Alfons Prießner MSc. Exploring predictors of electric vehicle adoption and preferences for electric vehicle product bundles DISSERTATION submitted in fulfilment of the requirements for the degree of Doctorate in Social Sciences and Economics Alpen-Adria-Universität Klagenfurt Faculty of Management and Economics Supervisor Univ.-Prof. Dr. Nina Hampl Alpen-Adria-Universität Klagenfurt Department of Operations, Energy, and Environmental Management First Evaluator Univ.-Prof. Dr. Nina Hampl Alpen-Adria-Universität Klagenfurt Department of Operations, Energy, and Environmental Management Second Evaluator Univ.-Prof. Dr. Rolf Wüstenhagen University of St. Gallen Institute for Economy and the Environment Klagenfurt am Wörthersee, January 2019
161
Embed
Exploring predictors of electric vehicle adoption and ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Alfons Prießner MSc.
Exploring predictors of electric vehicle adoption and
preferences for electric vehicle product bundles
DISSERTATION
submitted in fulfilment of the requirements for the degree of
Doctorate in Social Sciences and Economics
Alpen-Adria-Universität Klagenfurt
Faculty of Management and Economics
Supervisor
Univ.-Prof. Dr. Nina Hampl
Alpen-Adria-Universität Klagenfurt
Department of Operations, Energy, and Environmental Management
First Evaluator
Univ.-Prof. Dr. Nina Hampl
Alpen-Adria-Universität Klagenfurt
Department of Operations, Energy, and Environmental Management
Second Evaluator
Univ.-Prof. Dr. Rolf Wüstenhagen
University of St. Gallen
Institute for Economy and the Environment
Klagenfurt am Wörthersee, January 2019
Affidavit I
AFFIDAVIT
I hereby declare in lieu of an oath that
• the submitted academic paper is entirely my own work and that no auxiliary materials
have been used other than those indicated,
• I have fully disclosed all assistance received from third parties during the process of
writing the thesis, including any significant advice from supervisors,
• any contents taken from the works of third parties or my own works that have been
included either literally or in spirit have been appropriately marked and the respective
source of the information has been clearly identified with precise bibliographical
references (e.g. in footnotes),
• to date, I have not submitted this paper to an examining authority either in Austria or
abroad and that
• when passing on copies of the academic thesis (e.g. in bound, printed or digital form),
I will ensure that each copy is fully consistent with the submitted digital version.
I understand that the digital version of the academic thesis submitted will be used for the
purpose of conducting a plagiarism assessment.
I am aware that a declaration contrary to the facts will have legal consequences.
Eidesstattliche Erklärung
Ich versichere an Eides statt, dass ich
• die eingereichte wissenschaftliche Arbeit selbstständig verfasst und keine anderen als
die angegebenen Hilfsmittel benutzt habe,
• die während des Arbeitsvorganges von dritter Seite erfahrene Unterstützung,
politische Förderanreize; Photovoltaik; Batteriespeicher; Produktbündelung;
Conjointanalyse; Clusteranalyse; Konsumentenpräferenzen; Zahlungsbereitschaft; Österreich
Introductory Chapter 1
1. INTRODUCTORY CHAPTER1
“When Henry Ford made cheap, reliable cars, people said, 'Nah, what's wrong with a horse?'
That was a huge bet he made.“
Elon Musk, CEO Tesla Motors
1.1 Background and problem statement
There is almost unequivocal agreement among scientists that the climate of our earth is
changing and that mankind plays a critical role in this process (Anderegg et al., 2010). The
latest report from the Intergovernmental Panel on Climate Change (IPCC) (2014) concludes,
clearer than ever, that climate change is caused by the release of greenhouse gas (GHG)
emissions triggered by activities of the human population. Particularly the burning of fossil
fuels causes the release of CO2 emissions, the major GHG, into the atmosphere. Almost 80%
of the GHG emissions in the earth’s atmosphere have occurred since the 1970s, mainly driven
by population and economic growth (Archer and Rahmstorf, 2010). This increase in GHG
emissions is the most likely reason for the rise in temperature wich is causally associated with
several challenges such as droughts, higher sea-levels, etc. (Dow and Downing, 2011).
The IPCC argues that humanity is at risk if global warming were to continue at the
current rate, and calls for counter measures (IPCC, 2014). Former US president Obama made
the urgency of this topic very clear in his State of the Union Address in 2015, saying: “No
challenge poses a greater threat to future generations than climate change” (CNN, 2015).
Hence, in 2015 the members of the United Nations Framework Convention on Climate Change
(UNFCCC) set an ambitious target (also referred to as the Paris Agreement) to combat climate
change by limiting the global average temperature increase to well below the most noted 2
degrees Celsius target. They agreed to implement measures and policies in various GHG
producing industries to prevent temperature from increasing more than 1.5 degrees Celsius
above pre-industrial levels (UNFCCC, 2015). The International Energy Agency (IEA) claims
that to achieve this, an even more ambitious target for every short-term reduction of CO2
emissions is required, which would mean that we should employ “every known technological,
societal and regulatory decarbonization option” (IEA, 2016b: 5).
1 Please note that if specific notes are made to the papers, they are referred to as Paper 1 (Priessner, Sposato and
Hampl, 2018), Paper 2 (Priessner and Hampl, 2018a) and Paper 3 (Priessner and Hampl, 2018b); several sentences
of this chapter are drawn from Paper 1, Paper 2, and Paper 3 without explicit citation.
Introductory Chapter 2
One promising pathway to achieve this UNFCCC goal is to decarbonize the transport
sector which is responsible for almost 25 percent of global energy-related GHG emissions
causing global warming (IEA, 2016a). To reduce the carbon footprint of transportation climate
experts and politicians pursue a variety of possible actions, such as incremental technology
improvements for internal combustion engines (ICEs), hybridization, biofuels, shifting to mass
transportation systems (e.g., urban rail, bus systems, etc.), reducing trip distances, and vehicle
sharing concepts (IEA, 2016a, 2009). Currently they pin their expectations largely on replacing
the predominant use of conventional ICEs dependent on fossil fuels with electric ones (IEA,
2016a; IPCC, 2014; UNFCCC, 2015).
These electric vehicles (EVs)2 could have a zero-emission level during usage if coupled
with decarbonized energy production (e.g., Bleijenberg and Egenhofer, 2013; Granovskii et al.,
2006; IEA, 2016a; Sims et al., 2014). Hence, various studies emphasize the need for a major
uptake of EVs within the next three decades to reduce the transportation sector’s impact on
climate change, and to meet GHG reduction targets (EEA, 2018; IEA, 2016a, 2009; IPCC,
2014; Sims et al., 2014; UNFCCC, 2015). For instance, IEA modelling forecasts that by 2030
we would need 35% of the global vehicle sales to be electric (comprising battery electric, plug-
in hybrid, and fuel cell vehicles ranging from two-wheelers to light-commercial vans or trucks)3
to achieve the UNFCCC target. Translated, this means that 150 million EVs fueled by
renewable energy should be in the market by 2030, which in the interim requires a sizable
growth of EV stock and a major deployment of EVs during the 2020s (IEA, 2016a).
Consequently, over the past decade an interest in EVs has resurfaced among
practitioners, consumers, policy makers, and also researchers. Players in the automotive and
the battery industry have ongoingly been working on new and improved technologies (e.g., in
the field of battery storage) to overcome some of the motives for not purchasing an EV (e.g.,
range anxiety, purchase price, etc.) (Wesseling et al., 2015). Consumer preferences are
changing toward more sustainable mobility solutions which also include EVs with more
efficient and zero-carbon emission electric engines (Thiel et al., 2014). Policy makers and
governments around the world (Sierzchula et al., 2014), ranging from China (Zhang et al., 2013)
to the USA (Diamond, 2009), and to members of the European Union (Kley et al., 2011; Lieven,
2015; Mannberg et al., 2014; Sierzchula et al., 2014) have introduced policies and subsidies
2 In this dissertation, the author will focus on EVs which are battery electric (BEV) or plug-in hybrid electric
(PHEV) vehicles, which can be charged from an external source of electricity. Hybrid electric vehicles (HEVs)
are thus excluded from this definition. Further, fuel cell electric vehicles powered by hydrogen are not in the scope
of this definition. 3 This dissertation only focuses on EVs in the light-duty segment used for regular passenger transport (IEA, 2018).
Introductory Chapter 3
that promote electromobility. One regularly quoted example of successful policy support in EV
sales growth, comes from Norway (Langbroek et al., 2016), which is the third largest market
for electrically chargeable vehicles, after the US and China. Further, Norway is the undisputed
leader in new registration of EVs, meaning every third newly bought car in Norway is an EV
(Electric Vehicle World Sales Database, 2017). Moreover, research has been dealing with EVs
from various angles, e.g., in economical, engineering, or transportation literature (Granovskii
et al., 2006; Holland et al., 2015; Newman et al., 2014; Niesten and Alkemade, 2016; Rezvani
et al., 2015; Steinhilber et al., 2013; Zivin et al., 2012). Therefore, the transition to
electromobility has gradually been gathering pace, so that by 2016 EV sales had passed the one
million unit target (IEA, 2018).
However, despite all these efforts, sales growth across the entire EV market (besides
Norway and China) has fallen short of the growth rates required to achieve the targets of the
Paris Agreement (IEA, 2018, 2016a). Possible reasons are manifold, ranging from customer
concerns (e.g., Egbue and Long, 2012), technical problems (e.g., Lu et al., 2013), and economic
challenges (e.g., Dimitropoulos et al., 2013), to lock-in effects (e.g., Steinhilber et al., 2013) or
status-quo bias (Newman et al., 2014). Particularly, according to the literature, user concerns
about driving range, purchase price, and charging time are the most citied barriers to adoption
since the early 1980s (e.g., Beggs and Cardell, 1980; Bunch et al., 1993; Egbue and Long, 2012;
Schuitema et al., 2013). Additionally, consumers perceive EVs as cars of the future (Burgess
et al., 2013) or a work in progress (Graham-Rowe et al., 2012) due to technological and
infrastructural developments being unpredictable, which again has a negative impact on the
intention to purchase an EV.
On top of that, EVs’ effectiveness in combatting climate change has been also disputed
in the literature (Ellingsen et al., 2016; Sandy Thomas, 2012; Zivin et al., 2012). Experts argue
that, in production, depending on the size of the battery, power mix in production, location of
production, etc., EV’s CO2 footprint seems to be not superior or even worse than that of an ICE
(Ellingsen et al., 2016). Further, the energy requirements for raw material extraction and
processing are significantly higher than for ICEs (EEA, 2018). Several environmentalists also
argue that despite ICEs being replaced by EVs, certain regions hardly lower their GHG
emission at all, due to their current power supply mainly being produced by CO2-emissioning
resources (Holland et al., 2015; Zivin et al., 2012).
Nevertheless, there seems to be considerable agreement that over the entire life-cycle of
an EV (from raw material extraction to vehicle recycling) the CO2 footprint is below an ICE’s
when it is fueled by power from renewable energy sources (e.g., Bleijenberg and Egenhofer,
Introductory Chapter 4
2013; EEA, 2018; Ellingsen et al., 2016; Granovskii et al., 2006; IEA, 2016a; Sims et al., 2014),
because the “largest potential reduction in GHG emissions between a BEV and an ICE occurs
in the in-use phase, which can more than offset the higher impact of the raw materials extraction
and production phases” (EEA, 2018: 7). Hawkins et al. (2012) already projected this in arguing
that, assuming the current electricity mix in Europe, across its life-cycle an EV produces 17-
21% less GHG emissions than comparable diesel fueled vehicles, and 26-30% less than petrol
vehicles 4 . The benefits of moving to EVs could be even larger if we could increase the
development of renewable energy and the circular economy, including vehicle sharing and
product design, reuse and recycling (EEA, 2018).
To summarize, the question of how to increase consumer adoption for (greener5) EVs
is subject to lively discussion in various fields (e.g., academic research, industry, politics), but
has still not been exhaustively answered. This doctoral thesis aims to contribute some novel
insights to this discourse by focusing on two relevant sub-research fields. First, scholars argue
that it is very important to understand predictors of EV adoption to increase the possible role of
EVs in the global transportation system (Axsen et al., 2016). Second, to benefit from EVs in
countering climate change, research often suggests the bundling of EVs with, e.g., photovoltaics
(PV) and battery storage (BS) (Delmas et al., 2017), but very little has been researched from a
consumer preference perspective (Cherubini et al., 2015). Therefore, this thesis aims to
contribute to the following two overall research questions:
A) What are the major predictors of an EV adoption?
B) What are the consumer preferences for EV-PV-BS product bundles?
1.2 Theoretical background and research questions
There is quite an extensive literature on the adoption of EVs, and across the past decade
research interest has been focused on various topics (e.g., consumer preference, EV adoption
barriers, EV policy effectiveness, characteristics of early EV users, environmental impact of
EVs) to promote its acceptance and adoption rate (cf., Liao et al., 2017; Rezvani et al., 2015).
The author of this doctoral thesis focuses on two aspects in the field which appeared to require
additional research, namely: (A) a better understanding of the predictors of EV adoption (Axsen
4 A Nissan LEAF was compared to an ICE (Mercedes A 170 (petrol) and Mercedes A 160 (diesel)), assuming a
total lifetime mileage of 150,000 km. All vehicles are comparable in size, mass, and performance characteristics. 5 In this thesis the term greener will be used in referring to a more sustainable and environmentally friendly usage
of an EV (i.e. zero-emission) which can be ensured when EV’s power is supplied purely by renewable energy
sources.
Introductory Chapter 5
et al., 2016) and (B) the preferences for EV add-on products such PV and BS when they are
offered in the form of EV product6 bundles (Cherubini et al., 2015). In this chapter some
theoretical background on the need for research in these two fields, and derived more specific
research questions, will be provided.
(A) Predictors of EV adoption
Several recent research papers argue that a stronger focus on early-adopter customer
segments in product development and EV policy implementation is needed to gain a mass
market movement toward purchasing EVs (Green et al., 2014; Larson et al., 2014; Wesche et
al., 2016). Therefore, stakeholders and scientists have a pronounced desire to learn more about
early7 and potential8 EV adopters and the related EV adoption predictors. To date, predictors of
EV adoption/non-adoption have been researched widely, from North America (Axsen et al.,
2016) to Europe, which includes Norway (Nayum et al., 2016) and Germany (Plötz et al., 2014).
Initial insights of these studies suggest that socio-demographic and psychological
factors, but also policy incentives, significantly differentiate between EV owners, potential EV
adopters, and non-adopters not willing to purchase an EV in the near future. These findings are
in line with Stern (2000) who argues that personal characteristics (e.g., socio-economic),
attitudinal factors (e.g., several psychological parameters), and contextual forces (e.g., policy
incentives or regulations) trigger pro-environmental behavior and eventually could facilitate the
adoption of high-priced products such as greener cars. Nayum and Klöckner (2014)
interestingly added to this topic that the predictive power of socio-demographics for purchase
intention of environmentally friendly vehicles (such as EVs) might be lower when
psychological variables are included in the analysis of individual-related predictors. Based on
the relevant literature, this dissertation will research socio-demographic, psychological, and EV
policy incentives as predictors of EV adoption to gain a comprehensive impression of the
Austrian market.
In addition to assessing the impact of EV predictors on the Austrian market, this
dissertation aims to address two research gaps. First, existing literature has largely neglected to
6 In general, the term “product” refers to both goods and services (Stremersch and Tellis, 2002). In this dissertation
the author focuses on the products PV and BS, which are offered as EV add-on products and are referred to as EV-
PV-BS product bundle. 7 Early adopters are defined by the author as the people who have a positive attitude towards EVs and/or are
willing to purchase an EV as their next car or already own an EV. This category corresponds to the Innovators and
Early-Adopters segments identified by Rogers (2003) (approx. 15% of population). 8 Potential adopters are defined by the author as the people who generally have a positive attitude toward EVs
and/or are willing to purchase an EV, but not as their next car. This category corresponds to the Early Majority
segment identified by Rogers (2003) (approx. 34% of population).
Introductory Chapter 6
consider the influence the psychological variable cultural worldview9 has on the propensity to
purchase an EV. Previous research has shown that cultural worldviews could be predictive of
people’s attitudes toward climate change (Kahan et al., 2007), or of their acceptance of climate
change mitigation policies (Hart and Nisbet, 2011) or clean technologies (Cherry et al., 2014;
Sposato and Hampl, 2018).
Second, there is a broad discussion ongoing in the literature and among stakeholders,
on the effectiveness of policy measures to increase the EV adoption. On the one hand, scholars
argue that policy incentives increase market penetration of EVs (e.g., Langbroek et al., 2016;
Mannberg et al., 2014; Sierzchula et al., 2014). On the other hand, practitioners as well as
experts criticize the use of purely financially related incentives as not being effective in
convincing non-adopters (e.g., Egbue and Long, 2012; Spiegel, 2017).
Therefore, the derived specific research questions which this doctoral thesis aims to
answer are:
• What are the major predictors (socio-economic and psychological) of EV adoption?
• How do the potential adopter’s cultural worldviews (individualism and egalitarianism)
impact the adoption of EVs?
• How do EV related policy incentives affect the adoption of EVs?
(B) Consumer preferences for EV-PV-BS product bundles
Bundling products with additional products or services is a well-known and well-
researched tool for increasing consumers’ acceptance and willingness to pay (WTP) (e.g.,
Eppen et al., 1991; Stremersch and Tellis, 2002). It has been shown that bundling positively
influences the launch of new products (Simonin and Ruth, 1995), improves consumers’
valuation of innovative products, and increases the consumer’s purchase intention regarding
these products (Reinders et al., 2010).
In the automotive industry, add-on services or products have a long history and are
considered crucial for customer satisfaction and acceptance. They range from car maintenance
and repair services, financing, leasing, insurance, banking, and many different entertainment
services offered to ensure exclusive customer experience (e.g., Godlevskaja et al., 2011). Add-
on services or products have become not only key to customer satisfaction, but also to
promoting new products in the vehicles market (Fassnacht et al., 2011). Therefore, scholars
argue that the add-on offerings need to be consciously reviewed and adopted according to
9 Cultural worldview is defined as “a general perspective from which a person sees and interprets the world”
(Cherry et al., 2014: 563).
Introductory Chapter 7
changing customer needs or new technologies (Stauss, 2009). Cherubini et al. (2015) even
claimed that add-on components are relevant factors in purchasing decisions for EVs and
proposed product bundles or product-service bundles, as a key success factor to create a critical
mass of adopters. They specifically called for more research in this field.
However, to date, only the preferences for single add-on services (such as intelligent
parking or navigation packages) and not for EV product bundles (such as EV-PV or EV-PV-
BS) have been research in clean technology. Prior research focused on add-on services such as
navigation packages for charging services or intelligent parking and paying, for mobility
guarantees, or battery leasing (Fojcik and Proff, 2014; Hinz et al., 2015). Such research
evaluated the preferences and effects of electricity charging stations being available (e.g.,
Brownstone et al., 2000), or of the vehicle-to-grid (V2G) services on the acceptance of EVs
(Parsons et al., 2014). In contrast, EV add-on products such as PV and BS (in form of an EV-
PV-BS product bundle) have not been researched from a consumer preference perspective,
despite the high interest of automotive and electric utility players, as well as the ability to reduce
the EV’s CO2 emission (Delmas et al., 2017). Moreover, since consumer preferences are quite
diverse in the field of clean technologies (cf. Kaufmann et al., 2013; Salm et al., 2016; Tabi et
al., 2014), the identification of potential adopter segments and the description of their
preferences is an additional research gap to which this thesis tries to contribute.
Further, from a WTP perspective about EV product bundles, there is scant literature.
There is a broad literature stream on the WTP for EV attribute improvements (e.g., Ewing and
Sarigöllü, 2000; Hackbarth and Madlener, 2016; Tanaka et al., 2014). However, only a few
studies analyze the WTP for an EV if offered in a bundle with add-on services or warranties
(cf., Ensslen et al., 2018; Fojcik and Proff, 2014). For EV product bundles (e.g., PV-BS add-on
products), there are, to the author’s best knowledge, no extant WTP studies on the attributes of
such possible bundles.
Another angle which requires research, is the influence of individual-related
characteristics on the evaluation of EV product bundle attributes and the related WTP. Liao et
al. (2017) quite recently argued in their comprehensive literature review on EV consumer
preference studies that besides EV related attributes, individual-related characteristics, such as
socio-demographic or psychological factors, are major parameters influencing or moderating
EV utility. A literature review has identified a lack of research on the influence of (potential)
EV drivers’ characteristics on EV-PV-BS product bundle preferences and on the WTP (cf., Liao
et al. 2017 or Li et al., 2017 for details).
Introductory Chapter 8
Overall, this dissertation aims to close the above-mentioned gaps by trying to answer
the following research questions:
• What are the preferences for the EV-PV-BS product bundles and how do these
preferences differ across various adopter segments?
• What is the WTP for attributes of EV-PV-BS product bundles?
• How do individual-related attributes influence the preferences and WTP for EV-PV-BS
product bundles?
1.3 Overview research papers and objectives
The objective of this doctoral thesis is to shed some light on the research gaps and
questions described above, by means of presenting three research papers. The research
objectives and publication status of each of these three papers are described below in more
detail.
Paper 1 of this thesis is titled “Predictors of electric vehicle adoption: an analysis of
potential electric vehicle drivers in Austria.” Considering the different effects of socio-
demographic and psychological variables, it investigates distinctions between early EV
adopters, potential EV adopters, and non-adopters regarding EV purchases or EV purchase
intention. In doing so, this paper is the first to test the predictive effect of cultural worldviews
(Cherry et al., 2014; Kahan et al., 2007) on the adoption/non-adoption of EVs. Further, Paper
1 contributes to the discussion of policy incentives by evaluating whether EV-related policy
incentives have a positive effect on EV adoption (cf. Egbue and Long, 2012). Based on an
Austrian wide representative sample of 1,000 respondents, this research paper discusses
interesting novel insights on the main predictors, and the role of cultural worldviews and policy
incentives in the acceptance of EVs. Additionally, for practitioners, the paper identifies and
characterizes various potential adopter sub-segments, shedding some light on the topic of EV
adopter heterogeneity and how the preferences for policy incentives differ among potential EV
adopter sub-segments (cf. Langbroek et al. 2016). Paper 1 is co-authored by Robert Sposato
and Nina Hampl10 and published in Energy Policy, Volume 122 in 2018 (5-Year Impact Factor:
5.038/VHB: B). Furthermore, earlier versions of this paper have been presented in 2017 at the
10. Internationale Energiewirtschaftstagung (IEWT) in Vienna/Austria, 5. Energiekonzept
Kongress in St.Gallen/Switzerland and at the Electric Vehicle Symposium (EVS) 30 in
Stuttgart/Germany.
10 The data analysis, as well the writing of the paper, was almost entirely done by the lead author (Alfons Priessner).
Introductory Chapter 9
Paper 2, titled “Can product bundling increase the joint adoption of electric vehicles,
solar panels and battery storages? Explorative evidence from a choice-based conjoint study in
Austria,” takes a closer look at product bundling, in particular the bundling of EVs with
renewable power supply systems (PV with/without BS) in Austria. Since this type of EV-PV-
BS product bundle is not offered on the market yet, this study is, to our knowledge, the first to
analyze the preferences and the market potential of such EV-PV-BS product bundles. Based on
a conjoint experiment with 39311 potential EV drivers in Austria, the paper took a closer look
at different product bundle attributes and the preferences of potential EV drivers. Further, this
paper aims to provide a more finely grained view on consumer preferences, and hence,
identifies sub-segments of potential adopters, characterizing them via socio-demographic and
psychological factors, and identifying various product preferences across the segments by
running market simulations. Another objective of this paper, particularly relevant for
practitioners, is to evaluate the market potential of EV-PV-BS product bundles, as well as to
describe the relation between bundling complementary clean technologies and the increase in
intended adoption rate of such technologies. This paper is co-authored by Nina Hampl12 and
has received an invitation for “revise and resubmit” from Ecological Economics (5-Year Impact
Factor: 4.803/VHB: B). Moreover, this paper was presented at the 1. Advanced Demand
Modelling Workshop for Electromobility and at the 6th International PhD Day of the Austrian
Association of Energy Economics (AAEE) in 2018 in Vienna/Austria. In May 2019 this paper
will be also presented at the largest EV conference worldwide, the EVS 32 in Lyon/France.
Paper 3 also deals with EV-PV-BS product bundles, specifically from a willingness to
pay (WTP) angle. Hence, this paper is titled “Exploring consumer heterogeneity in willingness
to pay for electric vehicles product bundles.” The paper again used data from the Austrian-
wide conjoint experiment with 61611 potential EV adopters to arrive at the following research
objectives. First, the paper aims to estimate the WTP for the specific product bundle attributes
and related policy incentives. Second, the paper has the objective of evaluating the influence of
socio-demographic and psychological parameters, as well as by EV experience on potential
adopters’ preferences and their WTP for EV-PV-BS product bundles. This paper is co-authored
by Nina Hampl13. It has been submitted to Transportation Research Part A (5-Year Impact
11 Details about filtering the sample for the conjoint experiment (from the total sample of N=1,251) will be
provided in the next sub-section (Methodology and data), and in the dedicated sections in Paper 2 and Paper 3. 12 The data gathering and analysis, as well as the writing of the paper, was almost entirely done by the lead author
(Alfons Priessner). 13 The data gathering and analysis, as well as the writing of the paper, was almost entirely done by the lead author
(Alfons Priessner).
Introductory Chapter 10
Factor: 3.809/VHB: B) and is currently under review. Further, this paper has been accepted for
the 11. IEWT conference, which takes place in February 2019 in Vienna/Austria.
Table 1 provides an overview of all three papers. It summarizes the title, authorship,
research objectives, as well as the theoretical foundations (see details above Theoretical
background and research questions), methodologies applied and sample size (see details below
in Methodology and data). Also, it comprises information on the current publication status of
each paper.
TABLE 1. OVERVIEW OF RESEARCH PAPERS THAT MAKE UP THIS DISSERTATION
No. Author (s) Title Research Objectives Theoretical
Background Methodology Sample Publication Status
1 Priessner,
Alfons; a,c
Sposato,
Robert; a
Hampl, Nina
a,b
Predictors of electric
vehicle adoption: an
analysis of potential
electric vehicle
drivers in Austria
1) Predictors of EV adoption (incl.
cultural worldviews)
2) Effectiveness of EV policy
incentives on EV adoption
3) Heterogeneity of potential EV
adopters in characteristics and
EV policy preferences
EV adoption
literature,
cultural
worldview
literature, EV
policy literature
Multinomial logistic
regression, cluster
analysis (hierarchical
and k-means
clustering), factor
analysis, ANOVA
and chi-square tests
N=1,000 Published in Energy
Policy Volume 122,
2018: 701-714
5-Year Impact Factor:
5.038/VHB: B
2 Priessner,
Alfons; a,d
Hampl Nina
a, b
Can product bundling
increase the joint
adoption of electric
vehicles, solar panels
and battery storages?
Explorative evidence
from a choice-based
conjoint study in
Austria
1) Preferences for bundling EV
with add-on products (PV
with/without BS) and market
potential
2) Consumer heterogeneity for EV-
PV-BS product bundling
Bundling
literature, EV
and clean
technology
consumer
preferences
literature
Conjoint analysis
(choice-based
conjoint (CBC)
approach), latent
class cluster analysis,
share of
preference/market
simulations
N=393 Reviewed with
outcome Revise and
Resubmit by Ecological
Economics, 28.12.2018
5-Year Impact Factor:
4.803/VHB: B
3 Priessner,
Alfons; a,d
Hampl Nina
a,b
Exploring consumer
heterogeneity in
willingness to pay for
electric vehicles’
product bundles
1) WTP for EV-PV-BS product
bundle attributes
2) Influence of socio-demographic
and psychological variables on
the WTP/preferences for EV-PV-
BS bundles
EV consumer
preferences and
WTP literature,
EV adoption
literature
Conjoint analysis
(CBC approach),
WTP calculations,
HB model with
covariates
N=616 Under review at
Transportation
Research Part A,
15.11.2018
5-Year Impact Factor:
3.809/VHB: B
a Alpen-Adria-Universität Klagenfurt, Department of Operations, Energy, and Environmental Management b Vienna University of Economics and Business, Institute for Strategic Management c The first author is main author of this research paper; i.e., the first author developed the research questions, analyzed the data and wrote nearly the entire manuscript on his own. The
co-authors collected the data in their study “Erneuerbare Energien in Österreich 2016” (cf. Hampl and Sposato, 2017), provided coaching in developing the research questions and
analyzing the data as well as provided feedback during the development of the manuscript. d The first author is main author of this research paper; i.e., the first author developed the research questions, collected and analyzed the data, and wrote nearly the entire manuscript on
his own. The co-author provided coaching in developing the research questions and analyzing the data and provided feedback on each section and input for the conclusion section of
each paper.
Introductory Chapter 12
1.4 Methodology and data
All three papers of this doctoral thesis are reports on empirical studies that employed
quantitative research methods14. The target market of this dissertation is Austria, which had the
highest growth rate (128%) and the highest share of new registrations (1.2%) of EVs in the
European Union in 201615 (Electric Vehicle World Sales Database, 2017). Further, Hampl and
Sposato (2017) have shown that in Austria almost every second car driver could imagine
purchasing an EV. Hence, the Austrian population was considered as fairly suitable for analyzing
(A) predictors for EV adoption and (B) EV-PV-BS product bundle consumer preferences. Such a
population would be suitable for providing novel insights for countries at the beginning of EV
diffusion. Depending on the respective research questions and research context, different
methodologies and datasets were used, which are briefly described below, and given in more detail
in each paper.
In Paper 1 an Austrian-wide survey16 was conducted in 2016 to measure car drivers’
attitudes toward EVs and related policy incentives, and on their willingness to purchase. Further,
data on socio-demographic and psychological scales was collected (for further data details see
Paper 1). A market research company (meinungsraum.at) recruited a national-representative
sample of 1,000 respondents. Deloitte Austria and Wien Energie supported the survey with
expertise and funding. Other results of this survey have been published by Hampl and Sposato
(2017) in Erneuerbare Energien in Österreich 2016 (i.e., Renewable Energies in Austria 2016) and
in Sposato and Hampl (2018).
For analyzing the predictors of EV adoption, the author applied a multinomial logistic
regression (Backhaus et al., 2016). In addition, the author took a two-step approach to cluster
respondents into groups with minimized intra-cluster variance: hierarchical cluster analysis was
used to determine the optimal number of clusters, which served as input variables for the k-means
cluster analysis (Punj and Stewart, 1983). Finally, by applying ANOVA and chi-squared tests
14 Preparing for the quantitative research methods, the author conducted qualitative interviews (>10 interviews) with
experts or lead users (for more details see the Methods section in each paper) for each study. 15 The research proposal for this dissertation was accepted in June 2017. Hence the sales figures of 2016 formed the
most recent reference point for the decision regarding which country to focus on. 16 This survey is an annual survey covering several clean technology acceptance topics, i.e., public perceptions,
preferences, and willingness to invest in renewable energy and other low-carbon technologies across the Austrian
population. A sub-section of the survey deals with EV-related questions and topics, which were used for Paper 1.
Introductory Chapter 13
differences between the segments were identified along a set of socio-demographic and
psychological parameters, and the author compared respondents’ preferences for policy incentives.
Paper 2 and Paper 3 are both based on the same survey, including a choice-based conjoint
(CBC) experiment. A market research company called market recruited a representative sample of
1,251 respondents from Austria. The participants, who successfully and properly filled out the
survey, were financially rewarded, which is considered common practice in the market research
industry (cf. Gamel et al., 2016; Salm et al., 2016). Flatliners or speeders were reported back to the
market research company, and did not receive any reward. An additional benefit using participants
from a panel pool is their experience with longer surveys and with choice experiments which could
temper Jaeger et al.’s (2001) criticism that CBC gets more accurate results, if participants are
already used to CBCs due to certain training effects (for details see the respective sections in Paper
2 and Paper 3).
The final samples of the two papers differ due to different filtering questions. For Paper 2,
393 17 respondents passed the filtering questions for the final sample. For Paper 3, 616 18
respondents made up the sample. Both samples comprised future EV drivers with varying purchase
horizons aligned to different research objectives in each paper (for details see the respective
sections in Paper 2 and Paper 3). Sawtooth Software Lighthouse, the standard application for CBC
experiments in marketing research, was used to design, administer, conduct and analyze the survey.
Conjoint analysis is a frequently used market-research method, which aims at explaining
purchasing behavior even if the analyzed product does not exist on the market yet (e.g., Louviere
et al., 2008). In market research, conjoint studies are considered superior to directly asking
consumers about their decision criteria, because people have little insight into their decision-
making rationale, and they often have a recall bias or other information recovery failures (Golden,
1992). Even so, a number of scholars have discussed the methodological challenges of conjoint
analysis (e.g., Ben-Akiva et al., 1994; Louviere et al., 2000), and this has led to ongoing
advancement (e.g., Orme and Chrzan, 2017; Train, 2003).
Due to its benefits in studying decision-making, also regarding products which do not exist
yet, conjoint analysis has already successfully been applied to the context of EVs for decades (e.g.,
17 Filter criteria: Potential EV driver with a positive attitude toward EVs, willingness to purchase an EV as their next
car within the next 5 years. 18 Filter criteria: Potential EV driver with a positive attitude toward EVs, willingness to purchase an EV within the
next 10 years.
Introductory Chapter 14
Beggs and Cardell, 1980; Brownstone et al., 2000; Bunch et al., 1993; Ewing and Sarigöllü, 2000;
Hackbarth and Madlener, 2016; Hackbarth and Madlener, 2013; Hidrue et al., 2011; Hoen and
Koetse, 2014) and also for EV add-on studies (e.g., Hinz et al., 2015; Parsons et al., 2014). Table
2 provides an overview of the attributes of these studies and the methodology the author used in
each. Considering these earlier studies, the author also considers conjoint experiments as well
suited to investigate EV-PV-BS product bundle attributes in relation to each other from the
perspective of early/potential adopters.
Conjoint analysis uses an indirect questioning approach which divides the decision-making
process and products into underlying response preferences for particular attributes (referred to as a
decompositional approach) (Green and Srinivasan, 1990). This means that in a conjoint setting, a
respondent in the survey needs to make a decision (dependent variable) within several choice tasks
(i.e., between hypothetical but potential products, in this study EV-PV-BS product bundles). The
presented options vary along pre-defined attributes in their specific attribute levels (independent
variables) (e.g., Gustafsson et al., 2013; Louviere et al., 2000).
From the decisions made in the choice tasks, preferences for the attribute levels can be
derived in the form of average part-worth utilities and relative importance weights for each of the
attributes (e.g., Green and Srinivasan, 1990). Further, the part-worth data can be used to run latent
class cluster analysis (Sawtooth Software, 2004), share of preference market simulations with
varying attribute settings (Orme and Chrzan, 2017), WTP calculations (Orme, 2001) and
estimations on the covariates’ influence on the utility model (Orme and Howell, 2009). All these
methods have been applied in either Paper 2 and/or Paper 3.
Choosing the most relevant attributes and levels is a very critical task for the success of a
conjoint analysis study (Bergmann et al., 2006). Therefore, the author selected an elaborated
iterative process comprising a literature review, sales conversations with EV, PV, and BS retailers,
interviews with EV, PV, and BS lead users, and expert interviews. The survey development was
supported by the Austrian utility company KELAG with expertise and funding. Before going live
with the survey, the author conducted a pre-test with 45 EV supporters to verify the interpretation
of the attributes and test the relations among them. For further details about this applied iterative
process, please see details in the respective chapters in Paper 2 and Paper 3.
15
TABLE 2. OVERVIEW CONJOINT STUDIES ON ELECTRIC VEHICLES19
Source Attributes Number of choice tasks,
attributes, and levels
Respon-
dents
Analysis model
Hackbath and
Madlener
(2013)/(2016)
Purchase price, fuel type, fuel cost per 100 km, CO2 emission reduction, driving range,
fuel availability, refuel/recharging time, policy incentive
15 choice tasks / 8
attributes / 3 levels
711 Multinomial / Mixed Logit
Model (2013) / Latent Class
Model (2016)
Hinz et al.
(2015)
Range, charging time, motor power, purchase price, electricity cost per 100 km and 3
complementary services (IT-based parking space and payment, intelligent charging
station, augmented reality services)
14 choices tasks / 8
attributes / 2-4 levels
150 Hierarchical Bayes Model
Parsons et al.
(2014)
Price relative to favorite ICE, annual cash payback (V2G), required plug-in time per
a Rural Non-Techs vs. Undecided Individualists – p < 0.05, b Rural Non-Techs vs. Undiscerning Urbanites – p < 0.05, c Rural Non-Techs vs. Urban EV Supporters
– p < 0.05, d Undecided Individualists vs. Undiscerning Urbanites – p < 0.05, e Undecided Individualists vs. Urban EV Supporters – p < 0.05, f Undiscerning Urbanites
vs. Urban EV Supporters – p < 0.05
TABLE A.5 ANOVA SOCIO-DEMOGRAPHIC AND PSYCHOLOGICAL DIFFERENCES BETWEEN POTENTIAL ADOPTER CLUSTERS
a Rural Non-Techs vs. Undecided Individualists – p < 0.05, b Rural Non-Techs vs. Undiscerning Urbanites – p < 0.05, c Rural Non-Techs vs. Urban EV Supporters –
p < 0.05, d Undecided Individualists vs. Undiscerning Urbanites – p < 0.05, e Undecided Individualists vs. Urban EV Supporters – p < 0.05, f Undiscerning Urbanites
vs. Urban EV Supporters – p < 0.05
TABLE A.6 CHI-SQUARED TEST SOCIO-DEMOGRAPHIC DIFFERENCES BETWEEN POTENTIAL ADOPTER CLUSTERS
Exemption toll payment a, b, c, d, f 3.86 (0.96) 4.48 (0.72) 3.48 (0.78) 4.31 (0.89) 19.54 0.00
Free parking a, b, c, d, f 4.01 (1.08) 4.52 (0.74) 3.52 (0.75) 4.44 (0.87) 16.87 0.00
Scrapping premium a, b, c, d, f 3.86 (1.12) 4.41 (0.82) 3.11 (0.78) 4.24 (0.95) 19.55 0.00
Tax benefits company cars a, d, f 3.59 (1.21) 4.26 (0.95) 3.27 (0.77) 3.85 (1.35) 9.27 0.00
(Partially) deductibility of purchase price
in income tax return a, b, c, d, f 4.17 (0.94) 4.63 (0.57) 3.31(0.70) 4.55 (0.75) 36.64 0.00
Exemption standard fuel consumption and
car tax a, b, c, d, f 3.59 (1.21) 4.26 (0.95) 3.37 (0.77) 3.85 (1.35) 52.88 0.00
Bus lane usage a, c, d, f 3.27 (1.22) 3.80 (1.14) 3.13 (0.77) 3.90 (1.20) 8.39 0.00
Reserved special parking lots a, b, d, f 3.86 (1.05) 4.29 (0.90) 3.31 (0.83) 4.14 (0.83) 12.02 0.00
No speed limits a, d, f 3.31(1.23) 4.10 (1.02) 3.21 (0.89) 3.73 (1.36) 10.27 0.00
Free public charging a, b, d, f 4.56 (0.71) 4.79 (0.44) 3.40 (0.85) 4.73 (0.48) 63.63 0.00
Legally prescribed number of public
charging stations a, b, d, e, f 4.06 (0.88) 4.52 (0.66) 3.38 (0.72) 4.10 (0.87) 23.76 0.00
Regulation of internal combustion engines
a, d, e, f 3.14 (1.11) 3.84 (1.14) 3.25 (0.88) 3.10 (1.19) 9.32 0.00
a Rural Non-Techs vs. Undecided Individualists – p < 0.05, b Rural Non-Techs vs. Undiscerning Urbanites – p < 0.05, c Rural Non-Techs vs. Urban EV Supporters – p < 0.05, d Undecided Individualists vs. Undiscerning Urbanites – p < 0.05, e Undecided Individualists vs. Urban EV Supporters – p < 0.05, f Undiscerning Urbanites vs. Urban EV
Supporters – p < 0.05.
TABLE A.7 ANOVA POLICY INCENTIVES SUPPORT BETWEEN POTENTIAL ADOPTER CLUSTERS
Paper 1 68
Scale/Dimension Items Source(s)
Cultural worldview:
individualism-
communitarianism
The government interferes far too much in our
everyday lives.
Kahan et al. (2007, 2011);
Cherry et al. (2014)
Free markets – not government programs – are
the best way to supply people with the things
they need.
The government should do more to advance
society’s goals, even if that means limiting the
freedom and choices of individuals. (Recoded)
Cultural worldview:
egalitarianism- hierarchy
We have gone too far in pushing equal rights in
this country. (Recoded)
Kahan et al. (2007, 2011);
Cherry et al. (2014)
Our society would be better off if the
distribution of wealth were more equal.
Discrimination against minorities is still a very
serious problem in our society.
Pro-environmental
attitude
I would say of myself that I am environmentally
conscious.
Whitmarsh and O'Neill
(2010)
Being environmentally friendly is an important
part of my personality.
I would describe myself as someone who cares
about the environment.
Pro-technological
attitude
I see the digitization as ...
... opportunity for better networking of objects
of daily life.
... possibility of networking with people
worldwide.
... essential facilitation of communication and
the handling of everyday things.
... possibility of access to fast, up-to-date, and
extensive information and knowledge.
... danger to the privacy of the individual ("glass
man"). (Recoded)
… problematic with regard to hacker attacks.
(Recoded)
... predominantly negative development
regarding the safety of people. (Recoded)
TABLE A.8. PSYCHOLOGICAL MEASUREMENT SCALES USED IN SURVEY
69
PAPER: 2: CAN PRODUCT BUNDLING INCREASE THE JOINT ADOPTION OF
ELECTRIC VEHICLES, SOLAR PANELS AND BATTERY STORAGE?
EXPLORATIVE EVIDENCE FROM A CHOICE-BASED CONJOINT STUDY IN
AUSTRIA24
Priessner, Alfons*; Hampl, Nina*,#
ABSTRACT
Although electric vehicle (EV) sales have been increasing recently, EVs can only contribute to
mitigating climate change if their required power is generated from renewable energy sources.
Hence, a product bundle of EVs with photovoltaic (PV) solar panels in combination with battery
storage (BS) for households could be instrumental in improving EV adoption rates and thus
also their carbon footprint. We conducted a choice-based conjoint experiment with 393
respondents in Austria to investigate the effect of EV-PV-BS product bundles on purchase
intention. Our data show that a majority of potential EV drivers, given the choice, would prefer
to purchase an EV in such a bundle. Further, the purchase intention for a PV and BS is twice as
high in a bundle with an EV than standalone. Segmentation analysis identified four potential
customer segments, which we labelled “Price-Sensitive Non-Owners”, “Energy Self-Sufficient
Owners”, “Economically Rational Owners” and “Likely Non-Adopters”. The segments
specifically differ in their product preferences, which highlights a need for designing
customized bundle offerings. Moreover, we show that policy incentives are more effective
when product bundles are labelled with prices tags already discounted by subsidies. We draw
implications for practitioners and policymakers, as well as proposing areas of further research.
Keywords: Electric vehicle; photovoltaic solar panel; battery storage; product bundling;
choice-based conjoint; latent class analysis
Highlights:
• A significant share of EV drivers prefer purchasing an EV in a bundle to standalone
• Product bundling increases the adoption rate of complementary new technologies
• Heterogeneity in bundle preferences requires variety in bundle offerings
• Product bundles should be advertised with purchase prices reduced by subsidies
24 Early version of this paper has been presented at the 1. Advanced Demand Modeling Workshop for
Electromobility at 16.03.2018 and at 6th International PhD Day of the Austrian Association of Energy Economics
at 07.09.2018 in Vienna/Austria. Furthermore, this paper is accepted for the EVS 32 in Lyon/France from 19.05.-
22.05.2019 and has received an invitation for “Revise and Resubmit” from Ecological Economics at 28.12.2018
* Department of Operations, Energy, and Environmental Management, Alpen-Adria-Universität Klagenfurt
# Vienna University of Economics and Business Institute for Strategic Management
Paper 2 70
1. INTRODUCTION
Sales of electric vehicles (EV), i.e. battery-driven electric (BEV) or plug-in hybrid
electric (PHEV) vehicles, have been increasing over the last couple of years for several reasons
(IEA, 2018). First, climate change combating organizations have long been arguing for low-
carbon emission modes of transportation such as EVs to meet greenhouse gas reduction targets
(IPCC, 2014; UNFCCC, 2015). Second, after years of lip service, European car manufacturers
have committed to adjusting their engines and platforms towards electrical fleets within the
next decade (e.g. VW, Mercedes, BMW). Thus they are following EV pioneers such as Tesla
(Bloomberg, 2018). Third, as from 2025 a few countries (e.g. Norway) will implement a ban
on internal combustion engine sales (IEA, 2018).Global EV registrations reached the one
million threshold in 2016 (IEA, 2016a) and they are growing exponentially (IEA, 2018).
However, critics of EVs emphasize that the shift from fossil-fuelled to power-fuelled
engines only combats climate change if coupled with decarbonized electricity production (e.g.
Bleijenberg and Egenhofer, 2013). Several experts even predict a rise in greenhouse gas
emissions in certain regions due to EVs replacing fossil-fuel cars (Holland et al., 2015; Zivin et
al., 2012). Producing power from renewable energy sources such as wind and solar is a growing
trend which many policy makers encourage (e.g. European Commission, 2014). However, the
larger part of power globally is still generated from fossil energy sources (e.g. coal, gas) (IPCC,
2012).
Hence, the proportion of renewable power used by EVs needs to be increased to reduce
their carbon footprint. A possible solution is to offer EVs in combination with photovoltaic
(PV) solar panels and battery storage (BS) for producing and storing renewable energy at
residential sites. Such product bundles would help to reduce the CO2 emission of EVs. Delmas
et al. (2017: 235) already argue that “households that invest in both solar panels and electric
vehicles, and size their solar system to offset the additional electricity used by their vehicle, can
eliminate their carbon footprint from household and transportation activities”.
Besides this environmental benefit, EV-PV-BS product bundling could also be an
incentive that will increase EV adoption, since as a niche market EVs are mostly purchased by
technologically or environmentally discerning people (Axsen et al., 2016; Hidrue et al., 2011).
Also, PV and BS are still fairly new products on the market and despite high growth rates they
still suffer from a relatively low level of adoption for several reasons (IEA, 2016b; IPCC, 2012).
One reason remains the high installation cost, but another is the limited knowledge among
consumers which coincides with a higher level of perceived risk and potentially also a less
Paper 2 71
positive attitude towards these products. This is quite a common circumstance for many newly
launched products (cf., Jhang et al., 2012; Reinders et al., 2010; Schilke and Wirtz, 2012). For
a customer to adopt a new product, the risk-adjusted value must exceed the purchase price
(Kalish, 1985) or, in other words, the benefits have to outweigh the costs (Wang et al., 2008).
In marketing literature, one strategy often proposed to decrease the perceived risk and
consequently to increase customers’ willingness to purchase (new) products, is product
bundling (Eppen et al., 1991; Reinders et al., 2010; Simonin and Ruth, 1995), which entails
packing various products into one offering. Due to decreased search and assembly costs, such
bundling also increases consumer convenience (Harris and Blair, 2006; Kim et al., 2008).
Moreover, according to Reinders et al. (2010) consumers specifically value product bundles
with high fit, i.e. the bundle has products that are either complementary or, at least, related.
Consumers view the products in focus here (EVs and PVs (with/without BS)) as
complements that together generate higher customer value than separately (Agnew and
Dargusch, 2017; Delmas et al., 2017). Thus, it seems prudent to offer these products in a bundle.
Cherubini et al. (2015) already identify (product/product-service) bundles as one of five key
success factors in accelerating the diffusion of EVs, and call for further research to develop
such bundles and test their effectiveness. To date, research has covered only the impact of single
add-on services and not how multiple product or service add-ons affect the preference for EVs
(Fojcik and Proff, 2014; Hinz et al., 2015).
Against this background, our paper has the following research objectives. First,
following calls of Cherubini et al. (2015) to study the effect of multiple add-ons, we aim to
understand whether bundling EVs with renewable power supply systems (PV with/without BS)
has a positive impact on the joint adoption of these three products, thus constituting an effective
strategy to increase the share of “greener” EVs on the road. Investigating the combination of
the three products (EV, PV, BS) and their related consumer preferences are considered a novel
contribution to the literature in this field. Second, building on previous research evidence that
consumer preferences in various clean technology fields are quite heterogeneous (Kaufmann et
al., 2013; Salm et al., 2016; Tabi et al., 2014), our paper aims to contribute to extant literature
by identifying different customer segments and by investigating how they differ in their
characteristics (socio-demographic and psychographic variables) and preferences regarding
product attributes and policy incentives. To achieve these research goals, we conducted a survey
including a choice-based conjoint (CBC) experiment with a unique sample of potential EV
drivers in Austria (n = 393). With the derived part-worth utilities we performed a latent class
Paper 2 72
analysis to identify customer segments with distinct preferences related to EV-PV-BS product
bundles. Our findings show that a significant share of potential EV drivers prefer purchasing
an EV bundled with complementary clean technologies to a standalone product. However, the
results also highlight the necessity of customizing product bundles to the preferences of
different customer segments.
This paper is structured as follows: in Section 2, we give information on product
bundling and its typologies. We also provide a brief literature review on the benefits product
bundling and bundling in the context of clean technologies have for consumers. In Section 3,
we introduce our methodological approach and dataset. Section 4 presents the results of our
survey and choice experiment, including the latent class analysis. Section 5 concludes the paper
and discusses implications for marketers and policy makers, as well as limitations and areas for
further research.
2. LITERATURE REVIEW
2.1 Definition and typology of bundling strategies
The concept of ‘bundling’ was first introduced by Adams and Yellen (1976) in the mid-
seventies and has since then been applied and analysed in the field of economics and marketing
(cf., Guiltinan, 1987; Stremersch and Tellis, 2002; Yadav, 1995). A multitude of definitions
exist (Chiambaretto and Dumez, 2012); however, the one most widely used defines bundling
as “the sale of two or more separate products in one package”25 (Stremersch and Tellis 2002:
57).
Stremersch and Tellis (2002) introduced a standard typology that summarizes all
characteristics bundles can have in a single framework using two dimensions, namely bundle
form and bundle focus. Regarding bundle form, the literature distinguishes mainly between
pure, mixed and unbundling strategies (Adams and Yellen, 1976; Guiltinan, 1987;
Schmalensee, 1984; Stremersch and Tellis, 2002). In a pure bundle the offered products are
sold only in a bundle and are not even available separately. This bundling strategy enables
companies to reduce customer heterogeneity (Schmalensee, 1984). Completely opposite is the
unbundling strategy in which all associated products are sold as separate items. This strategy
suits products with strong homogenous customer preference, thus these customers typically
25 In general, the term “product” refers to both goods and services (Stremersch and Tellis, 2002). However, in the
empirical study of this paper we focus on goods only.
Paper 2 73
prefer one product and not necessarily another associated one (Schmalensee, 1984). The mixed
bundling strategy has features of two extreme bundle forms, thus it makes products available
in a bundle and separately. This strategy is often considered the optimal strategy if consumers
have heterogeneous preferences (Guiltinan, 1987; Schmalensee, 1984; Stremersch and Tellis,
2002). Mixed bundling reduces the heterogeneity in consumers’ reservation prices (i.e., the
maximum a buyer is willing to pay) and hence taps into consumers’ surpluses more efficiently
(Schmalensee, 1984). However, the most optimal bundling strategy depends on various factors
and needs to be evaluated separately for each bundle (Arora, 2011; Lee and O'Connor, 2003).
Regarding the second framework dimension, bundle focus, the literature distinguishes
between price bundling and product bundling (Guiltinan, 1987; Reinders et al., 2010;
Stremersch and Tellis, 2002). Price bundling involves selling two or more products in a
discounted package, regardless of the integration level of the products in a bundle (Guiltinana,
1987; (Simonin and Ruth, 1995). In contrast, product bundling involves integrating and selling
the products in a package at any price, but generating value by adding complementary products.
Hence, the main items theoretically do not need to be specially discounted (Stremersch and
Tellis, 2002), even if most consumers expect bundled products to cost less than the products
sold separately (Heeler et al., 2007; Tanford et al., 2011). Overall, the literature argues that
consumers tend to purchase products separately if a bundle of products is not at least discounted
or offered with value-adding complementarity (Harris and Blair, 2006; Reinders et al., 2010).
Therefore, the differentiation between price and product bundling is of high managerial
relevance triggering different strategic choices depending on the particular company’s
objectives. Price bundling is a pricing and promotional tool which can be applied by the
marketing department for a brief time period at short notice. In contrast, product bundling is
deployed more strategically and at longer term to distinguish themselves from competition.
This requires a more holistic and collaborative approach along the company value chain
(Stremersch and Tellis, 2002). EV, PV and BS are highly complementary, but they are relatively
new and complex products (Agnew and Dargusch, 2017; Delmas et al., 2017). Therefore, a
product-bundling strategy seems to be more appropriate in this context and we, thus, focus on
product bundles in the remainder of our paper.
2.2 Consumer benefits from product bundling
In the last few decades, product bundling got substantial research attention (cf.,
Stremersch and Tellis, 2002). Previous studies identified three major consumer benefits of
Paper 2 74
product bundling: (1) increased value through complementarity, (2) a decrease in (perceived)
risk, and (3) increased convenience.
Complementarity among bundle products is a prerequisite for successful product
bundling (Harris and Blair, 2006; Stremersch and Tellis, 2002). If the qualities of the products
in a bundle are unknown or unrelated to consumers, bundling might not be considered superior
to unbundling (Choi, 2003). In general, complementarity enhances the utility of one or more
jointly used products, which leads to a more positive evaluation of bundles (e.g. Reinders et al.,
2010; Simonin and Ruth, 1995; Stremersch and Tellis, 2002). For instance, Reinders et al.
(2010) argue that compared to a moderate fit, a high degree of fit between the products in a
bundle has a positive impact on the evaluation of those products and the intention to purchase
them in a bundle. Similarly, regarding complementary products, Simonin and Ruth (1995)
identify a moderating effect on the relationship between the prior attitudes to the individual
products and attitudes to the bundle. Stremersch and Tellis (2002) claim that well-integrated
products create value for consumers via, for instance, improved performance (e.g. workout
program and personalized dieting), seamless interaction (e.g. PC systems) or interconnectivity
(e.g. telecom systems).
Depending on the industry and individual characteristics, consumers consider
purchasing bundled products to be less risky than unbundled products. In industries, which lack
formalised technology standards, e.g., consumers perceive purchasing bundled products as
safer than unbundled ones (Lawless, 1991). A significant number of consumers who do not
know the products (knowledge uncertainty) or have less confidence in their ability to make a
1987; Urbany et al., 1989). In addition, researchers argue that product bundles increase
consumer acceptance by reducing customers’ perceived risks due to product spillover effects
(e.g. Choi 2003, Reinders et al., 2010, Simonin and Ruth 1995). For instance, Choi (2003)
developed a rationale for quality transfer from existing experience goods onto new experience
goods in a bundle based on the information leverage theory. He suggests that the use of a
product with established quality will benefit the new product in the bundle by overcoming the
asymmetry of information in the market.
Deciding to purchase more than one product in a single purchasing event is often
convenient to the customer (Stremersch and Tellis, 2002). This convenience benefit could be
influenced by the usefulness and ease of using the bundle. Schilke and Wirtz (2012), e.g., have
shown that consumers perceive bundled broadband services which include internet access, an
Paper 2 75
internet-linked telephone, and internet television as more favourable, if they also rate the
products’ usefulness and ease of use highly. Moreover, scholars argue that by purchasing
product bundles consumers will gain convenience benefits due to reduced search and assembly
efforts (Guiltinan, 1987; Harris and Blair, 2006; Kim et al., 2008). Kim et al. (2008) claim that
to avoid the complex purchase-decision process and to decrease search time, consumers choose
product bundles that online travel agents offer rather than themselves purchasing each product
separately. The more complex the product or the more unfamiliar the potential consumer is with
the product, the higher the perceived value of a reduction in search costs through integrated
product bundles (Harlam et al., 1995).
2.3 Product bundling in the context of clean technologies
To date very little has been published on product bundling in the context of clean
technologies, specifically those related to EV, PV and BS bundling. Several studies analyse the
EV bundled with add-on services. For instance, Hinz et al. (2015) or Fojcik and Proff (2014)
evaluated the impact of single add-on services (e.g. mobility guarantee, vehicle-to-grid (V2G),
IT-based parking, intelligent charging system, charging station finder, etc.) on the acceptance
of EVs. Their consensus is that add-on services could increase the purchase intention and
willingness to pay (WTP) for EVs depending on which services and mobility concepts are
added to the bundle. Particularly, V2G is a promising add-on service to EVs. By storing the
peak loads from renewable energy sources, an EV with V2G technology can contribute to the
balance of power grids (Sovacool et al., 2017). Parsons et al. (2014) show that EVs offered with
V2G services could increase the market acceptance if upfront payments or pay-as-you-go is
available for the V2G service. However, Hidrue and Parsons (2015) argue, that the WTP of
V2G-EVs (special type of EV that returns power to the grid) is still very low in relation to the
current and future cost of V2G-EVs. However, fast changing technology that could lower V2G-
EVs’ cost in the midterm future, could change this. Overall, bundling EVs with add-on services
is becoming increasingly important for e-mobility success (Laurischkat et al., 2016). Cherubini
et al. (2015) have even indicated product/product-service bundles as a key factor for increasing
EV acceptance, and called for it to be further researched.
Only a few studies have investigated the consumer preference related to PV and BS
bundling. Oberst and Madlener (2015), for instance, investigated German households’
preferences regarding prosumerism and their willingness to adopt renewable energy based
micro‐generation technologies. They find that households consider becoming prosumers by
Paper 2 76
investing in PV-BS systems with a high level of electricity self-supply, and that this is a more
important purchase driver than green electricity or a profitable investment. Galassi and
Madlener (2016) show that Italian PV owners and potential PV owners prefer a PV system with
a BS facility which is owned, controlled and maintained by an external company, e.g. a utility.
Hence, they argue that consumers prefer a “rent-your-roof” solution to a “plug-and-play” one.
Agnew and Dargusch (2017) also recently showed that accepting PVs increases when they are
bundled with a BS. But additionally, they pointed out that some safety, quality and knowledge
issues remain unresolved in the market, and call for instituting consumer “energy literacy”
measures. This would require a better understanding of consumers, their preferences and
expertise.
EV-PV combinations have been researched from different angles. For example, Klör et
al. (2017) considered how used EV batteries can be repurposed for storing power from
renewable sources (e.g. PV), and how add-on services could overcome consumers’ information
asymmetries in used EV battery purchases. Ida et al. (2014), contrastively, focused on consumer
preferences and conducted a stated preference analysis for i.a. PV and EVs. They derived
estimates for the penetration rates, potential reduction of greenhouse gas, and WTP in order to
foster the diffusion of such clean equipment in Japan. Although, they did not investigate
consumer preferences of these two products in a bundle, Delmas et al. (2017) did so quite
recently when they investigated the preference for jointly purchasing EVs with PVs in
California. They claim that such joint purchasing will continue to grow due to the products’
complementarity. Further, they argue that household sector emissions will decline if more
households drive EVs and own solar panels. In spite of consensus on the benefits and market
outlook regarding combining EVs with PV systems, there has to date not been much research
on EV-PV bundling from a consumer preference perspective.
3. METHODS AND DATA
3.1 Choice-based conjoint
We investigated potential EV drivers’ willingness and preferences to purchase EVs
bundled with PV and BS by applying a stated preference approach. More specifically, we
performed a choice-based conjoint (CBC) experiment (Orme and Chrzan, 2017). In contrast to
a revealed preference approach (dealing with actual behaviour), a stated preference approach is
based on behavioural intentions and responses to hypothetical choice situations (Adamowicz et
al., 1994; Ben-Akiva et al., 1994). This offers the advantage of running choice experiments for
Paper 2 77
products or situations that appear in the market only limitedly or not at all (Adamowicz et al.,
1994; Louviere et al., 2000). That the EV, PV and BS markets are still in an early stage of
diffusion, and that product EV-PV-BS bundles are currently not even available in the market,
have motivated our choice to conduct a CBC experiment.
However, the stated preference method comprises some hypothetical bias, meaning that
the stated and actual behaviours of the respondents are most likely not a perfect fit. Still, the
more familiar the respondents are with the survey setting, the lower the hypothetical bias
(Schläpfer and Fischhoff, 2012). Hence, the novelty of this product bundle (EV + PV + BS) to
an average Austrian consumer was managed by focusing the experimental design on a car (i.e.
EV) purchase to which we simply added a PV with/without a BS without going into
complicated technical details about these products. In addition, our study focused on
respondents who are very likely to become early adopters of EVs on the basis of positive
attitudes towards EVs which is a strong indicator of future EV purchase (Nayum et al., 2016).
Further, the phenomenon of hypothetical bias could be partially overcome by applying indirect
enquiring practices such as CBC experiments (Schläpfer and Fischhoff, 2012).
CBC is a well-established method in marketing research (Green and Srinivasan, 1990)
that has also been applied in the context of EVs (Ewing and Sarigöllü, 2000; Hackbarth and
Madlener, 2016; Hidrue et al., 2011) and small-scale renewable energy technology (Gamel et
al., 2016; Salm et al., 2016). CBC can be used for measuring consumer preferences for products
and services, and for simulating potential market sizes or defining post-hoc customer segments
based on consumer preferences (Gustafsson et al., 2013; Orme and Chrzan, 2017). This can be
achieved by simulating a buying situation, in which respondents need to choose the most
preferred option from several alternatives with varying attribute levels. By redoing this several
times the underlying preference for the attribute levels can be effectively elicited in the form of
average part-worth utilities for attribute levels and relative importance weights for each of the
attributes (Green and Rao, 1971; Green and Srinivasan, 1990; Gustafsson et al., 2013). The
theoretical foundation for these analyses is the classical utility theory which assumes that every
individual has a certain utility maximization attitude. Moreover, every product has a certain
utility for each individual, which can be defined as the sum of the part-worth utilities for the
various attributes of that product (Lancaster, 1966).
Paper 2 78
3.2 Survey and experimental design
We used Sawtooth Software Lighthouse, the standard application for CBC experiments
in marketing research, to design, administer, conduct and analyse the survey. The overall
questionnaire consisted of two parts. In the first part, we checked the respondents’ interests in
and attitude towards EVs, their willingness to purchase an EV including the time-horizon plus
a few first demographics (age, gender, federal state) for representativeness cross-checks in the
sampling process. Then respondents needed to answer psychographic questions about their
worldviews, technological affinity, and environmental identity (see Table A.1 in the appendix).
Next, they were asked to indicate their purchase intentions related to standalone PV and BS
systems as well as their WTP regarding their preferred EV.
The second part included the CBC experiment. In an introduction we provided
background details about the imagined context of purchasing an EV with the opportunity to buy
a PV and BS in a bundle with the car, and about the different options of PV and BS systems
ownership (owner or leaser with ownership option). Before starting the conjoint experiment,
we provided a sample choice task with all the variables including detailed descriptions of the
attributes. The selected attributes and attribute levels for the EV-PV-BS product bundles were
defined based on twelve interviews with lead users of at least one of these products and on sales
discussions with retailers. In addition, we conducted a pre-test of the CBC experiment with 45
potential EV drivers and several experts from the car and utility industry to confirm the
relevance and suitability of the chosen attributes and levels.26 Table 1 provides an overview of
the six attributes, a detailed description, and the corresponding attribute levels finally used in
the CBC experiment.
In the CBC experiment the respondents were shown a series of 12 choice tasks. Each of
the choice tasks presented three different product bundle alternatives and a non-option (in which
the person would prefer the EV without the power-supply add-on products). From this set of
four options the respondents had to choose their preferred option. The EV in each choice task
was fixed at 400 km range, 150 horse power, 40-60 minutes per full-charging27 and PV and BS
add-on products were added with varying characteristics. A full profile method was used, which
means that all attributes were presented for each set of alternatives (Orme and Chrzan, 2017).
26 The chosen attributes/levels are considered the most relevant for the purchase decision of the bundle in focus. 27 The car characteristics were taken from the Nissan Leaf 2.0, which was released at the beginning of 2018. The
Nissan Leaf model was the world’s best-selling electric car in 2017 (Bloomberg (2017).
Paper 2 79
After finishing the CBC experiment a few last demographics such as income, education,
household size, etc. were interrogated, which are used to characterize the customer segments in
the results section below.
TABLE 1: ATTRIBUTES AND ATTRIBUTE LEVELS IN THE CHOICE-BASED CONJOINT DESIGN
Attributes Attribute description Attribute levels
PV/BS add-on
(ownership)
The PV with/without BS can be purchased as owner or
leaser with ownership option. As a leaser no investment
costs incur, but you conclude a power purchase
agreement to receive power from the operator of this
system at very good price for 15 years. After this
period, the devices will become your property and you
will be able to get your electricity for free from your
own facility.
PV + BS owner (no monthly payment)
PV owner (no monthly payment)
PV + BS leaser with ownership option
(monthly payment)
PV leaser with ownership option
(monthly payment)
Self-
sufficiency
rate
The PV with/without BS can produce/procure a certain
percentage (25-100%) of the electricity demand itself,
and consequently save a percentage of the electricity
costs, similar to the total annual household cost.
Up to max. 25%
Up to max. 50%
Up to max. 75%
Up to max. 100%
Amortization
period
Period during which the investment costs for the PV
with/without BS will be returned to the owner/operator.
8 years
12 years
16 years
20 years
Provider The product bundle can by either purchased by an all-
in-one provider or by a set of dealers.
All-in-one car dealer/OEM
All-in-one utility
All-in-one specialist dealer
Diverse specialist dealers
Policy
incentive
The government of Austria can subsidise investments
in EVs, PV systems and BS.
0%
Up to max. 10%
Up to max. 20%
Up to max. 30%
Purchase price From this purchase price (which includes the EV and in
the case of the “owner” option, investment in a PV
system with/without BS), the state subsidy must still be
deducted to arrive at the final total price. Monthly
payments in the case of “leaser with ownership option”
are not included in this price.
EUR 25,000
EUR 30,000
EUR 35,000
EUR 40,000
EUR 45,000
3.3 Data collection and sample
The survey respondents were recruited by the market research company market in
February/March 2018 from their online panel pool of more than 20,000 active users in Austria.
The participants who succeeded to fill in the survey form properly, were financially rewarded,
as is currently common practice in market research (cf. Gamel et al., 2016; Salm et al., 2016).
A total sample of 1,251 respondents were invited based on quota sampling in order to ensure
Paper 2 80
representativeness, considering the distribution of gender, population by federal state and age
(see Table A.2 in the appendix).
From our total sample (N =1,251) we decided to remove four subgroups (Figure 1
illustrates the filter funnel). First, 438 respondents in the total sample who showed a negative
attitude towards EVs, were excluded. According to Nayum et al. (2016) a positive attitude
towards EVs, besides other factors, is an important predictor for purchasing an EV. Second, a
sub-segment of 189 respondents who were not willing to purchase an EV as their next car
despite a positive attitude towards EVs, were excluded. The remaining group of people with
positive attitudes towards EVs and willing to purchase one, (n= 624) represented approximately
50% of the total sample, which is in line with other studies in Austria (cf., Hampl and Sposato,
2017). These respondents could be further segmented according to their planned EV purchase
timeframe. Thus, third, since we aim to draw policy and marketing relevant implications from
this study, we focused only on the group of people willing to purchase an EV within the next 5
years (n = 431). Fourth, the focus sample was cleaned by removing 38 speeders 28 and
flatliners29, leaving a final sample of 393 respondents. These respondents provided 4716 choice
observations (12 choices completed by each of the 393 respondents), which is considered more
than sufficient for further analyses (Gustafsson et al., 2013).
As a final step, we applied statistical tests to investigate the representativeness of our
sample (see Table A.2 in the appendix). The results indicate that our final sample of 393
respondents only slightly differs from the Austrian population profile generally by being better
educated and having a higher average household income.
28 Respondents who were among the 10% that read the instructions of the CBC experiment the fastest, i.e. in less
than 18 seconds, while the mean was 60 seconds, and who completed the choice tasks among the 5% fastest
respondents, i.e. in less than 74 seconds, while the mean was 151 seconds, were excluded. 29 The average root likelihood (RLH) can be used as a measure of fit to assess data quality. In this study, as each
choice task presented four options, it would be predicted that each alternative would be chosen with a probability
of 25% (corresponding RLH of 0.25). All answer sheets scoring below 0.25 were removed.
Paper 2 81
FIGURE 1: FILTER LOGIC FROM TOTAL SAMPLE (N = 1,251) TO FINAL SAMPLE (N= 393)
4. RESULTS
4.1 Relative importance of attributes
As a first analysis of the CBC experiment, we report the relative importance scores of
the different attributes in Table 2. These scores describe the size of each attribute’s influence
on the purchase decision. The higher this importance score, the larger the difference between
each attribute’s highest and lowest part-worth utility, and the larger the contribution of this
attribute to the overall utility of, in our case, the product bundle. Due to standardization the
importance scores sum up to 100% across all attributes, which enables comparisons of effects
between attributes (Gustafsson et al., 2013). Unsurprisingly, the purchase price is ranked first
(31.1%) as the most important attribute for purchase decision. Ranked a close second and third
place are a PV/BS add-on (ownership) (18.7%) and a power self-sufficiency rate (16.1%). These
are followed by the amortization period (14.5%). The difference between the purchase price
and the other parameters confirms that price is the most important purchase driver, therefore
the adoption success of such product bundles are highly dependent on the cost-curve
development of EV, PV and BS (Seba, 2014). Interestingly, the attributes policy incentive
(10.6%) and provider (9.1%) are of minor importance. Further, the standard deviation of the
Paper 2 82
policy incentive attribute is relatively high, which points to some heterogeneity among survey
participants related to the attribute levels.
4.2 Part-worth utilities per attribute level
We chose the Hierarchical Bayes (HB) procedure, regarded the most advanced and
commonly used estimation method in recent CBC studies (cf., Hille et al., 2018; Hinnen et al.,
2017; Salm et al., 2016), to estimate the part-worth utilities per attribute level. The HB
procedure estimates part-worth utilities at the individual level for each respondent, which
enables a better hit ratio in predicting consumer choices as well as statistically more accurate
results (Orme and Chrzan, 2017) than traditional aggregate models (e.g. multinomial logit
analysis (McFadden, 1986)). The latter are criticized for losing a great deal of information by
aggregating data for all individuals (Gustafsson et al., 2013; Rossi and Allenby, 2003).
Table 2 gives the average part-worth utility for each attribute level of the CBC
experiment with the corresponding standard deviation and confidence interval. The utility
scores reflect a certain relative desirability of an attribute level compared to other levels of the
same attribute. The higher the utility score, the stronger the positive influence of the specific
attribute level on the potential EV driver’s decision to choose the bundled product. Negative
values indicate a lower desirability and a decrease in overall utility. All utilities are interval-
scaled and zero-centred, so that the sum of all utilities per attribute is zero. Further, the
magnitude of the utilities is highly dependent on the selected range of attribute levels.
Therefore, to compare utility values across attributes is not meaningful, while comparison
solely between different levels of a given attribute, is (Gustafsson et al., 2013).
The results unsurprisingly revealed that potential EV drivers most prefer lower prices,
shorter amortization periods and high sufficiency rates. The tipping point from positive to
negative part-worth utilities for the attribute amortization period lies between 12 and 16 years,
and for self-sufficiency rate between 50 and 75%. The results also reveal a negative preference
for product options with no policy incentive. Interestingly, there is a slight preference for
purchasing the product from one instead of multiple providers, but without a clear preference
for the type of provider. Furthermore, purchasing PV and BS in a bundle is preferred along with
ownership rather than renting.
Paper 2 83
TABLE 2: HIERARCHICAL BAYES MODEL ESTIMATION OF MEAN UTILITY VALUES AND MEAN
Diverse specialist dealers -21.17 17.73 -24.58 -14.18
Policy incentive (m = 10.55%; SD = 5.52)
Up to max. 30% 21.41 23.53 13.08 25.68
Up to max. 20% 7.00 23.66 0.71 13.62
Up to max. 10% -7.08 19.00 -14.37 -0.11
0% -21.32 26.09 -27.37 -15.31
Purchase price (m = 31.14%; SD = 12.52)
EUR 25,000 77.16 56.40 65.09 87.83
EUR 30,000 38.97 29.63 30.17 48.35
EUR 35,000 16.87 23.43 10.61 23.56
EUR 40,000 -41.67 36.55 -48.71 -28.99
EUR 45,000 -91.28 48.34 -95.22 -69.98
None -31.93 316.58 -43.31 14.01
a We use the average root likelihood (RLH) as an indicator for data quality and measure of fit, which is the
geometric mean of all predicted probabilities. Its acceptance threshold depends on the number of options per choice
task. Since we offered our respondents four options, the predicted probability of randomly choosing one of these
options is 25%. Thus, our RLH threshold is 0.25. Our RLH value is relatively high (0.647), which indicates that
our predictions are about 2.6 or 2.7 times better than the random chance level. b Mean relative importance scores per attribute (m) and corresponding standard deviation (SD) in parentheses. The
importance scores sum up to 100%. c Confidence interval.
Paper 2 84
Further analyses revealed that in a best-case scenario, where all attributes are set to the
most preferred level, 77.4%30 of our sample would purchase an EV with energy-supplying and
storing facilities (see Table A.3 in the appendix). And even in a base case scenario, with the
status-quo of product prices and specifications of February 2018, almost 40.6% of potential EV
adopters would prefer purchasing an EV with PV and BS (see Table A.4 in the appendix). In
addition, EV-PV-BS bundling increases the acceptance of PV and BS significantly: today,
40.6% of all respondents are willing to purchase PV and BS in a bundle with an EV, rather than
purchasing the PV as a standalone (25.3%31; p = 0.0332) or the BS as a standalone (18.7%7; p =
0.018).
4.3 Latent class analysis
4.3.1 Identification of the number of segments
To investigate potential heterogeneity in the EV drivers’ preferences, we performed a
segmentation analysis based on the final sample (n = 393). Latent class analysis is the most
frequently used procedure to cluster respondents based on preferences elicited in a CBC
experiment (Campbell et al., 2011; Garrod et al., 2012; Hille et al., 2017; Morey and Thiene,
2017; Tabi et al., 2014). It is also considered superior to other segmentation techniques
(Desarbo et al., 1995) because of its higher reproducibility and ability to create groups of similar
size (Sawtooth Software, 2004b).
The only downside of the latent class approach is its use of different starting points at
each computation. This can be overcome with two cross-checks. First, by re-running the model
several times. Second, by randomly splitting the sample into two parts and performing the
analysis separately on each segment (Sawtooth Software, 2004). Applying these to our case we
examined whether comparable segments emerged, which we confirmed. Each time the model
estimated the solutions from 2 to 6 segments, and we kept the solution with the highest chi-
square for our further analyses.
To determine the best model we used a subset of the recommended main criteria: percent
certainty, Consistent Akaike Information Criterion (CAIC), and chi-square (Desarbo et al.,
30 We applied a Randomized First Choice Model (Sawtooth Software Market Simulator) to estimate the share of
preference for the product bundle scenarios. For further details see for instance Orme and Chrzan (2017) and Hille
et al. (2018). 31 Purchase intention figures derived from direct questioning in part of the survey. 32 The test of significance was a chi-square statistic.
Paper 2 85
1995; Sawtooth Software, 2004b). Percent certainty is used to reveal how much better an
identified solution (number of segments) is compared to no segments; in other words, to check
how well the proposed segmentation solution fits the data. Therefore, the higher this figure, the
better the model. However, adding segments constantly increases it, hence we further used
CAIC (Sawtooth Software, 2004) developed by Bozdogan (1987) and adjusted from
Ramaswamy et al. (1993).
CAIC = -2 Log Likelihood + (nk + k - 1) x (ln N +1)
k = the number of groups
n = the number of independent parameters estimated per group
N = the total number of choice tasks in the data set
CAIC is most frequently used for determining the number of segments. In contrast to
percent certainty, CAIC indicates the best solution when researching its minimum (Sawtooth
Software, 2004b). Chi-square, like percent certainty, indicates whether a segmentation solution
is significantly better than the null solution. The measure is calculated by subtracting twice the
log-likelihood value of a null solution from two times the log-likelihood of the respective
grouping solution (Sawtooth Software, 2004).
TABLE 3: SUMMARY OF BEST REPLICATIONS LATENT CLASS ANALYSIS
Groups Percent certainty CAIC Chi-square
2 28.37 9754.12 3709.22
3 32.58 9401.88 4260.08
4 34.61 9335.56 4525.04
5 35.70 9391.60 4667.63
6 36.68 9461.46 4796.41
Instead of looking only at the highest and lowest number of these three variables, we
consider it helpful to look at the differences between the levels as well, as Sawtooth Software
(2004) has recommended. In all three cases we identified a slower increase for percent certainty
and chi-square and a soft re-increase in CAIC as described above from the solution of four to
five groups. Hence, the model we finally chose is the 4-group solution with a percent certainty
of 34.61, a CAIC of 9335.56, and chi-square of 4525.04 (see Table 3).
Paper 2 86
4.3.2 Description of the identified segments
TABLE 4: HIERARCHICAL BAYES MODEL ESTIMATION OF MEAN UTILITY VALUES PER SEGMENT
Segments
Segment 1:
Price Sensitive
Non-Owners
Segment 2:
Energy Self-
Sufficient
Owners
Segment 3:
Economically
Rational Owners
Segment 4:
Likely Non-
Adopters
Segment size n = 140 n = 83 n = 87 n = 83
PV/BS add-on (ownership)
PV + BS owner (no monthly
payment) -30.98 (-2.01) a 114.83 (14.50) 46.72 (7.15) -9.10 (-0.28)
a Price Sensitive Non-Owners vs. Energy Self-Sufficient Owners: p < 0.05. b Price Sensitive Non-Owners vs. Likely Non-Adopters: p < 0.05. c Price Sensitive Non-Owners vs.
Economically Rational Owners: p < 0.05. d Energy Self-Sufficient Owners vs. Likely Non-Adopters: p < 0.05. e Energy Self-Sufficient Owners vs. Economically Rational
Owners: p < 0.05. f Likely Non-Adopters vs. Economically Rational Owners: p < 0.05.
Paper 2 102
TABLE A.2. CHARACTERISTICS OF THE TOTAL AND FINAL SAMPLE IN COMPARISON TO THE
AUSTRIAN POPULATION
Variables
Total sample
(N = 1,251; in %)
Final sample
(n = 393; in %)
Austrian population
(in %) b
Gender
(x2 = 0.124, d.f. = 1,
p = 0.725) a
(x2 = 0.712, d.f. = 1,
p = 0.396)
Female 48.5 45.0 50.8
Male 51.5 55.0 49.2
Age
(x2 = 0.988, d.f. = 3,
p = 0.804)
(x2 = 1.852, d.f. = 3,
p = 0.582)
18-29 years 14.1 11.7 19.3
30-44 years 25.7 30.3 25.5
45-59 years 30.7 31.3 29.4
60-80 years 29.3 26.7 25.8
Education
(x2 = 36.996, d.f. = 3,
p = 0.000)
Compulsory school 2.3% 26.9%
Vocational training 35.9% 45.8%
High school 26.7% 14.6%
University 36.1% 12.6%
Household income/month
(x2 = 130.323, d.f. = 2,
p = 0.000)
25% percentile 2,500 1,601
50% percentile 3,000 2,611
75% percentile 4,000 3,995
Federal state
(x2 = 4.125, d.f. = 8,
p = 0.843)
(x2 = 0.667, d.f. = 8,
p = 0.999)
Burgenland 7.9% 2.5% 3.3%
Carinthia 5.8% 5.9% 6.4%
Lower Austria 19.9% 21.6% 18.9%
Upper Austria 20.8% 17.3% 16.7%
Salzburg 4.8% 5.3% 6.3%
Styria 14.3% 14.2% 14.1%
Tyrol 6.9% 6.9% 8.5%
Vorarlberg 5.0% 4.8% 4.4%
Vienna 14.5% 21.4% 21.4%
a The results from chi-square tests are included in parentheses, which show whether significant differences could
be identified between the total/final study sample and the Austrian population. b Source: STATISTIK AUSTRIA, 2018b).
Paper 2 103
TABLE A.3. BEST-CASE SCENARIO FOR THE SENSITIVITY ANALYSES
Best-case scenario
Option 1 Option 2 Option 3
PV/BS add-on (ownership) PV + BS
(Ownership)
PV + BS
(Non-ownership)
Would not choose any
of those options
Amortization period 8 Years 8 Years
Self-sufficiency rate 100% 100%
Provider All-in-one provider All-in-one provider
Policy incentive 30% 30%
Purchase price EUR 25,000 EUR 25,000
Individual simulated share of
preference for product
bundle options
48.2% 29.3% 22.6%
Aggregated simulated share
of preference for product
bundle options
77.4% 22.6%
TABLE A.4. BASE CASE SCENARIO FOR THE SENSITIVITY ANALYSES
Base case scenario
Option 1 Option 2 Option 3
PV/BS add-on (ownership) PV + BS
(Ownership)
PV + BS
(Non-ownership)
Would not choose any
of those options
Amortization period 20 years 20 years
Self-sufficiency rate 25% 25%
Provider Different Different
Policy incentive 0% 0%
Purchase price EUR 45,000 EUR 45,000
Individual simulated share of
preference for product
bundle options
23.4% 17.2% 59.4%
Aggregated simulated share
of preference for product
bundle options
40.6% 59.4%
PAPER 3: EXPLORING CONSUMER HETEROGENEITY IN WILLINGNESS TO
PAY FOR ELECTRIC VEHICLE PRODUCT BUNDLES34
Priessner, Alfons*; Hampl, Nina*,#
ABSTRACT
Electric vehicles (EV) are one major lever in decarbonizing road transportation, particularly if
they are coupled with renewable power. Bundling EVs with photovoltaic (PV) systems and
battery storage (BS) provides a possible solution, but consumer preferences and willingness to
pay (WTP) for such bundles have been limitedly researched. Therefore, we conducted a choice-
based conjoint study with 616 respondents in Austria who have a positive attitude towards EVs
and a purchase intention. Our data shows that the WTP for EV add-on products is still
significantly below the current market price. Further, consumers are willing to pay only a small
premium for the convenience of being served by an all-in-one provider. Moreover, higher EV
subsidies, appear generally to be less valued. Socio-demographic variables have a significant,
but rather small effect on the respondents’ preferences and WTP. Psychological variables, in
contrast, show a significant impact. For instance, technology-minded people are willing to pay
more for EV-PV-BS bundles, and environmentally-conscious respondents are more willing
than non-environmentalists to accept longer amortization periods and lower self-sufficiency
rates; also, they are less sensitive to higher purchase prices and valuable products without a
subsidy incentive. These findings have important implications for marketers and policy makers,
as well as for further research in this field.
Keywords: Electric vehicle, product bundling, renewable energy, conjoint analysis, customer
preference, willingness to pay
Highlights:
• WTP for EV add-on products (PV and BS) is significantly below current market price
• WTP a small premium for EV product bundles offered by an all-in-one provider
• Importance of subsidies decreases by increasing level of incentives provided
• Socio-demographics affect WTP for EV product bundle less than psychological
features
34 This paper is accepted for the 11. IEWT in Vienna, Austria from 13.02.-15.02.2018 and is currently under
revision in Transportation Research: Part A
* Department of Operations, Energy, and Environmental Management, Alpen-Adria-Universität Klagenfurt
# Vienna University of Economics and Business Institute for Strategic Management
Paper 3 105
1. INTRODUCTION
Electric vehicles (EVs) experienced their first hype more than a century ago (New York
Times, 1911), but their rise was prevented mainly by the success of Henry Ford’s Model T
(Kirsch, 2000). The true revival of interest in the EV came at the start of the 21st century when
Toyota launched its first mass-produced hybrid EV, and in Silicon Valley Tesla Motors started
up, producing luxury electric sport cars (Fialka, 2015). Today, EVs that replace vehicles with
fossil-fuel internal combustion engines (IPCC, 2014) are considered one possible lever for
reducing greenhouse gas (GHG) emissions in transportation, which is significant as the
transportation sector in the EU-28 causes almost 26% of all GHG emissions (EEA, 2017).
Although the electromobility transition is gathering pace (IEA, 2018), EVs’
effectiveness in combatting climate change is disputed in the literature (Sandy Thomas, 2012;
Zivin et al., 2012). Some experts argue that certain regions will not experience GHG emission
reduction despite EVs replacing fossil-fuel cars, due to their current non-sustainably produced
power supply (Holland et al., 2015; Zivin et al., 2012). Other regions are predicted to face a
significant long-term increase in demand for “green” power due to the surge in electromobility,
which requires investments additional to existing expansion plans in power generation from
renewable energy sources. For example, by 2030 EVs will account for 6% of all power
consumption in Germany, and by 2050 this share could increase up to 25% (Hacker et al., 2014).
Either way, the proportion of electricity from renewable energy sources used by EVs needs to
be increased to achieve the desired GHG emission reduction (e.g., Bleijenberg and Egenhofer,
2013; Holland et al., 2015).
One possible solution to this sustainability challenge is to purchase EVs in combination
with photovoltaic (PV) solar panels and battery storage (BS) for producing and storing
renewable energy at residential sites. Such product bundles could have a twofold benefit. On
the one hand, these EV-PV-BS product bundles support the reduction of EV GHG emissions
(Delmas, 2018). On the other hand, they could increase EV acceptance (Cherubini et al., 2015)
due to the complementarity of these bundle products (Reinders et al., 2010), which in turn
decrease consumers’ (perceived) risk (Choi, 2003) and increase their convenience (Stremersch
and Tellis, 2002).
Therefore, car manufacturers such as Tesla or Porsche have recently started to sell PV
systems or energy storages (cf. Porsche Holding, 2018; Tesla Motors, 2018). Also, a study
published by the German Federal Association for Solar Economy (Bundesverband für
Solarwirtschaft) in 2018 suggests that nine out of ten potential EV drivers living in a house
Paper 3 106
consider purchasing a PV system once they purchase, or even before they purchase, an EV
(BSW Solar, 2018). These findings emphasize the relevance of products producing renewable
energy for EV stakeholders. Moreover, they reveal the importance of better understanding
potential EV drivers’ preferences and their willingness to pay (WTP) for EV-PV-BS product
bundles.
In fact, a growing literature stream has already been studying the adoption of EVs and
related consumer preferences from different angles (Liao et al., 2017; Rezvani et al., 2015).
Liao et al. (2017) quite recently reviewed EV consumer preference studies and concluded that
(1) EV related attributes, such as infrastructure (e.g., charging points), policy incentives (e.g.,
worldview, pro-environmental attitude, and technology readiness), and EV experience
influence the preferences and WTP for EVs purchased in a bundle with PV or PV and BS.
Against this background, our paper addresses the following research questions: (1) What
is the WTP for specific EV product bundle attributes, and (2) to what extent are consumer
preferences and WTP for EV product bundles influenced by socio-demographic and
psychological parameters, as well as by EV experience? To answer our research questions, we
conducted a web-based survey and conjoint experiment with 616 potential EV drivers in
Austria. Based on this data we could determine customer preferences and the importance of
individual product attributes in consumer choice. We then calculated the WTP for attribute
levels. In addition, we modelled the impact of socio-demographic characteristics, psychological
characteristics, and EV experience on consumer preferences and WTP for EV product bundles.
Paper 3 109
Our paper is structured as follows: in section 2, we give information on our
methodological approach and dataset. Section 3 presents and discusses the results of our survey
and choice experiment, including WTP calculations. Section 4 concludes the paper and
discusses implications for research, marketers, and policy makers, as well as limitations and
areas for further research.
2. METHODOLOGY AND DATA
2.1 Conjoint analysis
Since the objective of our study is to investigate consumer preferences and WTP, we
used conjoint analysis as methodology (cf. Hinnen et al., 2017; Kaenzig et al., 2013). This
method is well suited to evaluate individuals’ preferences for hypothetical, but still realistic
purchase decisions. Developed and introduced by Luce and Tukey (1964) in mathematical
psychology in the mid-sixties of the last century, it has recently gained more and more
importance in a variety of research fields, such as marketing (Green and Srinivasan, 1990) or
entrepreneurship (Brundin et al., 2008). Moreover, conjoint analysis have been widely used in
recent studies investigating preferences for EVs (Beggs and Cardell, 1980; Brownstone et al.,
2000; Bunch et al., 1993; Ewing and Sarigöllü, 2000; Hoen and Koetse, 2014), WTP for EVs
(Hackbarth and Madlener, 2016; Hidrue et al., 2011; Parsons et al., 2014), for EV product
bundles (Fojcik and Proff, 2014; Hinz et al., 2015), or for clean technology product bundles
(Agnew and Dargusch, 2017; Galassi and Madlener, 2016; Ida et al., 2014; Oberst and
Madlener, 2015a).
The most widely applied conjoint design in research and practice is choice-based
conjoint (CBC) (Orme, 2009; Orme and Chrzan, 2017). The theoretical foundation for such
analyses is the classical utility theory which has two assumptions. First, every individual has a
certain utility maximization attitude. Second, every product or service has a certain utility for
each individual, which can be defined as the sum of the part-worth utilities for the various
attributes of this product or service (Lancaster, 1966; McFadden, 1986). Based on this theory,
products can be described by their most important attributes, and individual preferences for
attributes can be indirectly revealed in CBC experiments. In these experiments respondents
have to select their preferred option (dependent variable), from a range of choice objects (in
this study hypothetical EV-PV-BS bundles). Since the respondent repeats such a choice task
several times with varying attribute levels (independent variables), he or she needs to make
trade-offs between desired attributes. From the decisions made in the choice tasks the
Paper 3 110
underlying preference can effectively be elicited in the form of average part-worth utilities for
the attribute levels and relative importance weights for each of the attributes (Green and Rao,
1971; Green and Srinivasan, 1990; Gustafsson et al., 2013; Orme and Chrzan, 2017).
In market research, conjoint analysis studies are considered superior to simply asking
for consumer decision criteria, because people have little insight into their decision-making
rationale, or answers might be influenced by recall bias or other information recovery failures
(Golden, 1992). Additionally, direct answers related to preferences are often biased by social
desirability issues (Gustafsson et al., 2013). Another benefit of this approach is the opportunity
to simulate various product attributes in a controlled experimental setting in order to anticipate
and simulate specific choice contexts. This allows us to distill a number of implications for
policy makers and marketers (Ben-Akiva et al., 1994). Further, conjoint analysis is particularly
useful in immature markets for improving the product design or offering to best satisfy market
demand (Gustafsson et al., 2013; Louviere et al., 2000). Our choice to conduct a conjoint
experiment, was strengthened by our consideration that the products on which our study
focusses are still at the beginning of their diffusion process and currently not even sold in
bundles. Also, the methodological challenges that scholars have raised (e.g., Jaeger et al., 2001;
Louviere et al., 2008; McFadden, 1986), are constantly improving (e.g., Chapman et al., 2009;
Jaeger et al., 2001). Some limitations still assure the mainly exploratory nature of conjoint
analysis, which makes it well suited to investigating our identified research gaps and so
contributing in the area of EV-PV-BS product bundles.
2.1.1 Selection and description of conjoint attributes and levels
Selecting the relevant attributes and levels is the most critical part in a conjoint analysis
and hence must fulfil certain criteria. According to Bergmann et al. (2006) the attributes need
to be (1) relevant to the problem definition, (2) genuine and plausible, (3) understandable to
each respondent, and (4) provided with informative context.
Hence, we choose an elaborated iterative process to identify the most relevant attributes
and levels. To start with, we reviewed literature on EV product bundles (Delmas et al., 2017;
Ensslen et al., 2018; Fojcik and Proff, 2014; Hinz et al., 2015). Delmas et al. (2017) investigated
the joint offering of EVs and PVs, and argued that four parameters (i.e., price reduction, quality
improvements, innovative financing models, and policy subsidies) might positively influence
future demand for such product bundles. We took these suggested parameters together with
other criteria derived from literature on consumer preferences for PV and BS as a point of
departure (Agnew and Dargusch, 2017; Ida et al., 2014; Oberst and Madlener, 2015a). Next,
Paper 3 111
we conducted a qualitative analysis of product offerings on web-pages of major car
manufacturers and PV/BS producers and retailers in the German speaking area. Additionally,
we researched the policy incentive levels for EVs, PV and BS in Austria (BMVIT, 2017;
BMWFW, 2017; Photovoltaic Austria, 2018) 35 . As a next step, we conducted sales
conversations with EV, PV, and BS sellers to get more insight on the extant technological
standards and product benefits. Moreover, in semi-structured interviews with lead users of these
products, we talked about their purchase decision criteria and their WTP for each product
separately and in bundles. These conversations and interviews were conducted between August
and November 2017. Based on this qualitative data we identified a list of relevant attributes and
levels for the CBC design, which we refined once more, and then shortlisted in four expert
discussions with one car retailer, two utility company representatives focused on PV and BS
products, and one consultant focused on renewable energies and future of mobility. Finally, we
verified the interpretation of the attributes and levels in a pre-study with 45 respondents. A
subsample of these were also briefly interviewed to get additional feedback after they had
completed the survey.
To reduce the complexity of the choice experiment with three products (EV, PV, and
BS) for respondents, we decided to ask our interviewees whether they would be interested in
purchasing a PV system with or without BS as add-on, bundled with an EV36. This would draw
attention more to the bundle, than to the parameters of an EV. The latter have already been
researched quite comprehensively (cf. Liao et al., 2017). Further, the CBC literature suggests
limiting the number of independent variables to no more than six (Green and Srinivasan, 1990;
Orme, 2009). Hence, we finally selected six attributes for the CBC experiment, namely PV/BS
add-on (ownership), self-sufficiency rate, amortization period, policy incentive, provider, and
purchase price (see Table 1 for overview). To avoid a number-of-levels effect, the conjoint
design is symmetric with four levels per attribute, except for five levels for the pricing attribute
(Chapman et al., 2009).
The attribute PV/BS add-on (ownership) comprised two parameters. EVs could be either
bundled with a PV standalone or in a bundle with BS (Agnew and Dargusch, 2017). Further,
we included two financing options (Delmas et al., 2017), the first with ownership and no further
payments, and the second a non-ownership/leaser model with further payments , but no initial
35 At the time of the survey, Austria had different financial incentives for BS at the regional level. Since then, at
the federal level, a supplementary investment subsidy has been introduced for BS combined with a PV. In addition,
a national and several regional incentive programs incentivize the purchase of an EV. 36 The car of choice had the same characteristics as the Nissan Leaf 2.0, which was released at the beginning of
2018. The Nissan Leaf model was the world’s best-selling electric car in 2017 (Bloomberg (2017).
Paper 3 112
investment costs, as is reflected in the purchase price. If respondents were not interested in such
add-on products at all, they could still choose a none option, i.e., to purchase an EV only. This
attribute allowed us to assess whether potential EV adopters are interested in PV and BS add-
on products, whether they prefer PVs in a bundle with BS, and which ownership-/financing
option they find most preferable.
TABLE 1. ATTRIBUTES AND ATTRIBUTE LEVELS IN THE CHOICE-BASED CONJOINT DESIGN
Attributes Level 1 Level 2 Level 3 Level 4
PV/BS add-on
(ownership)
PV + BS owner (no
monthly payment)
PV owner (no
monthly payment)
PV + BS leaser with
ownership option
(monthly payment)
PV leaser with
ownership option
(monthly
payment)
Self-sufficiency rate Up to max. 25% Up to max. 50% Up to max. 75% Up to max. 100%
Amortization period 8 years 12 years 16 years 20 years
Provider All-in-one car
dealer/OEMb
All-in-one utility All-in-one specialist
dealer
Diverse specialist
dealers
Policy incentive 0% Up to max. 10% Up to max. 20% Up to max. 30%
a Purchase price has a 5th level at EUR 45,000. b Original equipment manufacturer.
The levels of the attribute power self-sufficiency rate reflect a range from 25% to 100%,
i.e., off-grid with full self-supply of power (Ida et al., 2014; Oberst and Madlener, 2015). Very
high self-sufficiency rates (> 50%) are only feasible with BS (Agnew and Dargusch, 2017).
However, we included the option to install a PV system with a power self-sufficiency rate of
up to 100%, which comprises the possibility of selling excess power back to the grid and so
gaining credit for future power purchases, or of storing the electricity in virtual power storages
provided by, e.g., utility companies. Such offerings are already available in the Austrian market
(KELAG, 2018; Wien Energie, 2018). This attribute allows us to evaluate which level of power
self-supply will be needed to increase acceptance of such a type of product bundle.
The attribute amortization period comprises four levels 8, 12, 16, and 20 years (Oberst
and Madlener, 2015a). Galassi and Madlener (2016), in contrast, fixed this level at 20 years.
With the current market prices for PV and BS, and an electricity price in the midfield of the
European Union (EUROSTAT, 2017), the current average amortization period for PV and BS
add-on products in Austria lies between 15 and 20 years, depending on the level of subsidies
received (cf. KELAG, 2018; Wien Energie, 2018). A survey by Hampl and Sposato (2018)
Paper 3 113
indicates an on average preferred amortization period of 10 years for PV systems in Austrian
households. With this variable we can test what amortization period potential EV adopters
would request for entire product bundles compared to purchasing individual products.
The levels of the attribute provider reflect the range of possible suppliers, from all-in-
one solutions provided by a car retailer, a utility company, or a specialty dealer to a set of
separate dealers providing these products. Richter (2013) already claimed that particularly for
small-scale renewable energy technologies such as PV or BS, an all-in-one business model is
recommendable for successful commercialization over scattered purchases. Such a product
bundle is not on the market yet, except for some pilot offerings by Tesla in the US (2018). But
since all three products are at the beginning of their diffusion curve and there is no clarity on
customer preferences regarding providers, our study specifically aims to assess the importance
of all-in-one providers for EV-PV-BS products. Further, the results will provide insight on the
type of company that the respondents prefer as a one-stop provider.
Policy incentives play a crucial role in EV acceptance (Lieven, 2015; Sierzchula et al.,
2014), and hence also in EV product bundle acceptance. Currently, the Austrian government
offers a broad range of incentives for EVs, PVs, and BS (BMVIT, 2017; BMWFW, 2017;
Photovoltaic Austria, 2018), with some regional differences. Hence, for simplicity, the levels
of this attribute range from 0% to 30% of total subsidies on the purchase price, corresponding
roughly to the subsidy offering in Austria at the time of the survey. By including the attribute
policy incentive, we can determine whether customers are willing to accept higher prices if the
EV-PV-BS product bundle is subsidized, what level of subsidies they consider to be sufficient,
or whether policy incentives are needed at all.
The levels of the attribute purchase price range from EUR 25,000 to EUR 45,000 and
are based on current list prices of the products included in the bundle (KELAG, 2018; Nissan,
2018; Sonnen, 2018). Previous research has shown that the purchase price is considered the
most important decision criterion (Agnew and Dargusch, 2017; Galassi and Madlener, 2016;
Hackbarth and Madlener, 2016; Hidrue et al., 2011; Oberst and Madlener, 2015a). The attribute
allows for estimating customers’ implicit WTP for the different levels of the other product
features, and makes market share forecasts possible in case of further EV, PV, and BS cost
decreases (Orme and Chrzan, 2017).
For this study, our respondents were invited to compare and show preferences in a series
of 12 choice tasks. We created a full-profile design using Sawtooth Software37, thus showing
37 One of the frequently used conjoint analysis software solutions in marketing research (cf. Hinnen et al., 2017;
Kaenzig et al., 2013; Kaufmann et al., 2013; Salm et al., 2016).
Paper 3 114
all attributes at the same time for each product in each choice task. Each of the choice tasks
presented three different product bundle alternatives and a none option (if the person were to
prefer an EV38 without the power-supply add-on products) from which the respondents had to
choose their preferred option. An example of a choice task is illustrated in Figure 1.
FIGURE 1. SAMPLE CHOICE TASK
2.1.2 Estimation algorithm
For data analysis we estimated individual part-worth utilities using a Hierarchical Bayes
(HB) model (Rossi and Allenby, 2003) implemented in Sawtooth Software. Recent studies
show that the results from an HB and traditional mixed-logit model are very similar (Chassot
et al., 2014; Hampl and Loock, 2013; Salm et al., 2016). The HB model has the advantage of
measuring preferences both on an individual level and, as is traditional, on an aggregated level.
By doing so, HB acknowledges the heterogeneity in consumer preferences. This is possible due
to the HB algorithm’s “hierarchical” nature, which means that HB consists of (1) a lower and
(2) an upper level (cf. Gamel et al., 2016; Kaenzig et al., 2013; Sawtooth Software, 2009). At
the lower (i.e., individual) level, the general assumption is that the probability of the ith
38 The parameters of the EV in each choice task were fixed at a 400 km range, 150 horse power, 40-60 minutes
per full-charging, which are the characteristics of the Nissan Leaf 2.0 (Nissan (2018).
Paper 3 115
individual choosing one option (k) among a set of options (j) is determined by a multinomial
logit model which can be described as follows (Sawtooth Software, 2009):
𝑝𝑘 = exp(𝑥′
𝑘𝛽𝑖)
∑ exp(𝑥′𝑗𝛽𝑖)𝑗
pk … probability that an individual i chooses the kth alternative in a given choice task.
xj … a vector of values describing the jth alternative in that choice task.
ßi … a vector of part worths for the ith individual.
At the upper level, the individual responses are pooled by assuming that the individuals’
part worths are described by the multivariate normal distribution ßi ~ Normal(α, D), where the
part-worth utilities (ßi) of the ith respondent are distributed with a vector of means α and a matrix
D of variances and covariances of the distribution of part worths across individuals (Sawtooth
Software, 2009).
The model parameters are derived from an iterative process applying a Monte Carlo
Markov Chain algorithm (cf. Gamel et al., 2016; Kaenzig et al., 2013; Sawtooth Software,
2009). To ensure convergence of the parameters we followed the approach proposed by
Sawtooth Software that recommends deleting the first 10,000 draws as burn-in of a total of
20,000 draws per respondent (Sawtooth Software, 2009). For a more detailed description of the
iterative estimation process of the parameters see (Sawtooth Software, 2009).
2.2 Measurement of socio-demographic and psychological parameters
In order to test the influence of socio-demographic and psychological characteristics on
private individuals’ preferences for EV-PV-BS product bundles, a set of variables (Table A.1
in Appendix) were used which were derived from the literature review in section 1. We
measured the variables in the course of the questionnaire accompanying the CBC experiment,
also using Sawtooth Software. On the one hand, socio-demographic variables such as gender,
age, educational level, income, and housing situation (apartment vs. house) were elicited. In
addition, we requested the respondents to indicate their experience with EVs. Answers were
given on a 4-point Likert scale with values ranging from (1) I own an electric car / owned an
electric car to (4) I have no EV experience at all.39
39 This variable was recoded in the covariate model to assess the impact of more EV experience.
Paper 3 116
On the other hand, the survey comprised statements designed to measure each
respondent’s cultural worldview, pro-environmental attitude, and technology readiness (cf.
Priessner et al., 2018). Following Cherry et al. (2014), we used eight items to measure the
cultural worldviews. This abridged version of the scale, which was originally developed by
Kahan et al. (2007), includes statements such as “The government should do more to pursue
social goals, even if it means restricting the freedom and choice of the individual”
(individualism-communitarianism) or “Our society would be better off if the distribution of
wealth were more equal” (hierarchism-egalitarianism). Answer options were presented on a 5-
point Likert scale ranging from (1) strongly disagree to (5) strongly agree. The scale had a
reliability score of α = 0.70 (“communitarian worldview”) and α = 0.56 40 (“egalitarian
worldview”). To test the influence of pro-environmental attitude (α = 0.69) we applied the scale
of Whitmarsh and O'Neill (2010) relying on four items. The scale includes items such as “Being
environmentally friendly is an important part of my personality”, which had to be rated on a 5-
point Likert scale ranging from (1) strongly disagree to (5) strongly agree. Technology
readiness (α = 0.85) was operationalized as participants’ agreement on a 5-point Likert scale,
ranging from (1) strongly disagree to (5) strongly agree), using eight statements from the
Technology Readiness Index (Parasuraman, 2000; Parasuraman and Colby, 2015) related to a
general attitude toward technology, such as “Technology gives people more control over their
daily lives.” Responses to the items of the cultural worldviews scale were averaged per
dimension so that, e.g., a higher score on the individualism-communitarianism questions
indicates a more communitarian worldview. The aggregation of responses for pro-
environmental attitude and technology readiness follows a similar logic, i.e., respondents with
higher values are perceived to have a more positive environmental attitude and a higher
technology readiness (see Table A.3 in the appendix for a summary and details on the
psychological variables).
2.3 Sample
The target population of this survey consisted of Austrians aged between 18 and 75
years who indicated an intention to purchase an EV within the next decade. Therefore, in our
sampling process we applied two filter questions at the beginning of the questionnaire. First,
respondents needed to indicate their attitude towards EVs on a scale from (1) very negative to
40 The egalitarian worldview was measured in the survey but was not included in the final model due to minimal
effect on potential EV adopters’ preferences and its low reliability score in the Austrian setting.
Paper 3 117
(4) very positive.41 A positive attitude ((3) positive and (4) very positive) is a good predictor of
both an EV purchase (Nayum et al., 2016) and the general acceptance of innovative products
(Jhang et al., 2012).
Second, the respondents needed to indicate the timeframe in which they intended to buy
an EV. Several research papers filtered potential EV drivers on the intention to purchase an EV
as their next car, but did not qualify the purchase intention related to the time horizon (Axsen
et al., 2015; Hackbarth and Madlener, 2013; Mabit and Fosgerau, 2011). However, the EV
market is still a very small niche in that 1.6% of all cars newly registered in the first half of
2018 in Austria were EVs (STATISTIK AUSTRIA, 2018a), and the uptake rates in Europe are
increasing, but slower than expected (IEA, 2018). Further, the average car age in Austria is 9.1
years (European Car Manufacturer Association, 2016). Therefore, we considered people willing
to purchase an EV in more than 10 years or without planning possibly to purchase an EV, as
not likely to purchase an EV as their next car. Hence, we excluded them from our target group
sample.
The respondents for the survey were recruited by the professional market research
company market in Spring 2018. They used their online panel pool of more than 20,000 active
users in Austria to invite interviewees via e-mail. Using participants from a panel pool has the
benefit of the pool’s experience with longer surveys and with choice experiments. Hence, we
could ameliorate Jaeger et al.’s (2001) criticism that CBCs get more accurate results if
participants are accustomed to CBCs due to training effects. The sample was drawn by quota
sampling, considering the distribution of gender, target population by federal state, and age. A
total sample of 1,251 survey participants were invited, of which 660 fulfilled both selection
criteria. We cleaned the sample by removing 44 speeders42 and flatliners43 (see filter funnel in
Figure 2).
Given the fact, that the market research company performed an iterative process in
recruiting respondents to ensure that the sample fulfills the predefined criteria for
representativeness, we cannot report a response rate, i.e., a ratio of participants of the survey
over the total number of potential interviewees approached. Consequently, the data might
41 For those indicating no attitude or preference towards EVs, we included an additional answer option: “I do not
know / I cannot say.” 42 Respondents who were among the fastest 10 percent in reading the instructions of the CBC experiment (less
than 20 seconds, mean 65 seconds), and who completed the CBC experiment among the fastest 5 percent of
respondents (less than 77 seconds, mean 158 seconds). 43 The average root likelihood (RLH) can be used as a measure of fit to assess data quality. In this study, as each
choice task presented four alternatives, the RLH predicts that each alternative would be chosen with a probability
of 25% (corresponding RLH of 0.25). All answers below 0.25 counted as “flatliners.”
Paper 3 118
comprise some non-response bias. Nevertheless, as described in Table A.2 in the Appendix, the
total sample represents the average Austrian population quite well in terms of gender, age, and
federal state. Only the final sample used for data analysis differed in income distribution, having
a higher proportion of better earning respondents. Further, people with a university education
seem to be somewhat over-represented in the sample, but similar differences with respect to
education level have been found in other EV studies (Axsen et al., 2016).
FIGURE 2. FILTER LOGIC FROM TOTAL SAMPLE (N = 1,251) TO FINAL SAMPLE (N = 616)
3. RESULTS AND DISCUSSION
In this section, we discuss the results of the conjoint analysis: (1) the relative importance
scores of each attribute, and (2) the part-worth utilities per attribute level. Subsequently, we
estimate the WTP for the features of an EV-PV-BS product bundle. In the last sub-section, we
present and discuss the results of a model (part-worth utilities and WTP) comprising covariates,
i.e., socio-demographic and psychological variables, as well as EV experience.
3.1 Relative importance of conjoint attributes
Our results are based on data from 616 future EV drivers with a positive attitude toward
EVs and an intention to purchase an EV within the next ten years. Each respondent conducted
12 choice tasks which leads to a total of 7,392 choices. One result from the CBC analysis gives
the relative importance scores of the different attributes, describing the size of each attribute’s
Paper 3 119
influence on the purchase decision (in our case, the decision to purchase an EV-PV-BS product
bundle). The relative importance scores are calculated by subtracting the highest and the lowest
part-worth utility within each attribute, and then standardizing these values to a sum of 100%
across attributes.44 Table 2 displays the average importance scores.
The most important purchase decision criterion is the purchase price at 30.6%, which
is in line with other EV, PV, or BS studies (Agnew and Dargusch, 2017; Galassi and Madlener,
2016; Hackbarth and Madlener, 2016; Hidrue et al., 2011; Oberst and Madlener, 2015a).
Ranked second, is the PV/BS add-on ownership model at 18.7%. Shih and Chou (2011) already
argued that the higher people’s concerns about an investment in renewable energy technologies
(e.g., regarding reliability, policy subsidy, electricity price, development of new technologies),
the more they value short-term, expensive non-ownership (i.e., leasing) contracts. The criterion
in third position (power self-sufficiency) measured at 16.4% ranked closely to the fourth
(amortization period) which measured at 15.0%. This order could suggest, that becoming a
prosumer (i.e., producing and self-consuming power) is slightly more important to potential
adopters than making a fast amortizing investment, as has also been indicated by Oberst and
Madlener (2015). Interestingly, policy incentives and the type of provider are of minor
importance at respectively 11.2% and 8.1%. This result allows the conclusion that lead users of
an EV product bundle find who provides their products to be negligible. Galassi and Madlener
(2016) already showed that the sales channel is of least importance in the decision to purchase
PV and BS bundles. Further, policy incentives seem to be no major purchase motive, as other
studies on clean technologies have also shown (Zhang et al., 2013). Additionally, Sierzchula et
al. (2014) pointed out that although policy incentives are correlated with the increase in market
share of EVs, they cannot ensure high EV adoption rates. Therefore, for them governmental
incentives are important in the early stage of the diffusion curve.
3.2 Part-worth utilities of attribute levels
The analysis of CBC data provides information on the average impact a particular
attribute level can have on the respondent’s decision to purchase an EV-PV-BS product bundle.
Table 2 displays the average raw utilities (coefficient estimates) of the HB model with the
corresponding standard deviations and confidence intervals. Each value represents the change
in utility of the total product when altering one of the attribute levels while keeping all others
44 The derived importance scores are dependent on the selected attributes and the definition of the attribute levels
(Orme and Chrzan (2017). For instance, illustrating policy incentives in exact EURs instead of relative cost saving
percentages may increase the importance of policy incentives (cf. Kaenzig et al. (2013).
Paper 3 120
equal. All values are zero-centered; thus, they sum up to zero within each attribute. This implies
that a positive coefficient increases and a negative coefficient decreases the utility of that
particular attribute level. The higher the value, the stronger the influence of the specific attribute
level on the purchase decision. The utility values are highly dependent on the selected range of
attribute levels. Therefore, it is only meaningful to compare utility values between different
levels of a given attribute (Orme, 2009). By converting part-worth utilities into monetary values
(see section 3.3. below on WTP) this scale effect can be eliminated, which then enables cross
attribute comparison (Orme, 2001). The none-option score stands for the utility potential EV
drivers gain if they do not choose any of the product bundles shown to the them. Hence, it can
be read as an investment threshold which needs to be exceeded by the sum of the utility values
for the attribute levels of the respective product bundle to trigger a potential purchase (Orme
and Chrzan, 2017).
As an indicator of data quality and measure of fit we used the average root likelihood
(RLH) (Orme and Chrzan, 2017). In this study, we assume that each alternative would be
chosen with a probability of one quarter (i.e., the RLH threshold is 0.25). The RLH was 0.65,
which indicates a good model fit (i.e., our model is 2.6 times better than the random chance
level).
TABLE 2. PART-WORTH UTILITIES OF THE DIFFERENT ATTRIBUTE LEVELS FOR THE DECISION TO
a Coefficient estimates are equal to the posterior population means across the saved draws, interval-scaled and
zero-centered within attributes. b Mean relative importance scores per attribute (m) and corresponding standard deviation (SD) in parentheses.
The importance scores sum up to 100%. c Confidence interval.
3.3 Willingness to pay for attribute levels
Conjoint analysis also enables conversion of the part-worth utilities to aggregated
monetary WTP values (Green and Srinivasan, 1990; Orme, 2010). This approach is commonly
applied in clean technology research with slightly different calculation methods and denotations
(Hackbarth and Madlener, 2016; Ida et al., 2014; Kaufmann et al., 2013; Salm et al., 2016).
Paper 3 122
In this study, we describe the WTP relative to a product bundle default option (i.e., EV
+ PV non-ownership, 25% sufficiency rate, 20 years amortization period, diverse specialist
dealers, 0% policy incentive). Our calculations aim to illustrate the willingness to pay a
premium for more desired product features. Thus, the WTP formula is determined as follows
(cf. approach in Salm et al., 2016):
𝑊𝑇𝑃 (𝑢𝑖𝑗) = (𝑢𝑖𝑗 − 𝑢𝑖𝑗 𝐷𝑒𝑓𝑎𝑢𝑙𝑡 ) ∗ 𝑝𝑚𝑎𝑥 − 𝑝𝑚𝑖𝑛
𝑢𝑝𝑗 𝑚𝑎𝑥 − 𝑢𝑝𝑗 𝑚𝑖𝑛
This approach involves calculating the difference between the part-worth utility (uij) of one
attribute level (j) (e.g., 8 years) and the default part-worth utility (uij Default) (i.e., 20 years) within
the same attribute (i) (e.g., amortization period). This difference is then multiplied by the price
of one utility unit (i.e., difference between the highest (pmax) and lowest (pmin) possible price)
divided by the utility difference between the highest and lowest price (upj max – upj min) (Orme,
2010). The results of these WTP calculations are displayed in Figure 3.
Before interpreting the WTP results, we want to refer to the explorative nature of our
analysis. Our study did not aim to calculate precise WTP values. We merely intended to test
the joint effects of different attributes and levels in an experimental setting which would allow
us to derive WTP estimates. Compared to a direct questioning approach, the indirect preference
measurement of a conjoint analysis has the advantage of overcoming biases such as social
desirability, which occur particularly commonly in decisions related to environmental issues
(Diekmann, 2017). Moreover, CBC is considered very suitable for testing preferences and WTP
for hypothetical products (Gustafsson et al., 2013). Still, a CBC design remains experimental
in that the respondent does not actually have to pay the price he or she indicates to be willing
to accept. Such a setting results in a gap between hypothetical and real WTP, referred to as a
“hypothetical bias” (List et al., 2006; Orme, 2001). In addition, actual WTP in real life depends
on several other aspects, such as status-quo bias (Samuelson and Zeckhauser, 1988) or
competition (Orme, 2001).
Considering the above, some scholars argue for complementing conjoint-based WTP
with additional data from incentive-compatible procedures, such as the Becker-DeGroot-
Marschak method (BDM) or actual point-of-purchase contexts that would reveal more accurate
WTP values (Wertenbroch and Skiera, 2002). However, despite being of high managerial and
research interest, EV-PV-BS product bundles are not available on the market yet. Therefore, in
the absence of sufficient BDM or actual purchase data, we have to rely on conjoint data. Further,
Paper 3 123
Miller et al. (2011:172) found that despite hypothetical bias, WTP values calculated on the basis
of conjoint data “still lead to the right demand curves and right pricing decisions.” Overall, the
WTP values presented here should be interpreted as upper boundaries and hence should be used
carefully.
One major insight this analysis brought, relates to the gap between WTP for EV add-on
products and current market prices (reference date October 2018). We noted an average WTP
for an increased self-sufficient energy supply (from 25% to 100%) at approximately EUR
9,500. In comparing the estimated WTP figures to current market prices for PV and BS in the
Austrian market(KELAG, 2018; Wien Energie, 2018)45, we identified a potential gap of 15 to
30%46. Nevertheless, assuming further cost curve effects of battery price47 toward USD 190 by
the end of this decade (and below the desiderated level of USD 100 per kWh by 2030)
(McKinsey, 2017) and further cost decreases for PV systems (the cost per watt has reduced by
11.4% per year, on average, since 197048 (Seba, 2014)), the market for EVs and their add-on
products is bound to become economically viable in the next decade (Seba, 2014).
Another interesting insight relates to the linearity of the WTP regarding improvements
in the self-sufficiency rate and amortization period. Following a clear order, the shortest
amortization period (i.e., 8 years) and the highest self-sufficiency level (i.e., 100%) have the
highest monetary value. Further, our WTP analysis reveals a similar WTP for the fastest
amortization and highest self-sufficiency attributes. Oberst and Madlener (2015) similarly
found that customers desire the highest self-sufficiency rate, as well as the fastest payback
periods for PV systems, as did Agnew and Dargusch (2017) for BS. Both the latter papers,
however, identified slightly diminished utilities toward the optimum value, which our research
did not. However, this is not considered unusual when conjoint analysis is applied in product
design research (Orme, 2010). Interestingly, there is a conflict of interest between these two
attributes, because with the current product offering one cannot achieve both optima; this is
only possible if virtual product storages are implemented (KELAG, 2018; Sonnen, 2018).
Regarding policy subsidy, potential investors assigned the highest WTP to the maximum
subsidy level, which is unsurprising. However, irrespective of the subsidy level, we noted that
the indicated WTP amounts are below the expected value of the incentive levels (e.g., 30% of
45 On average the PV + BS cost for a household with an annual electricity consumption of 5,000 to 7,000 kWh lies
between EUR 15,000 and EUR 19,450 (reference date October 2018). 46 This WTP gap is the “best case” since the calculated WTP figures should be interpreted as upper limits. 47 “From 2010 to 2016, battery pack prices fell roughly 80% from ~$1,000/kWh to ~$227/kWh” (McKinsey, 2017:
p.10) 48 In 1970 the cost per watt generated by a PV system was USD 100, which decreased to about 33 cents per watt
in 2014 (Seba, 2014).
Paper 3 124
EUR 45,000 = EUR 13,500 vs. EUR 6,011 WTP, equals 45% of real subsidy value; or 10% of
EUR 45,000 = EUR 4,500 vs. EUR 2,960 WTP, equals 66% of real subsidy value). This shows,
that with lower subsidy levels (e.g., 10%) the WTP is closer to the expected value of the subsidy
than with higher subsidy levels (e.g., 30%). Several studies in clean technology literature that
refer to, e.g., solar boilers, PV, energy conservation, co-generation of heat and power, and EVs,
indicate that policy subsidies hardly influence potential adopter decisions, and hence are often
under-valued (Kemp, 2000; Zhang et al., 2013). Fischer and Newell (2008) ranked the
effectiveness of subsidies for renewables as the second least effective measure out of six policy
options, but also added, that “an optimal portfolio of policies achieves emissions reductions at
a significantly lower cost than any single policy” (p. 142). Therefore, in line with Fischer and
Newell (2008), we conclude that policy makers should use some level of subsidy for products
that do not have an existing market, but then embedded in a set of policy incentives (including
taxes or regulations).
Concerning preferences regarding the provider, we find that switching from different
specialty dealers to certain all-in-one providers results in an increase in the WTP of between
EUR 1,693 and EUR 2,536. This suggests that respondents see value in purchasing all products
from one provider, but are not willing to pay a high premium for it. The results also indicate
that future EV drivers are largely indifferent about particularly who provides their product
bundle. While car manufacturers, utilities, and specialty dealers might all be interested in
offering such product bundles, the difference in WTP between these potential all-in-one
providers is rather low. The implications of this finding will be discussed in detail in section
4.2.
Lastly, we drew two specific conclusions from the findings on WTP for the attribute PV
+ BS add-on ownership model. First, our WTP data suggests that consumers would preferably
purchase PV and BS in a bundle, irrespective of whether they own or lease the bundle. This
seems to confirm the theory that bundling complementary products generates benefits for which
users, to a certain extent, are also willing to pay a premium (Reinders et al., 2010). Second,
respondents seem to differentiate between owning and leasing the add-on products in their
WTP. We notice a higher WTP for owning the PV and BS than for the non-ownership option.
An explanation here could be that people who prefer the non-ownership model experience
strong uncertainty regarding investment parameters such as price, product reliability,
technological development, or subsidy level, and hence are willing only to pay less for such
investments in total, and vice versa (Shih and Chou, 2011).
Paper 3 125
Note: Attribute levels of the default product (PV leaser with ownership option, 25% sufficiency rate, 20 years amortization
period, diverse specialist dealers, 0% policy incentive) are marked with an asterisk (*).
FIGURE 3. WTP FOR ATTRIBUTE LEVELS OF EV-PV-BS PRODUCT BUNDLE (RELATIVE TO
DEFAULT)
3.4 Impact of socio-demographic and psychological characteristics on part-
worth utilities and WTP
One of our research objectives is to test the influence of future EV drivers’ socio-
demographic and psychological characteristics and EV experience on their preferences and
WTP for EV-PV-BS product bundle attributes. Hence, we included five socio-demographic
variables (age, gender, income, education, and housing situation), three psychological variables
(communitarian cultural worldview, pro-environmental attitude, technology readiness), and
EV experience as covariates in the model (see Table 3 for details). A similar approach has been
successfully applied by Gamel et al. (2016) who analyzed the impact of age, asset valuation,
and environmental attitude on private individual wind power investment preferences.
According to Orme and Howell (2009), models including covariates could provide
additional information about respondents’ preferences, and hence could improve the share of
preference predictions through more accurate parameter estimates. Our RLH score improved
from 0.65 (cf. base model in section 3.2) to 0.73 (covariates model), which is considered quite
substantial (cf. Orme and Howell, 2009).
Paper 3 126
As recommended in the literature and to facilitate interpretation of the parameter
estimates, we zero-centered the covariates before adding them to the HB model (Orme and
Howell, 2009). Consequently, the mean value of each variable is zero and positive values
indicate an above-average score (e.g., income) or above-average agreement with the statements
measuring psychological constructs (e.g., cultural worldviews).
The first column in Table 3 displays the intercept, which equals the part-worth utility
value of an attribute level when all parameters included in the model are set to the mean (i.e.,
zero in our case). The formula for calculating the individual part-worth utility Betax for any
attribute level x follows the following logic (cf. Gamel et. al, 2016):
*Significant at the 0.05 level (parameter estimates are significantly positive/negative if more than 95% of the estimated parameter values in each iteration of the algorithm are positive/negative). † Significant at the 0.1 level (parameter estimates are significantly positive/negative if more than 90% of the estimated parameter values in each iteration of the algorithm are positive/negative).
Paper 3 132
4. CONCLUSION
To contribute to the global targets for reduced carbon emission, EVs need to be powered
with electricity produced from renewable energy sources. Consequently, a better grasp of EVs
bundled with PV and BS systems is highly relevant for promoting the diffusion of EVs in
individual transportation in a way that will limit environmental harm. This paper has aimed to
advance our current understanding of such EV-PV-BS product bundles by investigating private
individuals’ preferences and their WTP for such bundles. Further, our study aimed at shedding
light on how customers’ assessing values are influenced by socio-demographic and
psychological variables, as well as by self-assessed EV experience. We built our analyses on a
unique dataset of 7392 experimental choices of 616 respondents in Austria.
Our results show that the expressed WTP for EV add-on products (PV and BS) is still
fixed on amounts below current market prices. At the same time, the attribute purchase price
(30.6%) is the most important in directing a purchase decision, followed by PV/BS add-on
ownership (18.8%), and self-sufficiency rate (16.0%). Moreover, potential EV drivers have
some willingness to pay a premium for purchasing an EV-PV-BS bundle from an all-in-one
provider. Our study found influencing effects of socio-demographic and psychological
variables, and of self-assessed EV experience on product bundle preferences, as well as on the
WTP. However, our analysis showed that socio-demographic characteristics have relatively
minor effects on the potential EV drivers’ bundle preferences and their WTP, whereas the
psychological variables have a stronger effect. These findings have implications for further
research, as well as for practitioners and policy makers, which we discuss below.
4.1 Implications
Our study makes a two-fold contribution to EV literature and provides several avenues
for further research. First, following a call for EV product bundling research by Cherubini et al.
(2015), our paper is the first to analyze the WTP for the EV add-on products PV and BS in a
product bundle. In doing so, we show that the prices of current market PV and BS offerings do
not match the WTP of potential customers. However, if we assume an ongoing decrease in the
PV and BS cost curve, as described in section 3.3, we are likely to achieve grid parity within
the decade (Seba, 2014), which will make these add-on product offerings more appealing for
future EV drivers. Further studies can build on our insights and try to simulate the uptake in
market share by cost degression of EV, PV, or BS. Scholars could also investigate innovative
ownership models (Galassi and Madlener, 2016) in order to decrease initial investments. Even
though the average respondent in our sample prefers the ownership instead of the leasing option
Paper 3 133
related to PV and BS systems, we note some heterogeneity in preferences depending on the
socio-demographic and psychological profiles of the respondents. This finding could be of
interest in more detailed analyses.
Second, our study is the first to show that socio-demographic, psychological, and
experiential variables influence potential consumers’ preferences and WTP for EV product
bundles and their features. This is in line with findings in earlier EV literature (Bühler et al.,
2014; Nayum et al., 2016; Plötz et al., 2014). Related to socio-demographic variables, our
results indicate that more educated people have a better sense for currently feasible self-
sufficiency rates and amortization periods for PV and BS systems. One often discussed measure
for increasing awareness of renewable energy technology and investment criteria refers to
information and education campaigns (e.g., Islam, 2014; Islam and Meade, 2013). However,
the question remains whether such campaigns are effective in the context of the kind of complex
product bundles on which our study focuses. Thus, future research could investigate the effect
different measures have in supporting households’ investment decisions related to EVs and add-
on products. Such measures could range from awareness campaigns to more sophisticated
instruments such as online tools and apps that inform about relevant investment parameters.
Further, we find that the price people seem willing to pay for EV-PV-BS product
bundles is higher, when they have more EV experience, a pro-environmental attitude, or a more
technologically-ready mindset. Respondents with more EV experience are not only willing to
pay more for such product bundles, but also demand higher self-sufficiency rates from the add-
on products in focus, to facilitate their roles as prosumers. Future research could further
investigate the effect of experience, not only related to EVs but also related to PV and BS
systems (Agnew and Dargusch, 2017), on the willingness to adopt EV-PV-BS product bundles
or other EV add-on products. Another interesting finding of our analysis is that people with a