Content uploaded by Filip Mandys
Author content
All content in this area was uploaded by Filip Mandys on Apr 28, 2021
Content may be subject to copyright.
Electric Vehicles and Consumer Choices
Mandys, F.a*
a School of Economics, University of Surrey, United Kingdom
Abstract
Electric vehicles (EVs) have many attractive features compared to conventional vehicles (CVs). Their main
advantage lies in their significant economic benefits due to the much higher fuel efficiency, and substantial
environmental benefit of lower greenhouse gas emissions. However, the market share of EVs in the UK
remains low and their benefits will not be realized unless the government and the manufacturers can gain
crucial information, necessary to effectively support and speed up the adoption of EVs. Such contemporary
information for 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.
Highlights
Alternative fuel vehicles have significant benefits; their UK share remains low
Government and manufacturers need information on consumers, behavior and barriers
Potential UK early adopter: young, highly educated, from south, with higher income
UK early adopters primarily focus on purchase cost, performance and maximum range
Key barriers are high purchase cost and low range; battery research advisable
Keywords: alternative fuel vehicle, consumer behavior, discrete choice model, lasso, socio-technical
barrier, early adopter.
Word count: 7946 words
Abbreviations
AFV alternative fuel vehicle HEV hybrid electric vehicle
AIC Akaike information criterion IIA independence from irrelevant alternatives
BEV battery electric vehicle Lasso least absolute shrinkage and selection operator
BIC Bayesian information criterion OLS ordinary least squares
CV conventional vehicle ONC Ordinary National Certificate
EV electric vehicle PHEV plug-in hybrid electric vehicle
GCSE General Certificate of Secondary Education R&D research and development
* Corresponding author: f.mandys@surrey.ac.uk
2
1. Introduction
The very first attempts at producing an electric vehicle (EV) came as early as the 1830s,
with the first practical car built in London by Thomas Parker in 1884. However, EVs were very
soon driven out of the market by vehicles running on fossil fuels. Recently, commercial EVs are
making a comeback (due to advances in battery and hybrid technologies, as well as environmental
concerns [1], with their bright future being predicted by the manufacturers, governments and many
researchers. For example, BloombergNEF [2] predicts that the new car sales of EVs will grow from
2.7% in 2020, to 58% by 2040, France wants to end the sale of all fossil-fuel based vehicles by
2040, and UK considers banning the sale of all cars that are unable to travel at least 50 miles on
battery power by 2040, and reach net-zero emissions by 2050.
General alternative fuel vehicles (AFVs) can be separated into several different categories,
but primarily these are hybrid electric vehicles (HEVs) and battery electric vehicles (BEVs).
Hybrids run on a combination of gasoline and electric engines and are therefore cleaner than
conventional vehicles (CVs), while BEVs are fully clean, as they do not possess any form of
combustion engine. Due to lack of data, HEV’s and BEV’s are treated as one category of EVs in
this paper. Attractiveness of EVs for governments and consumers lies in their significant economic
and environmental benefits, caused by their cheaper fuel and lower emissions [3]. Even the
emissions of HEVs are about half of standard CVs [4]. Therefore, a large-scale adoption of EVs
can reduce transport emissions, related local health risks, slow down global warming and promote
the use of renewable energy [5][6], and should thus be promoted by the world governments.
Nevertheless, both manufacturers and governments have only limited amount of critical
information necessary to effectively support a large-scale and successful EV diffusion. This
includes, for example, the knowledge of consumers’ wants, what will make them consider a switch
from CVs, or the most serious barriers that slow the diffusion of EVs. This kind of information is
crucial to manufacturers, not only for optimal design purposes, but also to optimize their marketing
and advertising strategies for the correct consumer groups. Governments also value this
information highly for policy purposes, educational purposes, or, for example, for support
programs targeted at a certain typology of consumers. The described contemporary information for
the UK is currently missing in the literature, and this paper aims to fill this important gap.
Therefore, I analyze vehicle aspects, vehicle usage, and consumer profiles in the UK, using
consumer stated preferences survey data and a discrete choice model. The main goals of the study
include: firstly finding the most important characteristics of potential EV consumers. Secondly,
finding which vehicle characteristics chiefly affect the consumers’ desire to buy EVs. Thirdly,
analyzing how these aspects change with different consumer characteristics. And fourthly,
identifying the key barriers to EV adoption for potential EV consumers.
The rest of this paper is organized as follows. Section 2 reviews the main strands of relevant
literature. Section 3 presents the dataset used. Section 4 describes the econometric methods. Section
5 presents the results and discusses the findings. Section 6 concludes the paper.
3
2. Literature Review
Evaluation of consumer choices of electric vehicles represents a dynamic and contemporary
area of literature. While the topic is relatively new, and there are still many areas for potential
research, there exists a rich literature, some of which was well evaluated by e.g. Coffman et al. [7],
Liao et al. [8], or Hardman et al. [9], in their literature reviews. One of the early works exploring
the potential demand for electric cars is the paper by Beggs et al. [10]. They attempted to asses
which vehicle aspects (such as price, or maximum range) are the most important for the demand
for EVs. Their main finding was that consumers put a very high negative valuation on limited
vehicle range, low performance and long refueling period, and surprisingly low importance on
operating costs. Ewing and Sarigöllü [11] revisited this topic, exploring how the preferences
respond to improved performance and government regulation. Similarly, they discover that
consumers are most responsive towards performance attributes, in particular range, acceleration
and refueling time. Positives of government regulations, incentives, and environmental benefits
were found to be less important, as also concluded by e.g. Graham-Rowe et al. [6]. Results of
Hidrue et al. [12], support the conclusions of Ewing and Sarigöllü [11], stressing that higher fuel
costs and operational savings are mostly ignored by consumers, who rather focus on purchase price,
as in Larson et al. [13]. Similarly to Huang and Qian [14] in China, they found that high price,
particularly due to expensive batteries, is one of the main concerns of potential EV consumers. An
alternative result was reached by Krupa et al. [15] using a modified agent-based model of Eppstein
et al. [16], finding that 86% of consumers consider potential fuel savings as a crucial EV attribute.
Similar conclusions are reached in a recent study by Li et al. [17], finding that any incentives that
reduce operation costs should in turn increase the adoption rates by Chinese consumers.
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] also revealed that consumers that
were exposed to EV’s are less sensitive to previously found critical attributes, such as range [18].
Liao et al. [19] furthermore discovered that the type of EV business model affects the attractiveness
of EV adoption, with vehicle leasing being the most attractive.
One of the primary groups of consumers that policymakers and manufacturers are interested
in are the early adopters – the group buying EVs in the next few years. This first large group of
consumers is usually considered highly influential by researchers in deciding the success of a new
technology, according to Plötz et al. [20]. Their results show that the most likely early adopters in
Germany are men of middle age, in technical professions, and living in rural or suburban
households. These people travel many kilometers annually, benefit from the lower EV operational
costs, and have the funds to purchase an expensive vehicle. Hackbarth and Madlener [21] similarly
focus on German consumers, but rather find that potential EV buyers are younger, highly educated
and environmentally conscious individuals, as in Hidrue et al. [12], He et al. [22], or more recently
in Higgins et al. [23], using their multivariate analysis of variance. Similarly, Lane et al. [24] find
that EVs appeal significantly more to individuals who think about them in terms of their technical
and environmental components, compared to the more traditional households.
4
The socio-technical
1
barriers to an extensive EV diffusion were 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 the public charging infrastructure
could attract more “mainstream” buyers, not just early adopters. Further point of Egbue and Long
[5] is that consumers generally tend to resist new, unproven technologies. Therefore, incentives
such as subsidies or fuel taxes will have limited effect and governments should rather focus on
education of consumers of e.g. the non-financial EV benefits (such as environmental). Similarly,
Franke and Krems [28] discover evidence that consumers generally demand a significantly larger
range potential in their vehicles, than they 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 for many
consumers. However, hands-on experience with EVs and government programs, such as gasoline
taxes, tend to have a significantly positive effect on EV uptake [3][16].
Typically, the most widely used methodology of the discussed studies is the application of
a discrete choice model, where the decision of buying/switching to an EV is described as a choice
between two or more vehicle alternatives of various characteristics [8]. In most case, stated
preference data is used, mainly due to limited presence of EVs in many markets, causing lack of
market data. This means that the survey stated preferences and consumer actions may be only
hypothetical, and the questionnaires may provide large number of explanatory variables. The
typical identification problems of overfitting and multicollinearity are then solved by either using
a less detailed, more specific survey, or using a method, such as principal component analysis [22],
in order to extract only the important variables. Modeling-wise, the current standard in the literature
is to use some form of either a logit or a probit model [23]. This often includes a multinomial logit
[29] which assumes independence from irrelevant alternatives
2
(IIA), or a nested logit which
relaxes this assumption and allows for the correlation between various alternatives [14][30][31].
Accounting for taste parameters using a mixed logit [17][32], or a latent factor identifying hybrid
choice model of Ben-Akiva [33], is also popular [19][34].
3. Data
This research uses a stated preferences data set created by combining and quantifying the
answers to a UK survey run in 2014 and again in 2015 (with different individuals compared to
2014). Both of the survey years come from the UK Data Service catalogue, specifically the Electric
Vehicles Module of the Opinions and Lifestyle Survey. Combining and then quantifying the survey
answers yields a pooled cross-section dataset with two time periods of 2014 and 2015. The
Opinions and Lifestyle Survey is a multipurpose social survey, carried out by the Office for
National Statistics, while the Electric Vehicle Module is a more topic-specific set of questions from
the Department for Transport, that are included into the core questions which cover mainly
1
The term socio-technical includes social, political, economic, cultural and technological barriers.
2
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 alternative x must be the one chosen from the subset A. This implies that
consumers are homogeneous.
5
demographic topics. The module contains information about the respondents’ vehicle buying
priorities and their considerations of possibly purchasing an EV vs. a CV. The module surveyed
and interviewed UK adults aged 16 years or older and living in private households, using a multi-
stage stratified random sampling procedure
3
. The results from 2014 and 2015 contain 962 and 1034
respondents respectively, yielding a total of 1996 observations; however, due to missing answers,
this decreases to 1347 observations for the final dataset. The survey contains 76 questions about
consumer attributes (e.g. age or income), and vehicle attributes (e.g. ratings of performance or
maximum range). In the final dataset, 28 unique questions are used for variable creation.
3.1. Variable Construction
The main dependent variable is constructed from the survey question: “Which statement
best describes your attitude towards buying an electric car”. The answers are separated into two
groups. Firstly, individuals who have thought or are thinking of buying an EV and thus can be
considered more likely to buy an EV in the future, and secondly, those who either do not want to
buy an EV at the moment or who have not thought of buying it, and thus are unlikely to purchase
an EV in the near future. The main dependent variable is a binary variable, where the individuals
who thought about buying an EV are labelled as 1, and those who have not are labelled as 0. The
secondary dependent variable, used to identify the key barriers to greater EV diffusion, is
constructed in a similar way. Specifically it identifies the discouraged potential early adopters, by
labeling individuals who thought of an EV purchase but for some reason decided not to at that point
in time as 1, and the remaining individuals as 0.
Most variables for the dataset are created as dummy variables from the survey answers,
however, three variable types were constructed in a more specific manner. Firstly, the survey
answers about income came in the form of income bands, and to construct a single continuous
variable, the middle point of the band was assigned to each individual. Since the number of income
bands is very large, and the size of the intervals is very low, this method of variable creation stays
as true to the data as possible. Secondly, answers about the health of respondents came in the form
of a Likert scale
4
, and since the answer categories were equally spaced, this ordinal variable can be
treated as continuous [35], as in Agresti [36]. Nevertheless, even in the case where the spacing is
not equal across categories, many researchers make a strong case on treating ordinal variables as
continuous [35][37], since “the results are remarkably insensitive to spacing…except in the most
extreme cases” [37]. Thus, even variables without equal spacing are generally constructed as
continuous. Thirdly, the survey answers provide each respondent’s ranking of several vehicle
attributes (e.g. performance or range) and key barriers, according to their perceived importance.
Each of these vehicle attributes is thus constructed as a numerical variable where, for example, if
there are 7 vehicle attributes in a particular question, the most important attribute chosen would be
3
A sampling procedure where members of the population are divided into homogeneous subgroups before sampling,
where each subgroup is mutually exclusive. Samples are then taken in stages using increasingly smaller sampling units.
4
A type of symmetric scale typically used in questionnaires. It most frequently provides answers to a question in the
format “strongly disagree, disagree, neutral, agree, strongly agree”.
6
labelled with 7, the second one would be labelled with 6, and so on for all of the attributes. It is
assumed that if a person did not choose an attribute, that attribute is unimportant to them, and thus
is labelled with a 0. After performing the described operations, the constructed dataset contains 65
variables, where the descriptive statistics can be seen in table 1 below.
Table 1 Descriptive statistics of chosen important variables from the constructed dataset.
Variable Description
Mean
Standard Dev.
Minimum
Maximum
Respondent thought about EV purchase
Discouraged potential EV adopter
0,25
0,21
0,43
0,40
0
0
1
1
Surveyed in 2015
0,52
0,50
0
1
Respondent Demographics
Income
20 460
15 169
0
54 600
Household size
2,44
1,26
1
9
Male
0,51
0,50
0
1
Age
49,94
17,39
16
93
Married
0,54
0,50
0
1
No. of vehicles in the household
1,38
0,78
0
3
Health
4,10
0,88
1
5
Region of Habitude
North of England
0,26
0,44
0
1
Middle of England
0,23
0,42
0
1
South of England
0,34
0,47
0
1
Scotland
0,07
0,25
0
1
Education Level Attained
Higher education
0,39
0,49
0
1
A-Levels equivalent
0,15
0,36
0
1
Below A-Levels
0,22
0,41
0
1
Other qualification
0,11
0,31
0
1
Current Job
Full-time
0,56
0,50
0
1
Managerial and professional
0,35
0,48
0
1
Intermediate
0,11
0,31
0
1
Small employers
0,08
0,28
0
1
Lower supervisory and Technical
0,06
0,24
0
1
Routine and semi-routine
0,16
0,37
0
1
Student
0,05
0,21
0
1
Ratings of Vehicle Attributes and Barriers
Comfort
3,20
3,33
0
7
Positive environmental impact
2,19
2,68
0
7
Vehicle being fully electric
0,23
1,04
0
7
Style
1,68
2,42
0
7
Interior space
1,71
2,34
0
7
Reliability
3,48
2,42
0
7
Safety
2,28
2,25
0
7
Performance
0,29
1,06
0
7
Price
1,28
2,09
0
5
Recharging cost
0,54
1,42
0
5
Maintenance cost
0,57
1,45
0
5
Lack of choice
0,45
1,40
0
5
Lack of knowledge about technology
0,67
1,65
0
5
Maximum range
1,61
2,17
0
5
Recharging convenience
1,70
2,14
0
5
Technology proven
0,30
1,10
0
5
7
4. Methodology
4.1. Characteristics of Early EV Adopters and Importance of Vehicle Attributes
The first and second goal of this research is to find the most important characteristics of the
potential UK EV consumers, and to find out which vehicle attributes are the most important to
them. As the main dependent variable is a binary variable, I apply a binomial logit regression of
this dependent variable, on all the consumer and vehicle variables that were constructed in section
3. The logit is a well-established method, and provides more intuitive inference on the coefficients
than, for example, a probit regression. However, there is an issue that the number of attributes is
very large, which causes problems of overfitting, multicollinearity, and can lead to parameter
estimation issues and subsequent problems in interpretation [22]. To improve the estimation, an
important step is to build a model that only includes important attributes. Researchers, such as e.g.
He et al. [22], used principal component analysis to alleviate these issues. However, the final
extracted principal components are not the same as the original variables, and thus interpreting the
results of the analysis of the principal components and drawing conclusions from them can often
be problematic. Therefore, this paper applies the least absolute shrinkage and selection operator
(Lasso) to avoid these issues.
4.1.1. The Lasso
The Lasso is an econometric regularization
5
method, that is able to combine parameter
estimation and variable selection into a single step. The method shrinks the coefficients of those
variables that are irrelevant for the research to exactly zero, thus performing a selection of
attributes. Those variables that are not shrunk to zero by Lasso can be considered important for the
research and can be used further in the subsequent binomial logit regression. Considering the linear
regression model for this study:
where is the dependent variable, is the vector of independent variables,
are the unknown coefficients of the independent variables, is the error term,
is the no. of observations, and is the number of consumer and vehicle variables.
then the Lasso estimator
is defined as:
where is a tuning parameter, and are the estimated coefficients.
The most important parameter of equation (2) is the tuning parameter , which determines the
level of penalization that is applied, i.e. what kind of relevance a variable has to have in order for
its coefficient to not be shrunk to zero. If the tuning parameter is set too high, most coefficients
5
The process of regularization introduces additional information, in order to prevent the problem of overfitting.
8
will be eliminated and thus the analysis of the effects on the dependent variable can be limited. If
the tuning parameter is set too low, almost no coefficients will be eliminated, hindering what was
desired in the first place – elimination of coefficients. The tuning parameter can be determined
using three main methods: the Akaike information criterion
6
(AIC), the Bayesian information
criterion
7
(BIC), or cross-validation. Cross-validation can be seen as a middle ground between AIC
and BIC, and thus I use 10 rounds of cross-validation to determine the tuning parameter. However,
although that the Lasso has many desirable properties, it suffers from a handful of issues, such as
lacking the oracle properties
8
[38]. The method is also sensitive to the selection of the tuning
parameter . Therefore, I use a refined version of the Lasso instead – the adaptive Lasso.
4.1.2. The Adaptive Lasso
The adaptive Lasso method does have the oracle properties for a suitable choice of the
tuning parameter, and thus performs correct variable selection. The method is also much less
sensitive to the choice of the tuning parameter and estimates the non-zero coefficients with the
same efficiency as the least-squares estimator. The procedure of getting the adaptive Lasso
estimator is very similar to the ordinary Lasso:
where
is the adaptive weights vector,
is an initial estimate of the coefficients,
and is a positive constant for adjustment of the adaptive weights vector, set between
and
.
The main difference between the standard Lasso and the adaptive Lasso comes from the adaptive
weights vector , which applies different regularization for each coefficient. This means that the
penalization is adjusted for each coefficient and provides a stronger penalization to those
coefficients that are smaller. The initial estimate of the coefficients for the adaptive weights vector
is acquired using a standard ordinary least squares (OLS) regression. These coefficients are then
applied to the data points of each consumer profile variable. Therefore, while in the standard Lasso,
the relationship and the matrix of independent variables looks the following:
in the adaptive Lasso, the matrix is adjusted by the coefficients from the initial OLS regression:
6
An estimator that estimates the quality of a model, relative to each of the other models, and thus provides means for
model selection. The model with the minimum AIC value, relative to other models, is selected.
7
Closely related to the AIC. The BIC penalizes the number of parameters more than the AIC.
8
Oracle properties state that a procedure has to identify the right subset of true variables, with optimal estimation rate.
9
After applying 10-rounds of cross validation to determine the optimal tuning parameter, and
checking for multicollinearity issues, 26 final variables are found to be important and thus are used
in the subsequent regressions.
4.1.3. Binomial Logistic Regression
In order to find the characteristics of potential EV adopters and the importance of various
vehicle attributes, the relevant variables from the adaptive Lasso are used in a binomial logit
regression:
where is the dependent variable equal to 1 if a respondent is a potential EV buyers and 0
otherwise, are consumer variables, are vehicle variables, are control variables, are
coefficients of consumer variables, are coefficients of vehicle variables, are coefficients of
control variables, is the error term, is the number of observations, is the number of consumer
variables, is the number of vehicle variables, and is the number of control variables.
The binomial logit regression provides coefficients of the consumer variables, allowing for
interpretation of their effect on the EV dependent variable, as well as coefficients of the vehicle
attributes , providing means of inference of each variable’s importance for potential EV
consumers. Positive and significant variables will increase the chances of the individual to have
thought about EV purchase, and thus, the characteristics of the likely early EV adopters and key
vehicle attributes are revealed.
4.2. Effect of Consumer Attributes on Important Vehicle Attributes
The third goal of this paper is to explore how the importance of different vehicle attributes
varies with consumer attributes. In section 4.1., the most important vehicle attributes for potential
EV buyers were identified; however, this level of importance will vary among the buyers. For
example, assume that comfort has been found as a very important vehicle attribute. This attribute’s
importance is likely to vary with changes in consumer attributes, e.g. age. Perhaps older potential
EV adopters find comfort much more important compared to younger adopters. To investigate this,
I regress each of the significant vehicle attributes that were found using equation (6) on the set of
consumer attributes. Since the interest lies in what the potential EV adopters found important, only
the data of these individuals is considered in these regressions. As before, the adaptive Lasso
method is used before every regression, to use only those consumer attributes that are relevant.
10
The dependent variables in these regressions are the most important vehicle variables found
using equation (6). As discussed in section 3, vehicle attribute variables are constructed as
importance ratings, and thus binomial logit regression cannot be used. Therefore, ordered logit
regressions are applied instead:
where is one of the important vehicle attributes, are the consumer variables, are the
control variables, are coefficients of the consumer variables, are coefficients of the control
variables, is the error term, is the number of observations of potential EV adopters, is the
number of consumer variables, and is the number of control variables.
The ordered logit regressions provide a coefficient for each consumer attribute, and thus their
effect on each of the key vehicle attributes. From these results, it is therefore possible to conclude
what type of individuals put the greatest importance on each of the key vehicle attributes.
4.3. Key Adoption Barriers for Potential EV Buyers
The fourth goal of the paper is to identify the key barriers to EV adoption in the UK. The
interest in this case lies in those individuals who thought about an EV purchase, but for some
reasons decided to not proceed with it. These respondents can be labelled as temporarily
discouraged potential EV adopters, and the reasons why they decided to not purchase an EV are of
main interest for this part of the work. Therefore, I use binomial logit to regress the dependent
variable (a dummy variable labelling the discouraged potential adopters as 1, and everyone else as
0) on a set of adoption barriers stated by the respondents of the survey, and a set of control variables
in the form of consumer attributes. As in the previous parts, the adaptive Lasso is used before each
regression:
where is the discouraged adopter dependent variable, are the barrier variables, are
the consumer attribute control variables, are coefficients of the barrier variables, are
coefficients of the control variables, is the error term, is the number of observations, is the
number of barrier variables, and is the number of control variables.
The regression above provides coefficients of the barrier variables, and consequently their effect
on the discouraged potential EV adopters. Thus, if a barrier variable has a significant and positive
effect, an individual that ranked this variable highly will have higher probability to be a discouraged
potential EV adopter.
11
5. Results
5.1. Characteristics of Potential EV Adopters and Important Vehicle Attributes
5.1.1. Adaptive Lasso
The first and second aim of the paper is finding the characteristics of consumers that thought
of buying an EV, and the most important vehicle attributes. Firstly, let’s consider the results of the
adaptive Lasso, namely the consumer and vehicle variables that are discovered to be unimportant.
These include the year dummy, nationality, some regions of habitude, type of accommodation
ownership, number of children, presence of disability, full-time vs. part-time job, and some
occupation types. In particular, as expected, there is no difference between 2014 and 2015 in the
popularity of EVs versus CVs, as the surveys are only a year apart. Furthermore, no UK nationality
is found to be more prone to considering an EV purchase than any other one.
Almost half of the vehicle attributes were found to be insignificant, meaning that EV
adopters do not value these any more or any less compared to CV consumers. These attributes are:
comfort, interior size, the width of vehicle choice, reliability, technology establishment,
maintenance costs, vehicle tax, resale value, and insurance costs. It is unexpected that the
technology establishment is found irrelevant, as it was expected that non-adopters would feel that
the EVs are generally not proven. A possible explanation is that most respondents considered the
hybrid EV technology rather than fully electric, which is not new and may consequently not feel
unproven. The remaining consumer and vehicle variables are analyzed using a binomial logit
regression.
5.1.2. Characteristics of Potential EV Adopters
I first consider the discovered characteristics of the potential UK EV early adopters (table
2), focusing on the found key vehicle attributes in the next section. The propensity of being a
potential UK EV adopter is found to increase with: education, being a student, living in the
south/middle of the UK, marriage, lower age, and to small extent, income.
Education has by far the largest effect and is the most significant. Every level of education
has a positive effect on individuals’ consideration of buying an EV, and this effect is increasing in
magnitude for each education level. This strong impact was expected and found in many previous
studies. Since individuals with higher education have generally greater knowledge of the EV
positives and CV negatives, they can be expected to be favorable to a potential EV purchase. In
terms of occupation, the greatest positive effect is observed with individuals who are in full time
education, as expected. Students are in general people with more up-to-date knowledge of
technology and world issues in general, as well as a modern outlook on the world. It is thus expected
that students will be generally positive and open towards modern, clean technology. This is also
related to individuals’ age, where the expectation would be that younger people are less
conservative, more open to new, modern technology, and thus more in favor of a potential EV
purchase, which the regression results confirm. Location-wise, many regions of the UK were found
as insignificant, but a positive impact was found for West-Midlands, East-Anglia and South-West.
12
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 greater awareness of environmental
issues, and greater exposure to EVs. Furthermore, an expected positive impact is found for marriage
status, as discovered in most previous studies [26].
Table 2 Results of the binomial logit regression for consumer profile attributes. Notes: *** p<0.01, ** p<0.05, *
p<0.1, † p<0.15.
Thought of Buying an EV
Odds Ratio
Standard Error
P-value
Frequency of Public Transport Use
1.010
(0.035)
0.774
Frequency of Car Use
0.981
(0.051)
0.707
West Midlands
1.668
(0.402)**
0.034
East of England
1.457
(0.361)†
0.128
South-East
1.203
(0.236)
0.347
South-West
1.716
(0.430)**
0.031
Household Size
0.924
(0.066)
0.270
Male
1.193
(0.186)
0.258
Age (years)
0.988
(0.006)**
0.036
Married
1.393
(0.231)**
0.046
Number of Vehicles
1.041
(0.112)
0.709
Degree Level
2.470
(0.726)***
0.002
Higher Education
2.166
(0.710)**
0.018
A Levels
2.541
(0.845)***
0.005
ONC
1.526
(0.636)
0.310
GCSE A-C
1.344
(0.413)
0.336
GCSE D-G
2.054
(0.891)*
0.097
Other/Foreign Qualification
1.949
(0.620)**
0.036
Health (Likert scale)
0.875
(0.074)†
0.115
Lower Managerial/Professional
0.630
(0.119)**
0.014
Intermediate
0.516
(0.137)**
0.013
Lower Supervisory and Technical
0.551
(0.180)*
0.069
Semi-routine Occupation
0.543
(0.158)**
0.036
Routine Occupation
0.571
(0.191)*
0.094
Full-time Student
1.909
(0.686)*
0.072
Yearly income (ten-thousands of £)
1.082
(0.056)†
0.143
An unexpected result is found for income, which was expected to have a significant positive
impact, but this impact is unexpectedly small, with low significance. This could arguably point to
the falling price of EVs (due to better technology and cheaper components, such as lithium
batteries) and thus the gradual predicted blurring of the difference between EV and CV price tag.
With individuals expecting the price of EVs to fall in the future due to new EV cars (such as e.g.
Škoda Citigo), high income may no longer be a significantly important condition for future
potential EV purchase considerations.
The discussed results are consistent with the German findings of Hackbarth and Madlener
[21], and in line with the results of e.g. Hidrue et al. [12], He et al. [22] for the USA, and Higgins
et al. [23]. This supports the idea that the characteristics of 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 effective in convincing a large group of consumers
13
to adopt EVs. Alternatively, if they would rather want to increase the EV adoption likelihood of
other groups, then they should use the found characteristics as an indication of which typology of
consumers to avoid. To examine the robustness of the results, further binomial regressions were
run with interactions, and combining variables into categories. However, these variable changes
had very little effect on the results, and the interaction variables were generally found irrelevant
using adaptive Lasso.
5.1.3. Most Important Vehicle Attributes for Potential UK EV Adopters
Now let’s examine the most important vehicle attributes for the potential UK EV early
adopters. As seen from table 3, all attributes except for one are significant at the 10% level and
most are at the 5% level. The key attributes found are: purchase cost, performance/power,
maximum range, positive environmental effect, and the vehicle being fully electric.
Table 3 Results of the binomial logit regression for vehicle attributes. Notes: *** p<0.01, ** p<0.05, * p<0.1, †
p<0.15.
Thought of Buying an EV
Odds Ratio
Standard Error
P-value
Positive Effect on the Environment
1.163
(0.032)***
0.000
Fully Electric
1.293
(0.080)***
0.000
Style
0.932
(0.030)**
0.030
Safety
0.951
(0.032)†
0.132
Performance/Power
1.128
(0.071)*
0.058
Purchase Cost
1.102
(0.040)***
0.008
Recharging Cost
0.905
(0.050)*
0.069
Lack of Knowledge
0.796
(0.045)***
0.000
Range
1.085
(0.035)**
0.012
Convenience of Recharging
0.941
(0.032)*
0.075
A typical key vehicle attribute for potential EV buyers, found in all previous studies
[13][14], is purchase cost. A vehicle purchase represents a major expenditure for any individual,
and since the price of EVs is still higher than CVs, this is even more true for potential EV buyers.
A similar magnitude of importance is found for performance and vehicle range. Especially range
was found to be of major importance in most similar studies [26][28]. EVs in general have lower
maximum range compared to 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 and development (R&D) to focus on
improving batteries to lower their cost and make them last longer, as both these attributes are of
major importance to potential buyers. The importance of performance may be explained by the fact
that EVs are in general much quieter than CVs, and the lack of engine noise may be seen by some
as lack of performance. Furthermore, expected small and insignificant result can be seen for safety,
as both CV and EV buyers can be expected to have a similar average desire for vehicle safety.
The more unexpected results involve the EV environmental benefits, and recharging
factors. Potential EV adopters are found to put significant importance on their vehicle having a
positive impact on the environment. However, past research [6] found rather mixed results, with
potential adopters usually not putting major value on this aspect. Therefore, these results make the
14
UK more similar to Germany [21], rather than USA or Canada, as environmental concerns are
found to be significant among the potential buyers. Furthermore, the discovered low importance of
convenient recharging and recharging costs lends support to those studies which conclude that fuel
savings are not fully appreciated by consumers and are not discounted in the same way as by the
experts.
5.2. Varying Importance of the Key Vehicle Attributes
Since 5 vehicle attributes were found to be key for the potential UK EV adopters in the
previous section, I now move to the third goal of this paper – to explore how the importance of key
vehicle attributes vary with changes in consumer attributes (table 4 below). To recapitulate, the key
vehicle attributes are: cost, performance, range, environmental benefit, and the vehicle being fully
electric. Those vehicle attributes that show a low number of significant independent variables can
be concluded to not vary too much with consumer variables, and vice versa.
The importance of cost varies the most among potential EV buyers, showing the highest
number of significant consumer variables out of the 5 key vehicle attributes. The results show that
those living in the east of UK put lower importance on EV cost, compared to other regions.
Similarly, younger, more educated males, who own their accommodation tend to find, on average,
the EV cost more important. The greater importance of EV cost for younger people is
understandable, as younger people do not usually have the savings for a significant expense such
as an EV purchase. Similarly, people who spent their savings on an accommodation purchase may
be reluctant for another large purchase in foreseeable future, and thus care more about EV cost.
The importance of performance is less varied among the potential adopters, with significant
attributes including frequency of public transport use, gender, age and profession. Performance is
found to be less important to potential adopters that frequently use public transport. Since these
individuals use cars infrequently, they are unlikely to be car enthusiasts, and thus not extensively
interested in performance. On the other hand, younger males give higher importance to
performance when choosing a vehicle, likely caused by males being on average interested in cars,
their speed and engine power, and younger generations being naturally more thrill-seeking.
The importance of range for the potential adopters varies mainly over two consumer
attributes – region of habitude and occupation type. Individuals from southern UK are significantly
more likely to rate vehicle range as important. This difference likely stems from driving routines
and commuting patterns between northerners and southerners. People from managerial and
professional jobs likewise rate maximum range as more important. Additionally, while the effect
of income is positive and significant at the 10% level, the magnitude of the effect is very low, even
for an increase of £10 000 in yearly income. Thus, any inference on the effect of income on the
importance of vehicle range would be dubious.
15
Table 4 Results of the 5 ordered logit regressions for varying importance of key vehicle attributes. Notes: Notes: ***
p<0.01, ** p<0.05, * p<0.1, † p<0.15.
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)
The importance of the vehicle environmental benefit varies the least among the potential
buyers, with no variable significant at 5% or higher level. The only variables with some
significance, even at the generous 15% level, are region of habitude, some occupations, and
marriage status, however, these have generally low significance and a small effect even if
significant. Therefore, environmental benefit is an important attribute to the potential EV buyers
in general, and this importance is not particularly strong or weak for any specific group of
consumers.
The last of the 5 attributes considered is the importance of the vehicle being fully electric
(BEV), which varies among potential buyers mainly by frequency of public transport and car use,
region of habitude, and marriage status. The strong negative effect of frequent car users may point
to their worries about BEV range or charging convenience. Additionally, married potential EV
16
buyers are more likely to be interested in buying a BEV, possibly meaning that they are less worried
about a major expenditure due to their partner’s income cushion and support.
5.3. Key Barriers for the Potential EV Adopters
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 identified are high purchase cost and low maximum
range. Several barrier variables that were removed after the adaptive Lasso analysis closely relate
to the unimportant variables found in sections 5.1.1. and 5.1.3 (e.g. safety, car tax, or insurance
cost), showing robustness of the found results.
Table 5 Results of the binomial logit regression for the key barriers to EV adoption. Notes: Notes: *** p<0.01, **
p<0.05, * p<0.1, † p<0.15.
Discouraged Potential Early Adopter
Odds Ratio
Standard Error
P-value
Vehicle Range
1.05
(0.04)†
0.117
Lack of Knowledge
0.81
(0.05)***
0.001
Lack of Choice
0.96
(0.05)
0.408
Purchase Cost
1.07
(0.04)*
0.081
Recharging Costs
0.87
(0.05)**
0.015
Surveyed in 2015 (vs. 2014)
0.98
(0.15)
0.907
Freq. of Car Use
0.93
(0.05)
0.168
Southern UK
1.22
(0.29)
0.398
Middle UK
1.27
(0.32)
0.332
Northern UK
0.76
(0.20)
0.303
Accommodation Owned
1.32
(0.26)
0.159
Household Size
0.99
(0.08)
0.866
Male
1.17
(0.19)
0.341
Age (years)
0.99
(0.01)
0.294
Married
1.14
(0.21)
0.479
No. of Vehicles
0.88
(0.10)
0.262
Higher Education
2.39
(0.74)***
0.005
A Levels Equivalent
1.78
(0.61)*
0.090
Below A Levels
1.46
(0.48)
0.255
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
Routine/Semi-routine Jobs
0.63
(0.17)*
0.081
Small Employers
1.55
(0.42)†
0.105
Full-time Student
2.23
(0.84)**
0.033
Yearly income (ten-thousands of £)
1.05
(0.06)†
0.367
In line with previous results found in this paper, high cost and low vehicle range are found
as two key attributes. Even though the significance of these barrier variables is on the lower side,
when also considering the similar results of sections 5.1. and 5.2., it is safe to conclude that these
results are valid. Since price and range are also found as the most important vehicle attributes for
EV buyers (section 5.1.3.), unsatisfactory price or range will understandably form major barriers
to overall EV diffusion [28]. These findings support the strategy of focusing R&D on battery
improvements, as these would both reduce purchases cost and increase maximum vehicle range.
The remaining barrier variables confirm the results of the previous sections of this paper,
like the unimportance of recharging factors. Notably, lack of knowledge about EV technology is
17
found with a significant negative effect, meaning that respondents considering this to be a barrier
are unlikely to be the discouraged potential buyers. This means that lack of knowledge is a barrier
to those that never considered an EV purchase, but is not for those who already considered it. Thus,
if the government seeks to persuade the discouraged potential adopters, EV education would be
ineffective. However, educating those that never considered an EV purchase represents a promising
government policy.
6. Conclusions
The aim of this paper was to investigate the interests and motivations of UK EV early
adopters in detail, while providing information for government and manufacturer policies. More
specifically, this consisted of identifying the key characteristics of UK early adopters, finding their
important vehicle attributes, exploring how the importance varies among the potential buyers, and
pin-pointing the key barriers to EV diffusion. To achieve this, I used a state preferences UK survey
from 2014 and 2015, and applied the adaptive Lasso technique for variable selection before
implementing binomial and order logit regressions.
The research results offer valuable information to manufacturers, advertisers, marketers and
the UK government, for increasing the adoption rates of EVs. The key characteristics identified
show that the probability of being a UK EV early adopter increases with education, youth, being a
student, living in the middle/south, marriage and, to a low extent, income. Policymakers and
marketers should take these findings into account, with younger and educated groups being a
promising target. Consistently with previous studies, key vehicle attributes 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, R&D should focus on
improvements such as batteries, to both lower purchase price, and increase maximum range.
Additionally, positive environmental effect of EVs is important to the UK potential adopters,
something which was not frequently found in other countries. Surprisingly, recharging factors and
fuel savings were found unimportant, signifying that larger fuel savings do not offset the higher
cost in the consumers’ eyes. Lastly, high cost and low range were also found as the key barriers to
wide-scale diffusion of EVs in the UK. This adds robustness to the previous results, and further
strengthens the proposed focus of R&D on batteries, to both reduce the cost and increase range at
the same time. As expected, lack of knowledge about EV technology was not found to be a barrier
for EV adopters, but rather for people who never considered an EV purchase. Therefore, the
government should target its educational efforts on these individuals, if it aims to expand the base
of the potential EV buyers.
A potential addition to the paper could be a wider data set with different types of alternative
fuel vehicles (AFVs), as opposed to only CVs vs. EVs. However, as noted before, getting relevant
data on this kind of vehicles is quite difficult, due to the low penetration of the UK market. As this
paper is based on stated preferences, an interesting area for further study would be to use revealed
preferences, and thus actual market actions of the UK EV consumers. Additionally, looking at the
EV problematic from the supply rather than demand side, could bring interesting insight and further
useful information for manufacturers, marketers and the UK government.
18
Acknowledgements
I would like to specially thank Dr. Shivani Taneja for her help and comments during the research of this
paper. Furthermore, I would like to thank my colleagues from the 16th IAEE European Conference in
Ljubljana, for their helpful comments and discussion.
This research did not receive any specific grant from funding agencies in the public, commercial, or not-
for-profit sectors.
References
[1] Motavalli, J. (2012). The road ahead for gasoline-free cars. The Futurist, (March–April), 6–7.
[2] BloombergNEF. (2020). Electric Vehicle Outlook 2020. Bloomberg Finance L.P. [Available online
from: https://about.bnef.com/electric-vehicle-outlook].
[3] Gallagher, K., & Muehlegger, E. (2011). Giving green to get green? Incentives and consumer
adoption of hybrid vehicle technology. Journal of Environmental Economics and
Management, 61(1), 1-15.
[4] Jaramillo, P., Samaras, C., Wakeley, H., & Meisterling, K. (2009). Greenhouse gas implications of
using coal for transportation: Life cycle assessment of coal-to-liquids, plug-in hybrids, and hydrogen
pathways. Energy Policy, 37(7), 2689-2695.
[5] Egbue, O., & Long, S. (2012). Barriers to widespread adoption of electric vehicles: An analysis of
consumer attitudes and perceptions. Energy Policy, 48, 717-729.
[6] Graham-Rowe, E., Gardner, B., Abraham, C., Skippon, S., Dittmar, H., Hutchins, R., & Stannard, J.
(2012). Mainstream consumers driving plug-in battery-electric and plug-in hybrid electric cars: A
qualitative analysis of responses and evaluations. Transportation Research Part A: Policy and
Practice, 140-153.
[7] Coffman, M., Bernstein, P. and Wee, S. (2017). Electric vehicles revisited: a review of factors that
affect adoption. Transport Reviews, 37(1), pp.79-93.
[8] Liao, F., Molin, E. and van Wee, B. (2017). Consumer preferences for electric vehicles: a literature
review. Transport Reviews, 37(3), pp.252-275.
[9] Hardman, S., Jenn, A., Tal, G., Axsen, J., Beard, G., Daina, N., Figenbaum, E., Jakobsson, N.,
Jochem, P., Kinnear, N., Plötz, P., Pontes, J., Refa, N., Sprei, F., Turrentine, T. and Witkamp, B.
(2018). A review of consumer preferences of and interactions with electric vehicle charging
infrastructure. Transportation Research Part D: Transport and Environment, 62, pp.508-523.
[10] Beggs, S., Cardell, S., & Hausman, J. (1981). Assessing the potential demand for electric
cars. Journal of Econometrics, 17(1), 1-19.
[11] Ewing, G., & Sarigöllü, E. (2000). Assessing Consumer Preferences for Clean-Fuel Vehicles: A
Discrete Choice Experiment. Journal of Public Policy & Marketing, 19(1), 106-118.
[12] Hidrue, M., Parsons, G., Kempton, W., & Gardner, M. (2011). Willingness to pay for electric vehicles
and their attributes. Resource and Energy Economics, 33(3), 686-705.
[13] Larson, P., Viáfara, J., Parsons, R., & Elias, A. (2014). Consumer attitudes about electric cars: Pricing
analysis and policy implications. Transportation Research Part A: Policy and Practice, 69, 299-314.
19
[14] Huang, Y. and Qian, L. (2018). Consumer preferences for electric vehicles in lower tier cities of
China: Evidences from south Jiangsu region. Transportation Research Part D: Transport and
Environment, 63, pp.482-497.
[15] Krupa, J., Rizzo, D., Eppstein, M., Brad Lanute, D., Gaalema, D., Lakkaraju, K., & Warrender, C.
(2014). Analysis of a consumer survey on plug-in hybrid electric vehicles. Transportation Research
Part A: Policy and Practice, 64, 14-31.
[16] Eppstein, M., Grover, D., Marshall, J., & Rizzo, D. (2011). An agent-based model to study market
penetration of plug-in hybrid electric vehicles. Energy Policy, 39(6), 3789-3802.
[17] Li, L., Wang, Z., Chen, L. and Wang, Z. (2020). Consumer preferences for battery electric vehicles:
A choice experimental survey in China. Transportation Research Part D: Transport and
Environment, 78, p.102185.
[18] Franke, T., Cocron, P., Bühler, F., Neumann, I., Krems, J., F. (2012). Adapting to the range of an
electric vehicle – the relation of experience to subjectively available mobility resources. Proceedings
of the European Conference on Human Centered Design for Intelligent Transport Systems, Valencia,
Spain.
[19] Liao, F., Molin, E., Timmermans, H. and van Wee, B. (2019). Consumer preferences for business
models in electric vehicle adoption. Transport Policy, 73, pp.12-24.
[20] Plötz, P., Schneider, U., Globisch, J., & Dütschke, E. (2014). Who will buy electric vehicles?
Identifying early adopters in Germany. Transportation Research Part A: Policy and Practice, 67, 96-
109.
[21] Hackbarth, A. and Madlener, R. (2016). Willingness-to-pay for alternative fuel vehicle
characteristics: A stated choice study for Germany. Transportation Research Part A: Policy and
Practice, 85, pp.89-111.
[22] He, L., Chen, W., & Conzelmann, G. (2012). Impact of vehicle usage on consumer choice of hybrid
electric vehicles. Transportation Research Part D: Transport and Environment, 17(3), 208-214.
[23] Higgins, C., Mohamed, M. and Ferguson, M. (2017). Size matters: How vehicle body type affects
consumer preferences for electric vehicles. Transportation Research Part A: Policy and Practice,
100, pp.182-201.
[24] Lane, B., Dumortier, J., Carley, S., Siddiki, S., Clark-Sutton, K. and Graham, J. (2018). All plug-in
electric vehicles are not the same: Predictors of preference for a plug-in hybrid versus a battery-
electric vehicle. Transportation Research Part D: Transport and Environment, 65, pp.1-13.
[25] Axsen, J., Kurani, K., & Burke, A. (2010). Are batteries ready for plug-in hybrid buyers?. Transport
Policy, 17(3), 173-182.
[26] Li, W., Long, R., Chen, H. and Geng, J. (2017). A review of factors influencing consumer intentions
to adopt battery electric vehicles. Renewable and Sustainable Energy Reviews, 78, pp.318-328.
[27] Globisch, J., Plötz, P., Dütschke, E. and Wietschel, M. (2019). Consumer preferences for public
charging infrastructure for electric vehicles. Transport Policy, 81, pp.54-63.
[28] Franke, T., & Krems, J. (2013). What drives range preferences in electric vehicle users?. Transport
Policy, 30, 56-62.
[29] Achtnicht,M.,Bühler,G. and Hermeling,C. (2012). The impact of fuel availability on demand for
alternative-fuel vehicles. Transportation Research Part D: Transport and Environment,17(3),
pp.262-269.
20
[30] Train, K. (2003). Discrete choice methods with simulation. New York: Cambridge University Press.
[31] Train, K. (2009). Discrete Choice Methods with Simulation. Cambridge: Cambridge University Press.
[32] Helveston, J., Liu, Y., Feit, E., Fuchs, E., Klampfl, E. and Michalek, J. (2015). Will subsidies drive
electric vehicle adoption? Measuring consumer preferences in the U.S. and China. Transportation
Research Part A: Policy and Practice, 73, pp.96-112.
[33] Ben-Akiva, M., Mcfadden, D., Train, K., Walker, J., Bhat, C., Bierlaire, M., Bolduc, D., Boersch-
Supan, A., Brownstone, D., Bunch, D., Daly, A., De Palma, A., Gopinath, D., Karlstrom, A. and
Munizaga, M. (2002). Hybrid choice models: Progress and challenges. Marketing Letters, 13(3),
pp.163-175.
[34] Glerum, A., Stankovikj, L., Thémans, M. and Bierlaire, M. (2014). Forecasting the Demand for
Electric Vehicles: Accounting for Attitudes and Perceptions. Transportation Science, 48(4), pp.483-
499.
[35] Williams, R. (2017) Ordinal Independent Variables. University of Notre Dame.
[36] Agresti, A. (2007). An introduction to categorical data analysis. Hoboken, NJ: Wiley-Interscience.
[37] Pasta, D., J. (2009). Learning When to Be Discrete: Continuous vs. Categorical Predictors. ICON
Clinical Research, San Francisco, paper 248.
[38] Zou, H. (2006). The Adaptive Lasso and Its Oracle Properties. Journal of the American Statistical
Association, 101(476), pp.1418-1429.
