Content uploaded by Christian Schlereth
Author content
All content in this area was uploaded by Christian Schlereth on Jun 11, 2015
Content may be subject to copyright.
1 / 27
Fostering the Adoption of Electric Vehicles by Providing Complementary
Mobility Services: A Two-step Approach using Best-Worst Scaling and
Dual Response
Wenyan Zhoua
Christian Schlereth (Corresponding author)b
Oliver Hinza
Abstract: There is a substantial gap in research regarding the adoption of electric vehicles as a strategy to
remedy the climate problem and reduce oil consumption by integrating complementary mobility services. To
address this gap, we employ a two-step approach utilizing a hybrid stated preference method. Study 1 uses
Best-Worst Scaling and identifies the top three complementary mobility services consumers would prefer with an
electric vehicle. Study 2 applies Dual Response and analyzes the importance of these three services relative to
other technological and economic factors of electric vehicles. Our results offer evidence that complementary
mobility services may significantly foster electric vehicle adoption. Moreover, low purchase prices are less
important than low recurring costs, such as electricity costs. Finally, a segmentation strategy may be fruitful
because, e.g., men are more attracted by technological advantages than women and elderly consumers have a
higher preference for services that offer convenience.
Keywords: Adoption; Electric vehicles; Complementary mobility services; Best-Worst Scaling; Dual Response
Acknowledgements:
Acknowledgments The authors gratefully thank Luigi Bianco for his assistance during the data collection in
Study 1 & 2 and Dr. Donovan Pfaff from Bonpago for his financial support in study 2. We also thank Joséphine
Süptitz, the two anonymous referees as well as the editor Günter Fandel for their valuable comments and
excellent suggestions.
a Wenyan Zhou, Oliver Hinz, Technische Universität Darmstadt, Faculty of Business Administration and
Information Systems, Hochschulstraße 1, 64289 Darmstadt, Germany, Phone: ++49-6151-16-75221, Fax:
++49-6151-16-72220, zhouwenyan615@gmail.com, hinz@wi.tu-darmstadt.de.
b Christian Schlereth, WHU – Otto Beisheim School of Management, Chair of Digital Marketing, Burgplatz 2,
56179 Vallendar, Germany, Phone: ++49-261-6509-455, Fax: ++49-261-6509-509,
christian.schlereth@whu.edu.
2 / 27
1 Introduction
Fuel consumption is one of the main drivers of environmental pollution (e.g., CO2 emissions), and a total of 16%
of global anthropogenic CO2 emissions come from road transport vehicles (Olivier et al. 2013). Thus, emissions
from automobiles represent a significant challenge for researchers, representatives from the industrial community,
and policymakers. With 98% of all vehicles in the world still running on gasoline or diesel, pressure on global oil
supply is rapidly increasing, and peak oil demand is predicted to occur no later than 2030 (Aftabuzzaman and
Mazloumi 2011). Therefore, from the perspectives of both science and politics, it is crucial to remedy the climate
change problem and alleviate the problem of the oil demand gap.
One way to relieve both problems could involve a broader adoption of electric vehicles (EVs, including
Plug-in electric cars, hybrid electric cars, and hydrogen vehicles), which uses one or more electric engines or
traction engines for propulsion. Germany has made EVs part of its long-term oil policy and aims to reduce
emissions across all sectors by 40% by 2020 pursuing a massive EV strategy. This goal implies that the country
aims to have one million EVs on its roads by 2020. To achieve this ambitious goal, nearly 1.5 billion Euros ($1.9
billion) have been invested in research on EV subsidization and the development of e-mobility in general (Peard
2013). Those investments obviously stimulated the supply of EVs by the e-mobility industry. However, on the
demand side EVs face rather low levels of adoption. In 2012, only 4,157 out of a total of three million new
vehicles registered in Germany were EVs. In other countries the proportion of newly registered EVs is similarly
low. Although this number is double than that of the previous year, the situation remains unsatisfactory
(International Council on Clean Transportation 2013).
Thus, there seem to be a number of barriers to EV adoption. Previous research has noted the importance of
price (including purchase price and recharging costs), charging time, and driving ranges in the (non-)adoption of
EVs (Beggs and Cardell 1981; Bunch et al. 1993). However, along with the automobile industry’s work in this
regard, e-technologies have gradually improved (Wesseling et al. 2013). New battery technologies offer longer
ranges, more power and a shorter recharging time (Tie and Tan 2013) (e.g., lithium-ion technology by Toyota;
the new water-based battery by General Electric). In addition, government subsidization of this technology in the
form of subsidies and/or tax relief (Gärling and Thøgersen 2001) may help solve the problem from the cost
perspective.
Although complementary mobility services are likely to gain importance, they have mostly been neglected
in research, or prior research has treated these services as just another factor, which might increase costs.
Specifically, prior research has focused on repair and maintenance costs (Ewing and Sarigöllü 1998), service
station costs (Brownstone et al. 2000), and operating costs (Shepherd et al. 2012). Other studies mainly included
the availability of electricity charging stations (e.g., Bownstone et al. 2000; Mau et al. 2008; Potoglou and
Kanaroglou 2007). Instead of merely focusing on costs or mandatory infrastructure elements, this research
proposes that well-tailored complementary mobility services should be exclusively developed to improve the
holistic driving experience of EV. Examples include “Intelligent charging stations” or “IT-based parking space
and payment”. Our study encourages thinking about the availability of both exclusive and non-exclusive EV
services that might significantly affect consumer behavior. Consequently, our paper aims to explore the influence
of complementary mobility services on EV adoption and to formulate recommendations on how industry and
policymakers could use our results to more effectively encourage the adoption of EVs.
To achieve this goal, we conduct two types of discrete choice experiments that have a firm foundation in
sociology and behavioral research and are well-known for their ability to explain actual purchasing behavior,
even if the studied products do not yet exist on the market (Swait and Andrews 2003). Because the number of
possible complementary mobility services is quite large, we surmount the limits in the number of attributes of
3 / 27
the discrete choice experiment by conducting a two-step approach: first, we utilize Best-Worst Scaling to assess
the importance of identified potential complementary mobility services; second, we employ Dual Response, and
explore the influence of the most important complementary mobility services relative to other well-studied
attributes of EVs. This study shows the advantages of this novel two-step approach.
2 Literature reviews
This section summarizes the attributes prior research has examined with respect to EV preferences and
introduces the methodological foundation of our research. Subsequently, we also outline random utility theory,
which serves as the core of our theoretical framework in the two empirical studies.
2.1 Research on Electric Vehicle Preferences
Since 1981, following the first published study on EV demand (Beggs and Cardell 1981), practitioners and
scholars have focused on factors affecting the adoption of EVs by using revealed preference (RP) and stated
preference (SP) methods. RP can be used to forecast future demand based on past real-world decisions, whereas
SP can be used to incorporate new attributes of a product not yet available on the market. In this paper, we rely
on SP. Without claiming completeness, Table 1 lists studies in the adoption of EVs using RP or SP in the last few
years. Moreover, all attributes examined in previous studies can be assigned to one of four factors: technological,
economic, environmental, and complementary mobility services (see Table 1).
Better technologies improve purchase intentions by increasing consumers’ perceived ease-of-use and
perceived usefulness. For EVs, technological superiority is expressed by attributes such as charging time, range
per charge, motor power, acceleration, top speed, and multiple-fuel capability. Early in the 1980s, Beggs et al.
(1981) documented that limited range and long recharging time were the most significant barriers to the adoption
of EVs. Moreover, because the development of technology itself is a dynamic process, the relationship between
the adoption of a new technology (such as EVs) and its popularity in the market has also proven to be dynamic
(social contagion effect) (Axsen et al. 2009). Customers’ choices also can change the development of technology
(Mau et al. 2008). Because technological attributes continue to be frequently used in recent studies on EV
adoption (Dagsvik et al. 2002; Lieven et al. 2011), we also consider them in our study. However, we do not
devote much attention to the social contagion effect, which has been examined in other domains for new product
adoption (Hinz et al. 2014).
Another important factor considered in all studies is the attribute “costs”, which consist of EV purchase
price plus mileage-dependent operating costs. The mileage-dependent operating costs in the context of EVs
imply electricity costs or recharging costs. Previous research has shown that both the purchase price and the
mileage-dependent operating costs may significantly affect consumers’ decisions about adopting EVs, but no
unified conclusion about the optimal pricing strategy has been reached. Other economic factors, such as parking,
commuting, and repair and maintenance costs, have also been discussed in later research, but their influence on
choosing a vehicle is considered rather weak (Ewing and Sarigöllü 1998). As is well-known, price is always
directly connected to the level of performance, and studies have recently begun to estimate the willingness to pay
not only for an EV but also for high-tech features of the vehicle (Hidrue et al. 2011). In our study, we focus on
the two most important cost factors: purchase price and electricity costs.
4 / 27
Table 1: Electric Vehicle Studies using RP and SP after 2000
1
Study
Data
Method
(Econometric model)
Factors contributing to the adoption of EVs
Conclusions
Technolo-
gical
Econo-
mical
Environ-
mental
Complementary mobility services
Bownstone et al.
(2000)
7,387 households in
California
Joint SP/RP data
(multinomial and mixed logit)
╳
╳
╳
Electricity charging stations
Large heterogeneity in preference for fuel types could be due to
respondents’ different information sets and fundamental uncertainty.
Ewing and
Sarigöllü (2000)
881 respondents in
Canada
SP discrete choice experiment
(multinomial logit)
╳
╳
╳
Relative vehicle prices and performance levels as well as differential
commuting costs and times had modest effects on vehicle choice.
Dagsvik et al.
(2002)
622 Norwegian
residents
SP discrete choice experiment
(random utility models for ranking)
╳
╳
Alternative fuel vehicles appear to be fully competitive alternatives.
Driving range is an important attribute.
Potoglou and
Kanaroglou (2007)
902 respondents in
Canada
SP discrete choice experiment
(nested logit model)
╳
╳
╳
╳
Electricity charging stations
Reducing monetary costs, purchase tax relief and low emissions rates
encourage households to adopt a cleaner vehicle.
Mau et al. (2008)
1935 respondents in
Canada
SP discrete choice experiment
(multinomial logit)
╳
╳
╳
Electricity charging stations
Consumers' preferences in choosing between conventional and new
technologies can change with market conditions.
Axsen et al. (2009)
535 Canadians and
408 Californians
Joint SP/RP data
(multinomial logit)
╳
╳
╳
Investigates the ‘neighbor effect’ and uses SP and RP choice research
to improve the behavioral realism of an energy–economy model.
Hidrue et al. (2011)
3,029 respondents
in the US
SP discrete choice experiment
(latent class random utility model)
╳
╳
Estimates willingness to pay for EV attributes.
Hackbarth and
Madlener (2013)
711 people in
Germany
SP discrete choice experiment
(multinomial logit)
╳
╳
╳
╳
Electricity charging stations;
policy incentives
Conventional vehicles maintain dominance in market; increasing
driving range to that of conventional vehicles has same effect as
multiple policy incentives.
SP: Stated Preferences, RP: Revealed Preferences; “╳” indicates factors investigated in this paper.
2
5 / 27
Because EVs are supposed to solve emission problems, the environmental component is regarded as an
important aspect of the purchase. Scholars consider not only CO2 emissions (Shepherd et al. 2012) and the
reduction of pollution (Potoglou and Kanaroglou 2007), but also consumers’ level of environmental
consciousness (Ewing and Sarigöllü 1998). Ewing and Sarigöllü (1998) found that over one third of respondents
were willing to pay CAN $1000 more for a vehicle if it caused substantially lower emissions.
Today, services have become key not only to customer satisfaction but also to promoting new products in
the vehicle market (Fassnacht et al. 2011). Unfortunately, there is little research on this topic, especially for EVs,
a gap this study aims to close. To the best knowledge of the authors, previous research has mainly integrated the
availability of electricity charging stations, which is a mandatory infrastructure element for the success of EV
adoption. Hackbarth and Madlener (2013) predict that multiple policy incentives, such as permission for EVs to
drive in bus lanes or vehicle tax reductions, will raise the shares of EVs markedly. Although policy incentives are
important factors, their implementations tend to be exogenous to most car manufactures. Therefore, we propose
that research should keep pace with the development of the newest services complementing consumers’ driving
experience. We will introduce such services in detail later in Section 3.
Overall, even though all four main factors can have a significant influence on the adoption of EVs, in this
study we focus more on complementary mobility services. We take on the idea of examining technological and
economic factors which can help us to better understand additional routes to foster the adoption of electric
vehicles.
2.2 Theoretical Foundation
Because transactional data on complementary mobility services are not available, or likely do not exist, we turn
our attention to stated preference methods. These methods collect data using surveys, which cost less than setting
up test-markets and which simplify data analysis because factors of interest can be experimentally manipulated
and tested in a controlled research environment. According to Rao (2014), we classify stated preference methods
into four classes: (rating- or ranking-based) conjoint analysis, discrete choice experiments, self-explicated
methods, and hybrid methods (see Fig. 1).
Stated Preference
Methods
Rating- or
Ranking- based
Conjoint Analysis
Discrete Choice
Experiments Self-Explicated
Methods Hybrid Method
2-Step Approach
Within Same
Study
2-Step Approach
Across Studiies
Fig. 1: Classification of stated-preference methods
Rating- or ranking-based conjoint analysis, frequently used in the 1970s and 1980s, allows respondents to
rate or rank products, which are described by the particular attributes selected for the study (Green et al. 2001).
Those observations are then used to calculate respondents’ preferences by estimating the contribution of each
attribute and level to the observed outcome. The origins of these methods lay in psychology and are principally
associated with research dealing with ways to mathematically represent the behavior of rankings observed as an
6 / 27
outcome of systematic factorial manipulation (i.e., known as "factorial designs") of independent factors (also
known as “attributes”). As Louviere et al. (2010) outline, these methods rely on formal proofs about the
mathematical (algebraic) representations of rank orderings of orthogonal arrays (originally complete factorial
arrays). The effort required to perform these methods is rather low; however, the number of selected attributes
and levels should not be too high in order to keep the number of products, and thus respondents’ cognitive
burden, at a feasible level. In addition, deriving managerial recommendations, e.g., through what-if analysis or
counterfactual simulation, is only possible under strong and theoretically unsupported assumptions because the
observations do not readily translate into choices. Consequently, observations cannot be analyzed consistently
with neoclassical economic theory to simulate respondents’ choices, i.e., researchers must apply arbitrary
selected probability models (e.g., first choice, probabilistic choice, or logit), which substantially differ in their
managerial recommendations, though none of them are theoretically supported.
Instead of relying on ratings or rankings that are artificially translated into preferences, discrete choice
experiments allow respondents to repeatedly make choices between a set of alternative products. Given the
similarity to real-world purchase decisions, discrete choice experiments are able to explain actual purchasing
behavior well, and they have a firm foundation in sociology and behavioral research (Swait and Andrews 2003).
More specifically, with random utility theory, these experiments are backed-up by a long-standing, well-tested
theory of choice behavior that can take inter-linked behaviors into account (see McFadden 2001). Louviere et al.
(2010), in their comparison of discrete choice experiments with conjoint analysis, conclude that random utility
theory provides an explanation of the choice behavior of humans, not numbers.
1
However, because each
observed choice only provides a marginal amount of information, some researchers recommend conducting
discrete choice experiments only in the case of a few attributes (e.g., 3-8), which might become difficult as the
complexity of the studied product increases.
Self-explicated methods use direct assessments of attributes and their importance or apply various types of
adaptive models involving ratings of pairs of alternatives, often on the basis of a partial set of the attributes (see
Schlereth et al. 2014). Although most studies could not empirically demonstrate the superiority of discrete choice
experiments over conjoint or self-explicated methods (Schlereth et al. 2011), the same argument as against
conjoint analysis applies here, namely, that these methods also lack a direct link to respondents’ actual choices.
For the purpose of this research, we use a hybrid approach that also addresses the issue of a large number of
attributes and levels by limiting their number before presenting them to respondents. Essentially, hybrid
approaches involve two steps that can be made within one study (so that a respondent has to go through both
steps) or separated into two studies. An appealing aspect of the hybrid approach is that the researcher is free to
choose the most suitable stated preference method in each step.
In our study, we rely on two types of discrete choice experiments, given their firm theoretical and
behavioral foundations, and given that we are particularly interested in consumer choices of an EV, when adding
complementary mobility services. Step 1 is used to determine the most important attributes. Because sufficient
knowledge about product attributes of EVs are already available (see Section 2.1), we can concentrate our
attention on the selection of consumers’ most important complementary mobility services. Step 2 is a discrete
choice experiment using a limited set of attributes (i.e., the most important complementary mobility services
combined with the product attributes) to study consumer preferences for any of them. By this way, the limitation
with respect to the number of attributes and levels of conjoint analysis and discrete choice experiments is
mitigated.
1
See the subsequent Section 2.3.
7 / 27
2.3 Random Utility Theory as the Core of Modeling Consumer Decision Making
In both types of discrete choice experiments, later presented in the two empirical studies, respondents inspect
two or more alternatives and are assumed to choose the one which maximizes their utility. The observed choices
provide researchers with rich insights into consumer preferences. In marketing research as well as in economics
random utility theory is applied on such observations (McFadden 1974; Thurstone 1927). Random utility theory
assumes that respondent h decomposes his or her utility uh,i for alternative i into a deterministic component vh,i
and a stochastic component εh,i, i.e., uh,i = vh,i + εh,i. This means that, respondents’ utility can be decomposed into
components that are directly related to the attributes and levels shown to the respondent and components that are
known to the respondent, but they cannot be observed by the analyst. The stochastic component accounts for
Thurstone (1927) realization that respondents make errors in their choices, which means that they are not
necessarily always choosing the alternative with the highest deterministic utility. McFadden (2001)
generalization of Thurstone's RUT model provides tractable, closed-form models that accommodate choices
from sets of three or more alternatives (see Train 2009). More formally, the probability Prh,i of respondent h
choosing alternative i among a set of alternatives jϵJ is:
(1)
, , , , ,
Pr Prob(v v ; )
h i h i h i h j h j ji
,
We can rewrite that to:
(2)
, , , , ,
Pr Prob( ; )
h i h j h i h j h i
v v j i
.
Assuming a Gumbel distribution for the stochastic component εh,j, its density is
,
,
,
f( ) hj
hj e
hj ee
and
its cumulative distribution is
,
,
() hj
e
hj
Fe
. By assuming independence between all error terms j ≠ i, we
can rewrite Equation (2) as the cumulative distribution given the error term εh,i:
(3)
, , ,
,,
Pr | vv
h i h j h i
e
h i h i ji
e
.
Of course, observing εh,i is not possible, but weighting all possible values of εh,i in Equation (3) by their density
results in the following integral:
(4)
, , , ,
,
,,
Pr vv
h i h j h i hi
hi
ee
h i h i
ji
e e e d
.
Finally, we can apply algebraic manipulations and obtain the probability Prh,i of respondent h selecting product i
among a set of product alternatives J in a mathematically convenient closed form:
2
(5)
,
,
,
Pr hi
hj
v
hi v
jJ
e
e
.
2
See p. 34ff in Train (2009), especially Section 3.10, for the mathematical derivation of the algebraic manipulations as
well as the underlying assumptions.
8 / 27
3 Complementary Mobility Service Operationalization
In cooperation with a German consultancy firm, and based on elaborate discussions with EV experts (three
workshops with three experts plus four telephone interviews with CEOs of car sharing companies) as well as the
analysis of industry reports, we identified nine important complementary mobility services (see Table 2). The
nine selected services are frequently mentioned in media, and expectations are high that they might complement
EV technology. Some of the services can increase consumers’ perceived ease-of-use and usefulness by reducing
time-dependent costs (e.g., saving the time required to pay and park) or by increasing brand loyalty by forming
online communities (Algesheimer et al. 2006) (e.g., social network app in car); others offer new driving
experiences and make driving more exciting and intelligent (e.g., providing visual real-time updates on traffic
information).
All nine complementary mobility services can be classified into two classes. One class of services is
specific for electric vehicles and the other class can — in principle — also be integrated in traditional vehicles.
The first class includes “Intelligent charging stations” and “Vehicle-to-Grid” (V2G), two services that certainly
make only sense for EV. “Intelligent charging stations” is a demand side management instrument that could be
used to improve energy efficiency, reduce time of use, allow quick demand response, and enlarge the spinning
reserve (Palensky and Dietrich 2011). “Vehicle-to-Grid” (V2G) is an energy system that realizes large synergies
between the vehicle fleet and the electricity system. For society, the advantages of developing V2G include an
additional revenue stream for cleaner vehicles, increased stability and reliability for the electric grid, lower
electric system costs, and (eventually) inexpensive storage and backup capacity for renewable electricity
(Kempton and Tomić 2005). For consumers, V2G might serve as another source of income if electricity
providers offer real time prices for energy; they can charge their battery, when energy costs are low, and feed in
the electricity grid in phases of high demand and high prices.
9 / 27
Table 2: Complementary Mobility Services
Complementary Mobility Services
Explanation
Intelligent charging station
Intelligent charging stations simplify charging the EV battery. They enable to automatically identify
drivers and to bill energy consumption.
Vehicle-to-Grid (V2G)
To realize substantial synergies between the EV fleet and the electricity system, V2G refers to the return
of electricity from the battery of an EV into the electric grid. Drivers can help mitigate peak demand
shocks and earn money at the same time: They can charge the battery, when energy costs are low and feed
in the electricity grid in phases of high demand and high prices.
IT-based parking
and payment
IT-based parking systems directly guide drivers to parking spaces and allow them to pay easily and
automatically.
Drive-through for
bill payment
Bills may be authorized and paid directly from the EV for certain products or services (e.g., fuel bills,
parking fees, or tolls).
Connection to
mobility providers
By contracting with mobility providers, drivers may rent and switch batteries offered by mobility
providers. Moreover, mobility providers offer intelligent services (such as traffic or travel information)
that can also be booked.
Remote diagnostics
and updated supply
The software (e.g., operating system) adopted in EVs may be remotely controlled and updated by car
repair shops. Meanwhile, remote diagnostics may be offered in the event of errors or defects.
In-car apps, purely vehicle-related function
In-car apps are software applications that equip EVs with additional functions directly related to driving
(e.g., driver logs, electricity cost logs).
In-car apps, not purely vehicle-related
function
In-car apps that are not directly related to driving, e.g., social media or music apps.
Augmented reality services via head-up
displays
Augmented reality services automatically identify and project relevant information on the windshield via a
head-up display. The mentioned examples include navigation, information about electricity consumption,
prices for nearby recharging stations as well as hotel and restaurant recommendations.
The other seven services belong to the second class and can improve the utility of both, electric and
conventional vehicles. For example “Augmented reality services via head-up displays”, which is not restricted to
electric engines, use the windshield as a projection surface for displaying virtual content and may help drivers
detect and respond to traffic changes more quickly and increase navigational accuracy (Fadden et al. 1998),.
Although these seven complementary services would be or have already been used in conventional vehicles,
their availability might contribute more utility for electronic vehicles than for conventional cars. Taking an
example, bundling new complementary services like “IT-based parking” with EVs would make the EV market
more attractive. Advertisements and word-of-mouth conversations about self-parking could firmly grasp
customers’ attentions and create a strong innovation image for these companies and their products. Customers
always pay more attention to innovations and might regard such developments as a breakthrough that could help
them move quickly through the first stages (create product awareness) in the adoption process (Armstrong et al.
2010). Although EV development and service innovations are relatively independent, more innovative
companies should have more experience and passion on new product development. In turn, that could increase
the success of new products (higher sales and longer sale duration in this situation) (Cooper and Kleinschmidt
1987). Facts also supported our inferences: the world’s first experimental prototypes of automatic parallel
parking was developed at INRIA on a Ligier electric car in the mid-1990s (Paromtchik and Laugier 1998), and
BMW announced in January 2014 the following for their new “IT-based parking” service that is first exclusively
available for their i-Series (Brigl 2014).
10 / 27
Moreover, complementary mobility services could be aligned with the special needs of EVs. Strongly
market-oriented car manufacturers are well-advised to not only sell their EVs at a low price and to advertise the
technology itself but to also think about improving the holistic driving experience (Armstrong et al. 2010). For
example, the combination of “Augmented reality services via head-up displays” and EV can create a better
holistic driving experience. Obviously, the limited driving range is one of the largest barriers for EVs adoption.
“Augmented reality services via head-up displays” could provide access to vehicle and environment-related
information, such as the current driving conditions and battery charge, route guidance and nearest charging
stations and potentially, with which even could educate drivers in their energy efficient driving skills.
4 Study 1: Identification of Most Important Complementary Mobility
Services
Table 2 shows that the number of potentially interesting complementary mobility services is quite large. Since
many traditional methods like conjoint analysis cannot deal with too many attributes, we address the problem of
studying preferences of complementary mobility services in conjunction with other EV attributes as follows:
Study 1 applies Best-Worst Scaling to identify the most important complementary mobility services among those
listed in Table 2, which will serve later as an input for study 2. We also test whether simply counting how often a
complementary mobility service has been chosen as the best or worst alternative will lead to results similar to
those of more sophisticated estimation methods that have recently been proposed in the literature.
4.1 Best-Worst Scaling (Case 1)
Best-Worst Scaling, introduced in 1992 by Finn and Louviere, has recently grown in popularity. In case 1,
respondents view multiple choice sets that consist of a subset of attributes (in our study, the list of
complementary mobility services from Table 2) and are repeatedly asked to choose their least and most preferred
attributes.
3
Thus, the Best-Worst Scaling forces respondents to trade off attributes of varying attractiveness,
which have binary levels (e.g., exists and does not exist or applies and does not apply). Researchers can then
determine the preference for an attribute by comparing how frequently respondents have chosen that attribute
relative to other attributes in the choice set. Table 3 provides an illustrative example of a choice set employed in
study 1.
Table 3: Example Choice Set in Study 1
Most preferred
Complementary mobility services
Least preferred
X
IT-based parking space and payment
Intelligent charging station
Augmented reality services via head-up displays
X
Compared with other preference measurement methods, Best-Worst Scaling offers several unique
advantages. For example, compared with ranking methods, which are known to yield low accuracy and
consistency when there are more than seven attributes (e.g., Bettman, Johnson, and Payne, 1990), Best-Worst
Scaling can be applied to substantially more attributes. Best-Worst Scaling also avoids the assumption of equal
3
For a conceptual framework of Best-Worst Scaling, see Louviere et al. (2013). There are alternative methods with
supplement “case 2” and “case 3”, in which respondents either choose the most and least preferred level of a product
(case 2) or the most and least preferred alternative described by its attributes and levels (case 3).
11 / 27
differences between two subsequently ranked attributes, which leads to more realistic results. Compared with
verbal measurement scales (e.g., rating tasks or Likert scales), respondents do not use artificially numerical,
subjectively interpretable representatives for their preferences (e.g., agree-disagree scaling) but instead choose
among decisions that are easy to understand and can be made quickly. Results from this method provide a higher
degree of discrimination (Lee et al. 2008), and the interpretation of responses is consistent across respondents.
Therefore, it is a suitable method for cross-cultural studies and studies with respondents of heterogeneous
backgrounds or educational skills.
Another strength of Best-Worst Scaling is that its observations are easy to analyze because simply counting
best and worst choices is sufficient to obtain either individual or aggregate sample preference estimates (Finn
and Louviere 1992; Mueller Loose and Lockshin 2013). Imagine, for example, nine attributes and twelve choice
sets, each consisting of three of these attributes. In the case of a balanced design, each of the attributes appears
four times (= 12 x 3 / 9). Consequently, an attribute may generate Best-Worst scores ranging between -4 and +4,
depending on how frequently this attribute was chosen as the best attribute (+1), as the worst attribute (-1), or not
chosen at all (+0). Adding +5 to all Best-Worst scores will transform them into a range between 1 and 9, which is
the response we would observe in a nine-point rating task. The properties of this range depend on the number of
repetitions of each attribute. Analysis is simple and can be conducted without any proprietary software.
As an alternative to the count analysis, Marley et al. (2008) propose proper probabilistic choice models,
such as the MaxDiff model, which have their foundations in random utility theory for estimation.
4
The
probability BWC(j’,j’’) of choosing the pair of attributes j’ and j’’ in choice set C as the best and worst attributes
is calculated as follows:
(6)
In this model, vj’ is the deterministic utility of attribute j’ on the aggregate level, i.e., one parameter
representing the utility of all respondents. The probability BWC(j’,j’’) is calculated by maximizing the
differences between any chosen pair of the best and worst attributes j’ and j’’ through (vj’-vj’’) relative to the sum
of all possible combinations of attributes r and s in a choice set. The aggregate level utilities are sufficient for the
purpose of our study. Nevertheless, estimating the individual preferences is also possible.
We estimate Equation (6) using a Maximum Likelihood estimator, which was programmed in Matlab. The
denominator of Equation (6) uses permutations to obtain every combination of alternative r and s within a choice
set.
4.2 Set-up
We implemented study 1 including the Best-Worst Scaling task using the online platform DISE (Schlereth and
Skiera 2012). The questionnaire for study 1 consisted of three sections: 1) brief explanations of EVs, which were
followed by the presentation of the nine complementary mobility services listed in Table 2, each explained using
brief textual and pictorial descriptions; 2) the choice sets of Best-Worst Scaling; and 3) demographic and
socio-economic questions.
We presented the nine attributes in 12 choice sets consisting of three attributes each. Using a balanced
incomplete block design (see Table A14 in the Appendix), each attribute appeared four times with a pair
frequency of one. We described the complementary mobility services and provided pictures such that every
respondent could easily understand the services. Respondents were asked to choose the most and least preferred
4
See also Section 2.3.
' ''
, \{ }
exp( )
( ', '') ( ' '')
exp( )
jj
Crs
r s C r s
vv
BW j j j j
vv
12 / 27
complementary mobility service in each of the 12 choice sets.
4.3 Data
Study 1 was conducted in the first quarter of 2013 with a total of 251 completed questionnaires. We recruited
respondents through different channels, such as postings in specialized forums on car-related topics and inviting
colleagues, friends of the authors, and students from two major German universities to participate in the study.
As an incentive, we offered entry in a lottery for 3 gift vouchers valued at 20€ each. Therefore, we consider
study 1 to be a convenience sample rather than a representative sample. Table 4 summarizes the respondents’
demographic characteristics.
Table 4: Demographic Characteristics in Study 1
Gender
Age
Occupation
Male (68.5%)
Female (31.5%)
18-24 (25.9%)
25-34 (44.6%)
35-44 (11.6%)
45-54 (13.1%)
55+ (4.8%)
Unemployed (1.2%)
Employee (57.0%)
Workers (2.8%)
Civil Servants (0.8%)
Pensioners (0.4%)
Freelancers (4.0%)
Students (31.9%)
Others (2.0%)
N=251.
4.4 Results
Table 5 reports the average Best-Worst scores, their standard deviations over individuals in parentheses and the
results of the maximum likelihood estimation. “IT-based parking space and payment” and “Intelligent charging
station” are by far the most desirable complementary mobility services. “Augmented reality services via head-up
displays” and “Remote diagnostics and update supply” follow at third and fourth place, respectively.
“Drive-through payment” and “In-car apps not for purely vehicle-related functions” are not preferred by our
respondents.
13 / 27
Table 5: Results of the Best-Worst Scaling Analysis
Ranking
Additional mobility service
Best –Worst (Differences
between Best and Worst
Score)
Average Best-Worst
Scores
Maximum Likelihood
Estimates
1
IT-based parking space and payment
375
1.49 (1.96)
0.83 (0.70)
2
Intelligent charging station
330
1.31 (2.22)
0.74 (1.00)
3
Augmented reality services via head-up
displays
190
0.76 (2.38)
0.46 (0.89)
4
Remote diagnostics and update supply
129
0.51 (2.30)
0.25 (0.87)
5
In-car apps for purely vehicle-related
functions
74
0.29 (1.99)
0.13 (0.76)
6
Vehicle-to-Grid
-145
-0.58 (2.62)
-0.37 (1.44)
7
Connection to mobility agents
-208
-0.83 (2.45)
-0.49 (1.13)
8
Drive-through payment
-345
-1.37 (2.16)
-0.74 (0.92)
9
In-car apps not for purely vehicle-related
functions
-400
-1.59 (2.10)
-0.86 (0.77)
N=251.
A graphical comparison of the Best-Worst scores against the maximum likelihood estimates that we present
in Fig. 2 reveals that both estimation results are proportional to one another, which indicates high robustness.
Fig. 2: Best-Worst Scores (y) vs. Maximum Likelihood Estimates (x)
This proportionality demonstrates that the simple count analysis is sufficient to derive managerial
recommendations on the ranking of the complementary mobility services and underlines the strength of this
simple trade-off-based stated preference measurement method. However, it remains unclear, how the most
preferred complementary mobility services are perceived relative to technological and economic factors of EV
adoption. We address this gap in the subsequent study.
14 / 27
5 Study 2: Influence of Complementary Mobility Services Relative to
Other Electric Vehicles Attributes
In study 2, we explore the influence of complementary mobility services relative to well-studied factors of EV
adoption. We employ Dual Response (Brazell et al. 2006; Dhar and Simonson 2003), which enables researchers
to also estimate preferences for the levels (e.g., 25,000€ compared to 30,000€) of attributes (here, purchase
price). We subsequently describe Dual Response before presenting the study set-up, data collection, results and
additional managerial insights obtained from a counterfactual simulation.
5.1 Dual Response
Dual Response is a modification of the traditional discrete choice experiment (also frequently referred to as
Choice-Based Conjoint). Instead of combining product alternatives and a no-purchase option in a choice set,
Dual Response repeatedly asks two types of questions: First, which is the most attractive product alternative in a
choice set in a forced choice question without any no-purchase option? Second, in a free-choice question, would
they buy the chosen product alternative? This method is particularly useful if a high proportion of no-purchase
decisions is expected because researchers could then additionally observe the trade-off decision among various
products, thus yielding more accurate parameter estimates and more stable preferences (Brazell et al. 2006).
For the analysis, we apply random utility theory again and decompose respondent h’s utility uh,i for product
i into a deterministic component vh,i and a stochastic component εh,i, i.e., uh,i = vh,i + εh,i. Assuming a Gumbel
distribution for the stochastic component, we express the probability Prh,i for making the observed choices in the
forced and free choice questions, as follows:
(7)
,,
,,
,'
'
exp( ) exp( )
Pr exp(0) exp( )
exp( )
h i h i
hi hi
hi
iI
vv
v
v
.
The first term describes the probability of choosing product i as the most preferred product among the set of
products I, and the second term models the probability of purchasing it. Each vh,i assumes an additive functional
form, i.e., vh,i = βh∙Xi, where beta is a vector of preferences of respondent h for all attributes and a constant. Xi is
a vector specifying the attributes levels of each attribute in product i.
For the estimation, we employ Hierarchical Bayes, which is a powerful instrument that delivers the
distributions of the parameter estimates for respondents at the individual level despite the low number of
observations per respondent. The term “hierarchical” refers to the technique of iterating over the lower individual
level (i.e., respondents’ individual parameters) and the higher population level, assuming that the respondents’
parameters are described by a multivariate normal distribution. In each iteration, it draws candidate values from
the posterior distribution (i.e., the estimates) and “borrows” information from the distribution of the population
level to make predictions about each respondent’s parameters.
We use a normal distribution for all parameters except the price parameter. Here, we assumed a log-normal
distribution, which we multiplied afterwards by (-1) to ensure negative values. Thereby, we employed a vector
model for the attribute price on the deterministic utility and an effect-coded partworth model for all other
attributes. In particular, using a partworth model for electricity cost per 100km instead of a vector model as for
purchase price was motivated by the respective cost per 100km, which were close to 0. We employ standard
diffuse priors on the parameters with means 0.1 and standard deviations of 5, which imply vague prior
knowledge about the parameters. We obtain information about the posterior distribution based on 10,000
iterations that we obtained after discarding a sufficient number of (burn-in) iterations (also 10,000). Convergence
was assessed by examining the trace plot of the posterior’s likelihood. The estimator is programmed in Matlab
15 / 27
and is an extended version of the code provided by Train (2009).
5
For a detailed description of the estimator,
Bayesian concepts, and an introduction in prior and posterior distributions, we refer to Section 12 in Train
(2009).
5.2 Set-up
Study 2 compares the relative importance of complementary mobility services to the previously studied
attributes of the technological and economic categories. A literature review and experience reports on EVs were
used to identify the most important attributes and respective attribute levels (see Table A13 in the Appendix).
The attribute levels we chose (see Table 6) cover the largest part of the current electric vehicles market but also
incorporate likely technological improvements and slightly declining prices. The technological attributes are
“range per charge”, “charging time”, and “motor power”. The economic attributes are “purchase price” and
“electricity costs per 100 km”. The price is slightly lowered (but nevertheless reasonable according to Table A13
in the Appendix) compared with average market prices to reflect expectations about future price developments.
We then add the top three complementary mobility services to the list to explore their effects on the
purchase-decision process.
Table 6: Attributes and Attribute Levels Included in Our Main Study
Unit
Range
Levels
Range per charge
km
4
100; 175; 250; 325
Charging time
h
2
1; 4
Motor power
kW
2
40; 80
Purchase price
€
4
15,000; 20,000; 25,000; 30,000
Electricity cost per 100 km
€
4
1; 3; 5; 7
IT-based parking space and payment
[ ]6
2
supported; not supported
Intelligent charging station
[ ]
2
supported; not supported
Augmented reality services via head-up displays
[ ]
2
supported; not supported
Employing the techniques in Street and Burgess (2007), we created a D-optimal (4∙2∙2∙4∙4∙2∙2∙2) fractional
factorial design with 14 choice sets, i.e., 12 choice sets for the estimation and 2 for the holdouts (see Table A15
in the Appendix). These designs are known for their high efficiency and their suitability for a diverse range of
research questions. Each choice set shows three different EVs and subsequently asks the respondent in a separate
question whether he or she would buy the most preferred EV.
The questionnaire of study 2 consisted of two sections and was also implemented in DISE (Schlereth and
Skiera 2012) . First, we collected demographic information on gender and age and presented basic information
about EVs. At the end of this section, the respondents indicated their interest on a four-point rating scale. Only
the respondents who answered “Yes, I can imagine purchasing an electric vehicle” continued with the remainder
of the survey; for the other respondents, we assumed that their willingness to pay was below the minimum price
level in the discrete choice experiment. The second part continued with the discrete choice experiment.
5
See http://eml.berkeley.edu/~train/software.html.
6
The complementary mobility services are recorded as dummy variables, so there is no unit.
16 / 27
Table 7: Sample’s interest in purchasing an EV
Question
Can you imagine purchasing an electric vehicle?
Number of participants
Four-point rating
scales
Yes, I can imagine purchasing an electric vehicle.
168 ( 51.4%)
No, but I can imagine leasing an electric vehicle.
24 ( 7.3%)
No, but I can imagine using an electric vehicle as part of a car-sharing services.
57 ( 17.4%)
No, I cannot imagine using an electric vehicle at all.
78 ( 23.9%)
5.3 Data
We hired a market research firm that collected a representative sample of the German population with respect to
gender and age in April 2013. We obtained 327 completed questionnaires. A total of 168 of the respondents
(51.4%, see Table 7) reported having sufficient interest in purchasing an EV in the future and entered the
analysis, which is quite high compared with the 16% completion rate in California in 1998 and the 30%
completion rate in a large Swedish city in 2001 (Gärling and Thøgersen 2001). We excluded the fastest 10% of
the respondents (i.e., 18 respondents) to ensure that only respondents who did not click through the survey
entered the analysis. Therefore, a total of 150 completed questionnaires were considered for further evaluations.
Table 8 summarizes the respondents’ demographic characteristics.
Table 8: Demographic Characteristics in Study 2
Gender
Age
Occupation
Male
(56%)
Female (44%)
18-24 (10%)
25-34 (18.7%)
35-44 (15.3%)
45-54 (23.3%)
55-64 (22.7%)
65-74 (8%)
75-84 (2%)
Unemployed (1.3%)
Employee (45.3%)
Workers (7.3%)
Civil Servants (2.0%)
Pensioners (20.0%)
Freelancers (8.7%)
Students (8.0%)
Pupils (0.7%)
Others (5.3%)
Unspecified (1.3%)
N=150.
5.4 Results
Based on the Dual Response choices, we estimate the parameter values and derive importance weights of the
attributes (see Table 9). Signs and magnitudes of the parameter values are reasonable and provide face validity.
Internal validity is high, with a first choice-hit rate of 92.3% in the within-sample choice sets. Predictive validity
is also high, with 75.8% of the hold-out choice sets being correctly predicted.
The importance weights averaged across all respondents and listed in Table 9 demonstrate that electricity
cost (for 100 km) has the highest importance (25.03%) which means that the recurring cost is the most important
attribute when customers consider buying an EV. The aggregated parameter values of electricity cost gradually
decrease, and the maximum decline of 1.58 occurs when electricity costs per 100 km rises from 5€ to 7€. Range
per charge closely follows as second most important attribute (21.85%). We observe a substantial increase in
utility when the range per charge increases from 100 km to 175 km. In that case, aggregated parameter values
increase by 1.92. There is no substantial increase (only 0.43) in utility when the range increases from 250 km to
325 km. Surprisingly, purchase price only ranks fifth (10.37%); however, it has one of the highest standard
17 / 27
deviations (i.e., 11.56%), indicating heterogeneous preferences.
These numbers differ from results for traditional vehicles markets where purchase price almost means
everything. This is mainly due to two reasons: First, drivers preferring electric vehicles are likely to be
environment-friendly people or technology enthusiasts, who might be more concerned about emission rates or
technology innovation than about purchase price (Potoglou and Kanaroglou 2007). Second, the high fuel price is
probably one of the main reasons why people consider purchasing an EV (Hidrue et al. 2011). With this in mind,
they definitely pay more attention to electricity costs than to purchase price.
Table 9: Parameter Estimates of Study 2
Attributes
Levels
Aggregated parameter values
(Standard Deviation)
Average Importance weights
(Standard Deviation)
Constant
-0.88 (5.96)
Range per charge100 km
100 km
-1.87 (1.21)
175 km
0.05 (0.64)
21.85% (9.73%)
250 km
0.70 (0.69)
325 km
1.13 (0.85)
Charging time
1 hour
0.29 (0.49)
5.65% (4.68%)
4 hours
-0.29 (0.49)
Motor power
40 kW
-0.48 (0.83)
9.63% (9.33%)
80 kW
0.48 (0.83)
Purchase price
(per 1,000€)
-0.13 (0.24)
10.37% (11.56%)
Electricity cost per 100 KM
1€
1.54 (1.23)
25.03% (11.59%)
3€
0.86 (0.70)
5€
-0.41 (0.75)
7€
-1.99 (1.51)
IT-based parking space and payment
supported
0.80 (0.66)
10.38% (7.38%)
not supported
-0.80 (0.66)
Intelligent charging station
supported
0.78 (0.61)
10.42% (7.93%)
not supported
-0.78 (0.61)
Augmented reality services via
head-up displays
supported
0.38 (0.57)
6.67% (5.78%)
not supported
-0.38 (0.57)
N=150
Importance weights of all three complementary mobility services (“IT-based parking space and payment”,
10.38%; “Intelligent charging station”, 10.42%; “Augmented reality services via head-up displays”, 6.67%)
together exceed the importance of electricity cost and reach 28.01%. Furthermore, we conducted an analysis of
variance (One-way ANOVA) that indicated that gender has an influence on the importance weight of
“Augmented reality via Head-up displays” (see Table 10). We can reject the hypothesis that there are no
differences among 6 age groups on the importance weight of “IT-based parking space and payment” and
“Intelligent charging station” (see Table 11). Men are more attracted than women to technological innovations
(such as “Augmented reality via head-up displays”), and older people prefer complementary mobility services
that offer additional convenience more than younger people. Other socio-demographic characteristics have no
significant effect on the level of importance weights.
18 / 27
Table 10: Influence of Gender on Importance Weights of Properties
Female
N=66
Male
N=84
Significant differences
between groups
Augmented reality via head-up displays
(Average weights)
5.65%
7.47%
p< 0.1
N=150
Table 11: Influence of Age on
Importance Weights of Properties
18-24 year
N=15
25-34 year
N=28
35-44 year
N=23
45-54 year
N=35
55-64 year
N=34
65-84 year
N=15
Significant differences
between groups
IT-based parking space
and payment
(Average weights)
8.64%
9.14%
10.37%
10.19%
9.00%
18.03%
p< 0.001
Intelligent charging
station
(Average weights)
10.94%
10.58%
7.36%
8.77%
12.04%
14.48%
p< 0.1
N=150
6 Counterfactual simulation
Using a counterfactual simulation, we highlight the managerial insights emerging from the results of our two
studies. Managers usually want to acquire better knowledge about how prices or technical capabilities of EVs
affect adoption rates and other relevant variables to make informed decisions.
In a stylized scenario, we consider current market offers of EVs, which we label as follows: status quo, i.e.,
purchase price of 40.000€; charging time of 4 h; range per charge of 175 km; and electricity costs of 5€ per 100
km (see Appendix). For ease of illustration, we do not differentiate between different brands of EVs, we consider
static prices that do not change over time, and we neglect the associated costs within each improvement. We
predict market share si of buying an EV i using Equation (8). For each respondent h, we calculate their
deterministic utility vh,i of purchasing a specific EV i. We average this probability (where |H| indicates the
number of respondents) and multiply the value with 51.4%, i.e., the share of respondents ssufficientinterest who had
sufficient interest, to account for the fact that not everyone entered the discrete choice experiment.
(8)
,
sufficientinterest ,
exp( )
1
| | exp(0) exp( )
hi
ihi
h
v
ss Hv
.
The result helps us to assess the change in probability after improving each of the attributes one by one. We
also test the change in the probability when introducing the complementary mobility services. These predictions
might help EV manufacturers prioritize technical innovations and analyze potential increases in adoption rate
relative to the costs of these innovations. The results are summarized in Table 12.
19 / 27
Table 12: Results of the Counterfactual Simulation
Status Quo
40.000€
30.000€
5€ per 100 km
3€ per 100 km
175 km range
250 km range
Purchase probability
2.85%
4.01%
4.85%
3.50%
Change in probability
compared to status quo
+1.16%
+2.00%
+0.65%
Status Quo
IT-based parking
space and payment
Intelligent charging
station
Augmented reality
services via head-up
displays
With all three
mobility
services
Purchase probability
2.85%
5.15%
4.33%
3.93%
9.42%
Change in probability
compared to status quo
+2.30%
+1.48%
+1.08%
+6.57%
We obtain market share predictions of 2.85% for EVs. Decreasing purchases prices by 25% only increases
purchase probability by 1.16%. This is substantially lower than the 2.00% probability for EV manufacturers who
are able to develop new technologies that enable consumers to decrease recurring cost (the electricity costs per
100 km from 5€ to 3€). However, this increase in purchase probability is rather low compared to the impact of
adding our three complementary mobility services. The latter adds 6.57% to purchase probability. Support for
“IT-based parking space and payment” increases the share by 2.30%, and both “Intelligent charging station” and
“Augmented reality services via head-up displays” are able to foster EV adoption comparable to a 10,000€ price
cut. Therefore, complementary mobility services might be a good lever for fostering the adoption of EV. Still,
only a mix of improvements in the attributes can yield the anticipated success.
At first sight, this result might be surprising and seem counterintuitive, given that complementary mobility
services will improve the share of purchases more strongly than substantial price cuts. Nevertheless, we believe
the findings are justified for the following reason: We acknowledge that prices of approximately 30.000€ are
rather high and exceed Germany’s 2012 average prices for a new car (i.e., 26,780€) and for a second-hand car
(i.e., 12,730€).
7
Therefore, even with a 10,000€ discount, EVs would still be considered as premium products.
The segment of premium product buyers is known to be less price sensitive. Instead, potential buyers will be
more interested in improvements in the holistic driving experience. Thus, adding unique capabilities, such as the
proposed complementary mobility services, might be a better lever of EV adoption than just tweaking the
specifications or prices.
8
7 Conclusions
Building upon a two-step approach using Best-Worst Scaling and Dual Response, we not only estimate the
importance of different complementary mobility services but also enhance the general state of science in the
fields of EV adoption and survey methodologies. First, our findings show that the adoption rate for EVs is
expected to be higher than in previous years. Specifically, 51.4% of the respondents could actually imagine
buying an EV. Second, “IT-based parking and payment”, “Intelligent charging stations”, and “Augmented reality
services via head-up displays” are the top three services preferred by consumers. These services can indeed
foster the adoption of EVs along with other technological and economic factors. Third, instead of the purchase
7
Source: de.statista.com.
8
As a robustness test, we use another status quo with a purchase price of 30,000€ and electricity costs of 3€ per 100 km.
The general insights do not change, as reported in the Appendix.
20 / 27
price, recurring cost (electricity cost) is the most important attribute individuals consider when thinking about
adopting an EV. These results are very encouraging and useful for the electric automobile industry. Fourth,
hybrid stated preference methods have proved to be an effective and efficient survey methodology in the
adoption field.
The influences of specific complementary mobility services on the adoption of EVs are examined together
with technological and economic factors, which make the EV demand literature more comprehensive and
abundant. The top three complementary mobility services we selected from Best-Worst Scaling (study 1) have
high importance weights (“IT-based parking and payment”: 10.38%, “Intelligent charging stations”: 10.42%; and
Augmented reality services via head-up displays: 6.67 %), as the Dual Response (study 2) shows. Offering these
top three services could increase the purchase probability to 9.42%, which is a strong improvement compared to
the former 2.85%. Thus, our results confirm that these services may significantly affect the adoption rate of EVs
and should thus be carefully considered by policymakers and the automobile industry.
As previous studies have concluded, electricity costs and range per charge are two of the most important
factors that foster or hinder EV adoption. According to the results of our study, the range per charge should be
175 km or more, and electricity costs should be reduced as much as possible. However, unlike previous studies,
we found that the purchase price plays a minor role for the respondents in our German sample. Compared to
importance weights of electricity costs (25.03%) and the range per charge (21.85%), the importance weight of
purchase price, at 10.37%, appears to be relatively low (see Table 9 for full information). This finding should be
considered, and it might be beneficial to offer EVs for a higher purchase price and subsidize certain recurring
costs, such as electricity costs, through the purchase price increase. It might also make sense to use the best and
most expensive technologies to reduce electricity costs. Prospective buyers pay more attention to recurring costs
than to purchase price. Moreover, price elasticity is rather low compared to offering complementary mobility
services. Decreasing purchase price from 40,000 € to 30,000 only increases purchase probability by 1.16%
which is even less than the increase from adopting the “IT-based parking space and payment” and “Intelligent
charging” services. This result emphasizes that consumers consider the holistic driving experience as more
important than a lower purchase price. Thus, instead of a price-cutting strategy, offering complementary mobility
services seems to be a promising strategy.
Moreover, the analysis of the data obtained in this study reveals that for car manufactures and other
(potential) market participants, potential consumers exhibit a wide range of demand due to a heterogeneous
preference structure. For example, men are significantly more interested in EVs and electric mobility in general
and show more passion in augmented reality services via head-up displays compared to women, whereas older
consumers are attracted by high-quality parking and charging services. Therefore, a segmentation strategy may
be fruitful in the EV market.
From a methodological point of view, our paper’s contribution is to summarize the classification of
stated-preference method and then to propose and demonstrate a two-step approach (hybrid methods) using the
Best-Worst Scaling method prior to Dual Response. The first step serves to identify a subset of the most
preferred complementary mobility services. Best-Worst scaling is particularly helpful when preferences need to
be captured by a relatively small sample. Making the required decisions simply requires selecting the best and
worst attributes and thus is fairly simple for respondents. The interpretation of responses is consistent across
respondents, even in the presence of heterogeneity with respect to knowledge and cultural background. Also, the
data analysis is simple and does not require complex statistical knowledge. As we demonstrate with our study,
the simple count analysis using Best-Worst scores provides results consistent with the more sophisticated random
utility theory aligned MaxDiff model. In a second step, we integrated the most preferred complementary
21 / 27
mobility services in Dual Response to study their impact on purchase decisions relative to other attributes
frequently considered in prior research. By doing that, we reduced the number of choice sets and alternatives to
an acceptable level. We thus expect that this two-step approach will also be useful in other research domains.
Future studies might deep-dive beyond our reasoning in Section 6 and investigate further into the causes of
why the impact of purchase price was relatively low compared to the complementary mobility services. One way
might be to test for interaction effects between complementary mobility services and purchase price. Such tests
require different design generation processes that explicitly incorporate potential interaction effects, though. In
addition, we could not test, whether the number of levels or their ranges affected the high relative importance of
complementary mobility services. For example, we used 4 levels for price and only 2 levels for each
complementary mobility service (i.e., available and not available). However, here for example, De Wilde et al.
(2008) would predict that the higher number of levels for price would rather increase its importance compared to
the 2 levels attributes. Moreover, if the complementary mobility services become state-of-the-art and are no
longer innovative, it’s likely that their importance decrease quickly compared to price. As a result, car
manufactures as well as further research should keep pace with the times.
22 / 27
8 Appendix
Table A13: Marketing Research on Electronic Vehicles
Brand
Model
Purchasing
price
Range per
charge
Top
speed
Electric
cost9
Charging time
Motor
power
(€)
(km)
(km/h)
(€/100 km)
(h)
(kW)
Tazzarri
Zero
24,499
140
100
2.20
9
15
Renault
Kangoo Z.E.
23,800
160
130
3.88
3.96
44
Renault
Zoe
20,600
210
135
3.65
2.60
65
Volvo
C30 Electric
145
130
4.15
3.90
80
Mitsubishi
i-Miev
34,990
160
130
3.13
7
47
Citroen
C-Zero
34,164
150
130
3.15
6
49
Opel
Ampera
42,900
55
161
3.38
4
111
Chevrolet
Volt
41,950
55
161
3.38
6.50
111
Peugeot
iOn
29,393
150
130
3.15
3.51
49
Ford
Focus Electic
40,000
160
136
3.85
4.13
107
Citroen
Berlingo First
Electic
59,694
95
95
5.46
28
Buddy
Pure Mobility AS
26,989
120
80
2.20
8
13
Nissan
Leaf
35,000
160
145
3.75
3.90
80
Tesla
Roadster
118,000
395
201
3.30
215
BMW
Mini E
34,950
175
152
3.50
3.5
150
Daimler
Smart Electric Drive
23,000
115
112
3.00
4.29
30
Ford
Transit Connect
Electric
53,544
130
120
5.43
5.46
105
S.A.M.
Group
Sam EV II
16,600
100
90
2.00
5
11.6
DFM Mini
Auto
Van EQ 6380
15,988
120
85
5.00
4.03
Collected from: 1. http://www.adac.de/infotestrat/autodatenbank/suchergebnis.aspx
2. http://electric-car-database.com/de/?h=n
9
Electric cost is calculated by electric consumption (kWh/100km) times the electricity price in Germany
(0.25€ per kWh).
23 / 27
Table A14: Design of Study 1 (Each Row Represents a Choice Set and Each Cell the Attribute
Index of Each Alternative)
Alternative 1
Alternative 2
Alternative 3
2
4
8
3
5
9
4
7
9
1
2
3
2
5
7
3
4
6
2
6
9
1
8
9
1
4
5
Table A15: Design of Study 2 (Each Row Represents a Choice Set and Each Cell the Level-Index
of Each Attribute per Alternative)
Alternative 1
Alternative 2
Alternative 3
2
1
0
2
0
1
0
1
0
2
1
1
1
0
1
0
1
0
0
0
1
0
0
1
1
0
1
2
0
1
1
0
2
2
0
1
1
0
0
1
3
3
1
3
0
1
1
0
2
0
1
1
0
0
1
1
3
1
1
3
1
1
1
1
0
3
0
0
0
0
0
0
1
2
0
3
1
1
0
0
3
3
0
2
1
1
0
0
0
1
1
0
0
0
1
1
1
3
1
1
0
1
0
1
3
2
0
0
0
1
1
1
2
1
0
2
1
0
1
0
1
3
0
1
0
1
1
1
3
1
1
0
1
0
0
0
0
0
1
3
1
0
0
0
2
0
0
3
0
0
1
0
3
1
0
1
0
0
0
0
0
2
1
2
1
1
0
1
*1
2
0
2
0
0
0
0
0
0
0
3
1
1
1
1
2
3
1
0
1
1
1
1
2
2
1
0
0
1
0
0
1
1
1
3
0
0
0
1
3
3
0
2
1
0
1
1
3
2
0
3
0
0
0
1
1
1
1
0
1
1
1
0
0
0
1
1
1
1
1
0
1
2
0
0
1
0
1
1
2
3
1
2
1
0
0
1
0
0
0
1
0
1
1
0
3
0
1
2
0
1
0
0
0
3
0
3
1
0
1
1
2
1
1
1
0
1
0
0
*0
1
0
0
0
1
0
1
3
0
0
1
1
1
0
1
1
2
1
2
0
0
1
0
2
2
0
2
0
1
1
1
1
0
1
3
1
1
0
1
0
3
0
0
0
0
1
0
* Choice sets that were used for holdout predictions.
24 / 27
Table A16: Robustness Test of Counterfactual Simulation
New Status Quo
(30,000€ and 3€ per 100 km)
30.000€
20.000€
3€ per 100 km
1€ per 100 km
175 km range
250 km range
Purchase probability
6.79%
9.25%
7.96%
7.38%
Change in probability
compared to status quo
+2.46%
+1.17%
+0.60%
New Status Quo
(30,000€ and 3€ per 100 km)
IT-based parking space
and payment
Intelligent charging
station
Augmented reality
services via head-up
displays
With all three
mobility services
Purchase probability
6.79%
10.51%
9.31%
8.69%
17.77%
Change in probability
compared to status quo
+3.73%
+2.52%
+1.90%
+10.98%
25 / 27
9 References
Aftabuzzaman M, Mazloumi E (2011) Achieving Sustainable Urban Transport Mobility in Post Peak
Oil Era. Transport Policy 18 (5):695-702
Algesheimer R, Herrmann A, Dimpfel M (2006) Die Wirkung von Brand Communities auf die
Markenloyalität—eine dynamische Analyse im Automobilmarkt. Zeitschrift für
Betriebswirtschaft 76 (9):933-958
Armstrong G, Cunningham MH, Kotler P (2010) Principles of Marketing. Pearson Education, New
Jersey, pp 63-65
Axsen J, Mountain DC, Jaccard M (2009) Combining Stated and Revealed Choice Research to
Simulate the Neighbor Effect: The Case of Hybrid-Electric Vehicles. Resource and Energy
Economics 31 (3):221-238
Beggs S, Cardell S (1981) Assessing the Potential Demand for Electric Car. Journal of Econometrics 17
(1):1-19
Bettman JR, Johnson EJ, Payne JW (1990) A Componential Analysis of Cognitive Effort in Choice.
Organizational Behavior and Human Decision Processes 45 (1):111-139
Brazell JD, Diener CG, Karniouchina E, Moore WL, Séverin V, Uldry P-F (2006) The No-Choice
Option and Dual response Choice Designs. Marketing Letters 17 (4):255-268
Brigl S (2014) BMW at the Consumer Electronics Show (CES) in Las Vegas 2014.
https://www.press.bmwgroup.com/global/pressDetail.html?title=bmw-at-the-consumer-electro
nics-show-ces-in-las-vegas-2014&outputChannelId=6&id=T0162885EN&left_menu_item=n
ode__5238. Accessed December 2014
Brownstone D, Bunch DS, Train K (2000) Joint Mixed Logit Models of Stated and Revealed
Preferences for Alternative-Fuel. Transportation Research Part B: Methodological 34
(5):315-338
Bunch DS, Bradley M, Golob TF, Kitamura R, Occhiuzzo GP (1993) Demand for Clean-Fuel Vehicles
in California: A Discrete-Choice Stated Preference Pilot Project. Transportation Research Part
A: Policy and Practice 27 (3):237-253
Cooper RG, Kleinschmidt EJ (1987) New Products: What Separates Winners From Losers? Journal of
Product Innovation Management 4 (3):169-184
Dagsvik JK, Wennemo T, Wetterwald DG, Aaberge R (2002) Potential Demand for Alternative Fuel
Vehicles. Transportation Research Part B: Methodological 36 (4):361-384
De Wilde E, Cooke AD, Janiszewski C (2008) Attentional Contrast during Sequential Judgments: A
Source of the Number-of-levels Effect. Journal of Marketing Research 45 (4):437-449
Dhar R, Simonson I (2003) The Effect of Forced Choice on Choice. Journal of Marketing Research 40
(2):146-160
Ewing G, Sarigöllü E (2000) Assessing Consumer Preferences for Clean-Fuel vehicles: A Discrete
Choice Experiment. Journal of Public Policy and Marketing 19 (1):106-118
Ewing GO, Sarigöllü E (1998) Car Fuel-Type Choice under Travel Demand Management and
Economic Incentives. Transportation Research Part D: Transport and Environment 3
(6):429-444
Fadden S, Ververs PM, Wickens CD Costs and Benefits of Head-Up Display Use: A Meta-Analytic
26 / 27
Approach. In: Proceedings of the Human Factors and Ergonomics Society Annual Meeting,
1998. SAGE Publications, pp 16-20
Fassnacht M, Stallkamp C, Rolfes L (2011) Betriebsformen im Automobilhandel – Resultate einer
empirischen Untersuchung. Zeitschrift für Betriebswirtschaft 81 (11):1181-1203
Finn A, Louviere JJ (1992) Determining the Appropriate Response to Evidence of Public Concern: the
Case of Food Safety. Journal of Public Policy and Marketing 11 (2):12-25
Gärling A, Thøgersen J (2001) Marketing of Electric Vehicles. Business Strategy and the Environment
10 (1):53-65
Green PE, Krieger AM, Wind Y (2001) Thirty Years of Conjoint Analysis: Reflections and Prospects.
Interfaces 31 (3):S56-S73
Hackbarth A, Madlener R (2013) Consumer Preferences for Alternative Fuel Vehicles: A Discrete
Choice Analysis. Transportation Research Part D: Transport and Environment 25 (1):5-17
Hidrue MK, Parsons GR, Kempton W, Gardner MP (2011) Willingness to Pay for Electric Vehicles and
Their Attributes. Resource and Energy Economics 33 (3):686-705
Hinz O, Schulze C, Takac C (2014) New Product Adoption in Social Networks: Why Direction Matters.
Journal of Business Research 67 (1):2836-2844
International Council on Clean Transportation (2013) European Vehicle Market Statistics Pocketbook
2013.
Kempton W, Tomić J (2005) Vehicle-to-Grid Power Fundamentals: Calculating Capacity and Net
Revenue. Journal of Power Sources 144 (1):268-279
Lee JA, Soutar G, Louviere J (2008) The Best–Worst Scaling Approach: An Alternative to Schwartz's
Values Survey. Journal of Personality Assessment 90 (4):335-347
Lieven T, Mühlmeier S, Henkel S, Waller JF (2011) Who Will Buy Electric Cars? An Empirical Study
in Germany. Transportation Research Part D: Transport and Environment 16 (3):236-243
Louviere J, Lings I, Islam T, Gudergan S, Flynn T (2013) An Introduction to the Application of (Case 1)
Best–Worst Scaling in Marketing Research. International Journal of Research in Marketing 30
(3):292-303
Louviere JJ, Flynn TN, Carson RT (2010) Discrete Choice Experiments Are Not Conjoint Analysis.
Journal of Choice Modelling 3 (3):57-72
Marley A, Flynn TN, Louviere J (2008) Probabilistic Models of Set-Dependent and Attribute-Level
Best–Worst Choice. Journal of Mathematical Psychology 52 (5):281-296
Mau P, Eyzaguirre J, Jaccard M, Collins-Dodd C, Tiedemann K (2008) The ‘Neighbor Effect’:
Simulating Dynamics in Consumer Preferences for New Vehicle Technologies. Ecological
Economics 68 (1):504-516
McFadden D (1974) Conditional Logit Analysis of Qualitative Choice Behavior. New York, pp
105-142
McFadden D (2001) Disaggregate Behavioral Travel Demand's RUM Side: A 30-Year Retrospective. In:
Travel Behaviour Research: The Leading Edge. Pergamon, Amsterdam, pp 17-64
Mueller Loose S, Lockshin L (2013) Testing the Robustness of Best Worst Scaling for Cross-National
Segmentation with Different Numbers of Choice Sets. Food Quality and Preference 27
(2):230-242
Olivier JGJ, Greet Janssens-Maenhout, Muntean M, Peters JAHW (2013) Trends in Global Co2
27 / 27
Emmissions: 2013 Report. PBL Netherlands Environmental Assessment Agency
Palensky P, Dietrich D (2011) Demand Side Management: Demand Response, Intelligent Energy
Systems and Smart Loads. Industrial Informatics, IEEE Transactions on 7 (3):381-388
Paromtchik I, Laugier C Automatic parallel parking and returning to traffic maneuvers. In: Video Proc.
of the IEEE Int. Conf. on Robotics and Automation, 1998. Citeseer,
Peard E (2013) Electric Cars Slow to Gain Traction in Germany. Phys.org.
http://phys.org/news/2013-05-electric-cars-gain-traction-germany.html. Accessed July 2014
Potoglou D, Kanaroglou PS (2007) Household Demand and Willingness to Pay for Clean Vehicles.
Transportation Research Part D: Transport and Environment 12 (4):264-274
Rao VR (2014) Applied Conjoint Analysis. Springer, New York, pp 56
Schlereth C, Eckert C, Schaaf R, Skiera B (2014) Measurement of Preferences with Self-explicated
Approaches: A Classification and Merge of Trade-off-and Non-trade-off-based Evaluation
Types. European Journal of Operational Research 238 (1):185-198
Schlereth C, Skiera B (2012) DISE: Dynamic Intelligent Survey Engine. In: Quantitative Marketing
and Marketing Management. Springer, pp 225-243
Schlereth C, Skiera B, Wolk A (2011) Measuring Consumers’ Preferences for Metered Pricing of
Services. Journal of Service Research 11 (4):443-459
Shepherd S, Bonsall P, Harrison G (2012) Factors Affecting Future Demand for Electric Vehicles: A
Model Based Study. Transport Policy 20 (1):62-74
Street DJ, Burgess L (2007) The Construction of Optimal Stated Choice Experiments: Theory and
Methods. John Wiley and Sons, New Jersey, pp 647
Swait J, Andrews RL (2003) Enriching Scanner Panel Models with Choice Experiments. Marketing
Science 22 (4):442-460
Thurstone LL (1927) A Law of Comparative Judgment. Psychological Review 34 (4):273
Tie SF, Tan CW (2013) A Review of Energy Sources and Energy Management System in Electric
Vehicles. Renewable and Sustainable Energy Reviews 20 (1):82-102
Train K (2009) Discrete Choice Methods with Simulation. Cambridge University Press, New York, pp
34
Wesseling JH, Niesten EMMI, Faber J, Hekkert MP (2013) Business Strategies of Incumbents in the
Market for Electric Vehicles: Opportunities and Incentives for Sustainable Innovation.
doi:10.1002/bse.1834
