Kurani K, Burke A. Are batteries ready for plug-in hybrid buyers?
References Transport Pol 2010;17(3):173–82.
[26] Li W, Long R, Chen H, Geng J. A review of factors influencing consumer intentions
[1] Motavalli J. The road ahead for gasoline-free cars. Futurist 2012;6–7 to adopt battery electric vehicles. Renew Sustain Energy Rev 2017;78:318–28.
(March–April). [27] Globisch J, Plötz P, Dütschke E, Wietschel M. Consumer preferences for public
[2] BloombergNEF. Electric vehicle outlook 2020. Bloomberg Finance L.P; 2020. charging infrastructure for electric vehicles. Transport Pol 2019;81:54–63.
Available online from: https://about.bnef.com/electric-vehicle-outlook. [28] Franke T, Krems J. What drives range preferences in electric vehicle users?
[3] Gallagher K, Muehlegger E. Giving green to get green? Incentives and consumer Transport Pol 2013;30:56–62.
adoption of hybrid vehicle technology. J Environ Econ Manag 2011;61(1):1–15. [29] Achtnicht M, Bühler G, Hermeling C. The impact of fuel availability on demand for
[4] Jaramillo P, Samaras C, Wakeley H, Meisterling K. Greenhouse gas implications of alternative-fuel vehicles. Transport Res Transport Environ 2012;17(3):262–9.
using coal for transportation: life cycle assessment of coal-to-liquids, plug-in [30] Train K. Discrete choice methods with simulation. New York: Cambridge
hybrids, and hydrogen pathways. Energy Pol 2009;37(7):2689–95. University Press; 2003.
[5] Egbue O, Long S. Barriers to widespread adoption of electric vehicles: an analysis of [31] Train K. Discrete choice methods with simulation. Cambridge: Cambridge
consumer attitudes and perceptions. Energy Pol 2012;48:717–29. University Press; 2009.
[6] Graham-Rowe E, Gardner B, Abraham C, Skippon S, Dittmar H, Hutchins R, [32] Helveston J, Liu Y, Feit E, Fuchs E, Klampfl E, Michalek J. Will subsidies drive
Stannard J. Mainstream consumers driving plug-in battery-electric and plug-in electric vehicle adoption? Measuring consumer preferences in the U.S. and China.
hybrid electric cars: a qualitative analysis of responses and evaluations. Transport Transport Res Pol Pract 2015;73:96–112.
Res Pol Pract 2012:140–53. [33] Ben-Akiva M, Mcfadden D, Train K, Walker J, Bhat C, Bierlaire M, Bolduc D,
[7] Coffman M, Bernstein P, Wee S. Electric vehicles revisited: a review of factors that Boersch-Supan A, Brownstone D, Bunch D, Daly A, De Palma A, Gopinath D,
affect adoption. Transport Rev 2017;37(1):79–93. Karlstrom A, Munizaga M. Hybrid choice models: progress and challenges. Market
[8] Liao F, Molin E, van Wee B. Consumer preferences for electric vehicles: a literature Lett 2002;13(3):163–75.
review. Transport Rev 2017;37(3):252–75. [34] Glerum A, Stankovikj L, Thémans M, Bierlaire M. Forecasting the demand for
[9] Hardman S, Jenn A, Tal G, Axsen J, Beard G, Daina N, Figenbaum E, Jakobsson N, electric vehicles: accounting for attitudes and perceptions. Transport Sci 2014;48
Jochem P, Kinnear N, Plötz P, Pontes J, Refa N, Sprei F, Turrentine T, Witkamp B. (4):483–99.
A review of consumer preferences of and interactions with electric vehicle charging [35] Williams R. Ordinal independent variables. University of Notre Dame; 2017.
infrastructure. Transport Res Transport Environ 2018;62:508–23. [36] Agresti A. An introduction to categorical data analysis. Hoboken, NJ: Wiley-
[10] Beggs S, Cardell S, Hausman J. Assessing the potential demand for electric cars. Interscience; 2007.
J Econom 1981;17(1):1–19. [37] Pasta DJ. Learning when to Be discrete: continuous vs. Categorical predictors. San
[11] Ewing G, Sarigöllü E. Assessing consumer preferences for clean-fuel vehicles: a Francisco: ICON Clinical Research; 2009. p. 248.
discrete choice experiment. J Publ Pol Market 2000;19(1):106–18. [38] Zou H. The adaptive Lasso and its oracle properties. J Am Stat Assoc 2006;101
[12] Hidrue M, Parsons G, Kempton W, Gardner M. Willingness to pay for electric (476):1418–29.
vehicles and their attributes. Resour Energy Econ 2011;33(3):686–705. [39] Santos G, Davies H. Incentives for quick penetration of electric vehicles in five
[13] Larson P, Viáfara J, Parsons R, Elias A. Consumer attitudes about electric cars: European countries: perceptions from experts and stakeholders. Transport Res Pol
pricing analysis and policy implications. Transport Res Pol Pract 2014;69:299–314. Pract 2020;137:326–42.
[14] Huang Y, Qian L. Consumer preferences for electric vehicles in lower tier cities of [40] Hackbarth A, Madlener R. Consumer preferences for alternative fuel vehicles: a
China: evidences from south Jiangsu region. Transport Res Transport Environ discrete choice analysis. Transport Res Transport Environ 2013;25:5–17.
2018;63:482–97. [41] Li X, Clark C, Jensen K, Yen S, English B. Consumer purchase intentions for
[15] Krupa J, Rizzo D, Eppstein M, Brad Lanute D, Gaalema D, Lakkaraju K, flexible-fuel and hybrid-electric vehicles. Transport Res Transport Environ 2013;
Warrender C. Analysis of a consumer survey on plug-in hybrid electric vehicles. 18:9–15.
Transport Res Pol Pract 2014;64:14–31. [42] DeShazo J, Sheldon T, Carson R. Designing policy incentives for cleaner
[16] Eppstein M, Grover D, Marshall J, Rizzo D. An agent-based model to study market technologies: lessons from California’s plug-in electric vehicle rebate program.
penetration of plug-in hybrid electric vehicles. Energy Pol 2011;39(6):3789–802. J Environ Econ Manag 2017;84:18–43.[43] Office for National Statistics Social
[17] Li L, Wang Z, Chen L, Wang Z. Consumer preferences for battery electric vehicles: a survey division. Opinions and Lifestyle survey, electric vehicles module, february
choice experimental survey in China. Transport Res Transport Environ 2020;78: 2014 and february 2015. [data collection]. second ed. UK Data Service. SN: 7571;
102185. 2015. https://doi.org/10.5255/UKDA-SN-7571-2.