Transport Policy: Joachim Globisch, Patrick Plötz, Elisabeth Dütschke, Martin Wietschel T
Transport Policy: Joachim Globisch, Patrick Plötz, Elisabeth Dütschke, Martin Wietschel T
Transport Policy: Joachim Globisch, Patrick Plötz, Elisabeth Dütschke, Martin Wietschel T
Transport Policy
journal homepage: www.elsevier.com/locate/tranpol
Keywords: The lack of public charging infrastructure is often referred to as an important barrier to the diffusion of electric
Public charging infrastructure vehicles. As the construction of charging stations is a costly endeavour, the question arises as to how maximum
EVSE benefit for potential users can be achieved with limited resources. Therefore, our analysis deals with the factors
User perspective that influence the attractiveness of public charging infrastructure from the perspective of potential users. Our
Electric vehicle
analysis is based on the assessments of 1003 German car drivers on possible future charging infrastructure
Willingness to pay
Target groups
systems with different configurations regarding spatial coverage, charging duration and usage costs. We ex-
amined the preferences with regard to these features using a rating-based conjoint analysis. We also looked into
the question of whether groups of car drivers can be identified that are characterised by specific preference
constellations with regard to these features. Our key finding is that the majority of car drivers are unwilling to
pay a basic fee for the possibility of using public charging infrastructure. Nevertheless, there are subgroups that
value the public charging infrastructure more than other car drivers. In addition to implications for possible
business models, this result indicates that public charging infrastructure could be important for attracting other
target groups to electromobility besides classic early adopters of electric vehicles. Furthermore, our analysis
shows that the charging duration at charging stations in cities and along the highway has a strong influence on
the evaluation of the public charging infrastructure. The spatial coverage with charging stations in cities and
along the highway, on the other hand, has a weaker influence. A central conclusion from this is that the existence
of fast-charging stations should be prioritized over a close-meshed coverage with charging points when the
charging infrastructure is expanded.
1. Introduction more willing to make a detour than to accept waiting times for char-
ging. This raises the question of how many fast charging stations are
Battery electric vehicles (BEV) can reduce greenhouse gas emissions actually needed and to what extent electric vehicle drivers are willing
if powered with renewable energy. A barrier to their market diffusion is to pay for their presence or use? These questions are especially im-
the current low driving range delivered by batteries. Although it is portant as fast charging infrastructure requires large investments
possible to find user groups who can meet their driving needs without (Schroeder and Traber, 2012). At the same time, the question arises
public charging (see e.g. Jakobsson et al. (2016)), a broader market how much charging infrastructure needs to be set up in order to ef-
diffusion of BEVs requires either a noteworthy improvement in battery fectively support the spread of electric vehicles (Hardman et al., 2018).
technology or a more extensive charging infrastructure or progress in Thus, our study focuses on the following main research question: Which
both. factors influence private car drivers’ evaluation of public EVSE? Other
Sun et al. (2016) point out that there is comparatively little research questions linked to this are: What attributes of public EVSE are most re-
done on user perceptions of and preferences for charging in the field of levant to car drivers? Do different target groups have distinct preferences
user perception and acceptance of electric vehicles (EVs). This is sur- regarding EVSE? Our results on the latter question are particularly useful
prising as the need to charge the vehicle is one of the main differences for researching target groups for charging infrastructure, as there is a
to driving a vehicle with an internal combustion engine. Philipsen et al. research gap in this field (Sovacool et al., 2017).
(2016) come to the conclusion that electric vehicle drivers and people To address these questions, we conducted an empirical study with a
interested in electric vehicles want fast charging stations on motorways sample of 1003 German car drivers. Our analysis employs a hierarchical
and in public places. Their results also indicate that respondents are linear model to analyse data collected by a rating based conjoint
∗
Corresponding author.
E-mail address: joachim.globisch@isi.fraunhofer.de (J. Globisch).
https://doi.org/10.1016/j.tranpol.2019.05.017
Received 10 October 2018; Received in revised form 26 May 2019; Accepted 28 May 2019
Available online 06 June 2019
0967-070X/ © 2019 Elsevier Ltd. All rights reserved.
J. Globisch, et al. Transport Policy 81 (2019) 54–63
analysis to quantify the factors that influence the respondents’ evalua- The actual need for public EVSE is important as the scale of EVSE
tion of (hypothetical) public EVSE options. The outline of this paper is deployment is crucial for the direct and indirect costs of the system. The
as follows. The remainder of section 1 provides an overview of the economic dimension is addressed e.g. by Guo et al. (2016) who look at
literature that forms the basis for our study. Section 2 describes the the business perspective and investment planning for charging station
methodology. Section 3 presents the results of our analysis. The main providers. Similarly, Sadeghi-Barzani et al. (2014) look at how to
findings and conclusions are discussed in section 4. minimize the total cost of charging station investment including grid
In general, the topic of EVSE has been approached from different costs. The indirect costs of EVSE are analysed by Wang et al. (2013)
perspectives, in particular by studies analysing techno-economical as- who look at the distribution system with the objective of minimizing
pects and those that focus on the preferences of (potential) users of power losses and voltage deviations.
public EVSE. Both approaches examine which kind of EVSE is needed The cost of public EVSE leads to the question of who is willing to
where. Following these functional aspects, the question arises of the pay (how much) for its construction and use. It seems reasonable to
costs of EVSE and/or the users’ willingness to pay as the deployment of assume that the willingness to pay is influenced by the usefulness at-
EVSE has to be financed in some way. tributed to EVSE by individuals. Thus, the question arises whether there
In relation to the requirements and preferred locations for charging are target groups with higher or lower evaluations of EVSE or distinct
stations, gender, age and prior experience with battery electric vehicles requirements regarding EVSE.
were found to be important (Philipsen et al., 2016). The duration of the The results of the reviewed studies provide useful insights that will
recharging process as well as the location and density of EVSE may guide our own analysis and can be summarized as follows: The density
affect the (perceived) usefulness and costs of EVSE (Philipsen et al., of an EVSE grid is considered to be an important factor when evaluating
2016; Schroeder and Traber, 2012). In addition, Nicholas and Tal its usefulness. Density is not only defined by geographical distance – the
(2017) find that fast charging EVSE is used by EV drivers without re- time needed for a detour to a charging point is also important. Thus, the
charging opportunity at home as a replacement for home charging. location of EVSE can also be important for its utility as it defines the
Furthermore, Philipsen et al. (2015) conclude that a detour of 5 km or ease of its accessibility. The relevance of detour time and the few stu-
10 min to a fast charging station is seen as acceptable. Sun et al. (2016) dies that focus on fast charging also indicate that the duration of the
from Beijing, China, study the willingness to take a detour for rechar- charging process might be relevant for EVSE utility.
ging, focusing on the opportunity to fast-charge an EV. They find an
average willingness to detour of up to about 1750 m for private car 2. Methods
drivers on working days and 750 m on non-working days based on the
recorded mobility patterns of Japanese EV drivers. However, their The main aims of our study were to obtain insights about the trade-
sample was limited (24 drivers) and the results are certainly influenced offs between different attributes of EVSE and identify target groups. We
by the current state of existing infrastructure, i.e. do not necessarily conducted an explorative and inductive analysis. Likewise, the content
mirror the ideal situation desired by drivers. Sun et al. (2017) surveys of our data collection is guided by the results of the literature review.
actual EV users and highlights that nearly half the respondents want to The resulting research design and data collection are described in sec-
be able to find a charging station within a 5-min drive if they need to tion 2.1. Section 2.2 outlines how the collected data was prepared and
charge their car. Only 16% find it acceptable to drive for 10–20 min to a analysed.
charging station. Density of charging stations is also found to be sig-
nificantly correlated with EV satisfaction. Furthermore, Sierzchula et al. 2.1. Research design
(2014) identify EVSE density as positively correlated with PEV sales.
These user-centric analyses of preferred locations and density of Sampling procedure. Potential EVSE users are the most relevant po-
EVSE are mirrored by techno-economic studies. A common approach in pulation for our study. We defined potential EVSE users as persons who
such studies is to maximize the number of electric miles travelled or regularly use a passenger car (at least once a week) that is owned by
reduce the number of unfulfilled trips if all vehicles were BEV (Alhazmi themselves, a member of their household or a family member.
et al., 2017; Dong et al., 2014; Shahraki et al., 2015). For example, Therefore, car ownership and regular car use were the two basic re-
Alhazmi et al. (2017) optimize the allocation of EVSE with respect to quirements to participate in the survey. The requirement of car own-
electrifying as much car travel as possible. Dong et al. (2014) also op- ership aimed to ensure that the respondents represent potential EV
timize EVSE locations and analyse the number of range-constrained adopters. The precondition of regular car use was used as a proxy for
days and trips for the greater Seattle region. Shahraki et al. (2015) potential relevance of EVSE in the case of a (future) EV adoption.
perform a similar analysis applied to taxis in Beijing. To obtain a sample of the above described population, we tasked the
While the aforementioned studies focus on how the usefulness or fieldwork to a market research institute and defined screening criteria
utility of public EVSE can be maximized, it is important not to neglect and quotas for the sample. The screening criteria reflected the two re-
the question to what extent public EVSE is needed at all. For example, quirements that constitute the definition of the relevant population. The
the sample of Sun et al. (2017) consists mostly of EV drivers who quotas aimed to ensure that the sample sufficiently covers the relevant
usually charge at public charging stations (around 80%), but many population. To derive these quotas we used data of a previous study by
(around 40%) would prefer to charge at their own parking lot. From a Dütschke et al. (2013) who collected a sample that was representative
techno-economic point of view, the impact of EVSE on the market for the adult population of Germany. After limiting this sample to cases
diffusion of EVs has been analysed by various studies. Nicholas et al. who satisfied the two screening criteria, we used the demographic
(2012) conclude, based on mobility data from 48 households in the properties of the remaining subsample to define the quotas that our
Sacramento Area, for EVs with a range of 120 miles that 3.4% of the own sample had to meet. These quotas referred to gender, education
tours (accounting for 30% of the vehicle miles travelled) would not be and age (cf. Table 1). The resulting sample was collected via an online
possible without fast charging infrastructure. Gnann (2015) as well as survey and eventually comprised 1003 respondents from Germany.
Gnann and Plötz (2015) find that a large share of vehicle-owning Questionnaire design. The online survey consisted of four parts. The
households in Germany are equipped with garages and require EVSE first part contained questions about demographic information to check
only for long-distance travel. In addition, Jakobsson et al. (2016) con- the screening criteria and quotas at the beginning of the survey. The
clude that multi-car households can better handle the restricted range second part of the survey contained questions about the mobility be-
of BEVs and depend less on public charging infrastructure. Lin and haviour and mobility features of the respondent or his/her household.
Greene (2011) point out that home charging is much more important These questions were placed deliberately before the evaluation of EVSE
for the diffusion of EVs than EVSE. offers (which was the third part of the survey) in order to make
55
J. Globisch, et al. Transport Policy 81 (2019) 54–63
Table 1 Table 3
Sampling quotas. Mobility-related variables.
Gender Age (mean = 48 year) Variable Categories and descriptive statistics
Male 49.9% 18–30 years 16.7%
Female 50.1% 31–40 years 17.0% Car size (measured by cylinder volume) Small (< 1.4 L) 25%
Education 41-50 years 19.8% Middle (≥1.4 to ≤2 L) 63%
No graduation/GCSE EQF-Level 2* 40.3% 51–60 years 16.8% Large (> 2 L) 10%
GCSE EQF-Level 3** 32.8% Older than 60 years 29.7% Not specified 2%
A-level/university degree 26.9% Car acquisition New car 50%
Pre-owned car 50%
European Qualification Framework equivalent for *“Hauptschulabschluss” and Number of cars One car 68%
**“Mittlere Reife”. Two or more cars 32%
Parking place with socket Regularly available 47%
Not regularly available 53%
respondents aware of their mobility needs and behaviour. I.e. we aimed Frequency of trips with more than > 3 times per month 18%
to foster a framing effect so that the respondents evaluate the EVSE 100 km 3 times per month 11%
offers in the light of their personal mobility requirements. The fourth 2 times per month 13%
Once per month 16%
part of the questionnaire contained items to measure attitudinal char-
Less than once per 42%
acteristics. These items were placed at the end of the questionnaire to month
avoid the evaluations of the EVSE offers being influenced by perceived Annual mileage First quartile 7.000 km
social desirability or pressure to give answers in accordance with re- Median 10.000 km
ported attitudes. Third quartile 16.000 km
Besides the information needed to assign respondents to the sam-
pling quotas (cf. Table 1), additional demographic information was
information regarding the properties of EVSE offers that are acceptable
collected in the first part of the questionnaire in order to identify and
for customers. Therefore, a choice with three options (A vs. B vs. neither
describe target groups. This additional information is presented in
A nor B) would have entailed the risk that many respondents select
Table 2.
“neither A nor B”, thus limiting the practical relevance of our results.
Mobility needs and mobility behavior were surveyed in the second
Allowing more graduated evaluations mitigates this risk since these
part of the questionnaire. The respective descriptive statistics are re-
provide information about which factors affect attractiveness of EVSE
ported in Table 3. Car size and car acquisition are relevant information
offers even for respondents who would not choose either of the two
as these properties can influence the willingness or ability to adopt an
alternatives offered.
EV and thus facilitate some needs and preferences with regard to EVSE.
The EVSE offers were preceded by a short text and characterised by
The number of cars in the household, the availability of a parking place
different attributes, which are symbolized by the placeholders depicted
with a wall socket, the frequency of long distance trips and the annual
as “[xxx]” in Fig. 1. One placeholder (range of the hypothetical EV) was
mileage are considered to be decisive for the dependency on public
part of the introductory text and one of three options was displayed
EVSE and may further increase the need for EVSE in certain places, e.g.
based on a random selection: 150 km, 250 km or 350 km. The values
near to home or along the highway (Jakobsson et al., 2016; Gnann,
between 150 km and 350 km represent the range of EVs that were
2015; Gnann and Plötz, 2015).
available on the market at the time of the survey (December 2016).
Fig. 1 shows how these EVSE offers were depicted in the third part
Except from the range of the hypothetical EV the text was identical for
of the online survey. Each participant had to rate the attractiveness of
all respondents and provided context information. In particular, the
ten EVSE offers presented pairwise within five tasks. Each task con-
respondents were informed how much charging an EV would cost at
sisted of an introductory text, a table with information about the
home (€5/100 km) and at public charging spots (€7/100 km). As
properties of two EVSE offers and two six-point Likert-scales to evaluate
comparison, current fuel prices of 1.60 € (incl. VAT) and about 7 L/
the attractiveness of each EVSE offer (from 1 = “very unattractive” to
100 km real world fuel consumption amount to approx. 11 € per
6 = “very attractive”). We decided to conduct a rating-based conjoint
100 km range for a gasoline vehicle (cf. Gnann et al. (2014) and Plötz
analysis instead of a choice-based conjoint analysis, because a six-point
et al. (2014a,b) for a discussion of current and future fuel prices).
Likert-scale as the dependent variable allows more nuanced evaluations
The range of the hypothetical EV differed between the respondents
of the EVSE offers by the respondents.
but was the same over all five exercises. In contrast, the displayed at-
We considered it important to enable these more nuanced evalua-
tribute levels of the hypothetical EVSE offers (depicted in the table in
tions for two reasons: (1) The subject of the analysis (attractiveness of
Fig. 1) changed each exercise. Which attribute levels were displayed
EVSE) can be assumed to be psychologically distant (cf. Liberman et al.,
was randomly selected. Each attribute had five possible levels:
2008), because EVs are a new and unknown technology - only 1.5% of
our respondents actually owned an EV. Therefore, we expected re-
spondents would find it easier to answer using a graduated evaluation • Monthly basic fee: €0, €5, €10, €15 and €20.
of attractiveness than by making a decision where the only options are • Density of charging spots in cities: none, every 250 m, every 500 m,
every 750 m and every 1000 m.
A vs. B vs. neither A nor B. (2) Prior studies provide not enough
• Density of charging spots along the highway: none, every 25 km,
every 50 km, every 75 km and every 100 km.
Table 2 • The attribute levels of “charging duration in cities” and “charging
Demographic variables. duration along the highway” were identical: N/A (if density was
Variable Categories and descriptive statistics
“none”), 10 min, 30 min, 60 min or 120 min for 100 km range.
Occupation Full time employee 43% We included the attribute ‘monthly basic fee’ as our literature re-
Part time employee 11% view had identified the willingness to pay for the existence of public
Something else (student, 46%
retired, etc.)
EVSE is an understudied issue. In addition, we included density of EVSE
Household net income per month First quartile €1700 (in cities and along the highway) to address the need for detours for
(Mean = €2780) Median €2500 recharging and perceived mobility restrictions due to a lack of charging
Third quartile €3200 spots. The attribute levels for EVSE density in cities is based on the
56
J. Globisch, et al. Transport Policy 81 (2019) 54–63
findings of Sun et al. (2016). The attribute levels for EVSE density along answer (Weiber and Mühlhaus, 2014).
the highway was chosen proportional to the distances in cities. As the The items used as indicators for environmental consciousness and
duration of the charging process was identified as an important aspect technophilia contained 5 answer options including two affirmative
(Philipsen et al., 2016; Schroeder and Traber, 2012), we included it as options (I totally/rather agree), two depreciating options (I totally/ra-
an attribute. The attribute levels for the duration of the charging pro- ther disagree) and one neutral option (I partly agree and partly dis-
cess approximately reflect the technological performance of currently agree). This design is considered less advantageous by some methodo-
available charging systems (December 2016). In order to direct re- logical studies as the neutral option may indicate medium consent with
spondents' attention to the changed attribute levels, additional advice the statement as well as the general absence of a clear sentiment to-
was displayed after the respondent went on from the first to the second wards the statement (Trommsdorff, 1975; Green and Rao, 1970).
task. However, we decided to use five answer options to adhere as closely as
Part four of the survey measured attitudinal constructs. Our analysis possible to the original and validated form of the items of Wingerter
comprised the variables ‘environmental consciousness’ and ‘techno- (2014) and Neyer et al. (2016). In general, high values indicate affir-
philia’ as well as ‘general attitude towards EVs’. The constructs en- mation of the statements while low values indicate disagreement. In our
vironmental consciousness and technophilia were measured by existent analysis we considered the level of measurement of Likert scales to be
and validated scales that consist of multiple items as indicators of these quasi-metric (cf. Norman, 2010).
psychological constructs (Wingerter, 2014; Neyer et al., 2016). The
general attitude towards EVs was measured by a single item that has
been used by the authors in prior surveys. The variables and the re- 2.2. Data preparation and analysis
spective items that were used as indicators are shown in Table 4. The
descriptive statistics (mean, standard deviation and the range of values) The preparation of the data for analysis consisted of five steps: (1) a
are displayed in brackets. confirmatory factor analysis (CFA) to test whether the items depicted in
Like the evaluations of the EVSE offers, these items are Likert scales, Table 4 are valid and reliable indicators of the proposed theoretical
i.e. the respondents could choose different options to agree or disagree constructs listed in the left column. (2) Imputation of missing values to
with these statements. The item used as an indicator for the general reduce bias resulting from estimations impacted by missing values. (3)
attitude towards EVs had six options as possible answers – three affir- Rearranging the structure of the dataset so that the evaluated offers
mative and three depreciating options (I totally/predominantly/rather constitute cases, i.e. rows in the dataset (4) Excluding unsuited cases
(dis)agree). Unlike the evaluation of EVSE offers, there is an additional (i.e. evaluations of EVSE offers). from the dataset. (5) Dichotomising all
“not specified” option. This design has the advantage that respondents variables with non-(quasi)-metric level of measurement and mean
either have to reveal a general tendency or deliberately refuse to centring all non-binary variables. We explain the steps in more detail
below.
57
J. Globisch, et al. Transport Policy 81 (2019) 54–63
Table 4
Attitudinal constructs.
Variable Item text (mean; standard deviation; min/max)
Environmental consciousness It worries me when I think about the environmental conditions our children and grandchildren will have to live with. (3.5; 1.1; 1 to 5)
If we just carry on as before, we are heading for an environmental catastrophe. (3.7; 1.0; 1 to 5)
I am often angry and appalled when I read reports about environmental problems in newspapers or watch these kinds of TV programmes. (3.4;
1.0; 1 to 5)
Technophilia I am very interested in the latest technology developments. (3.4; 1.1; 1 to 5)
It doesn't take me long to learn to like new technology developments. (3.3; 1.1; 1 to 5)
I am always keen to use the latest technological devices. (2.8; 1.2; 1 to 5)
If I had the chance to do so, I would use the latest technical products even more often than I do at present. (3.0; 1.2; 1 to 5)*
General attitude towards EVs In general I think electric vehicles are a good thing. (4.4; 1.5; 1 to 6)
* Due to a lack of construct validity we exclude this from the analysis (cf. section 2.2).
The first step of data preparation addressed the variables that refer given a large enough sample. Thus, we consider our model fits the data
to psychological constructs and use multiple items as indicators, i.e. sufficiently (Weiber and Mühlhaus, 2014; Kline, 2011).
‘environmental consciousness’ and ‘technophilia’. The values of the In the revised model, the two factors (environmental consciousness
different indicators had to be combined to a common value for each and technophilia) also show factor reliability and construct validity as
variable for the analysis. The presence of factor reliability and construct illustrated by the parameters for the local fit reported in Table 6. We
validity is a prerequirement for this. Whether factor reliability and therefore combined the values of the respective indicators by calcu-
construct validity are given can be assessed by a confirmatory factor lating their average scores for each respondent.
analysis (CFA). An acceptable global fit in a CFA is a necessary but not In the second step of data preparation, we used Expectation-
sufficient condition for construct validity in a factor model. Maximization-imputation to estimate the values for missing data. This
Furthermore, the factors should feature convergent and discrimanent was necessary as listwise deletion (as the default option to deal with
validity, which refers to the local fit of the factors comprising the model missing data) presumes that missing data is missing completely at
consists of (Weiber and Mühlhaus, 2014; Kline, 2011). random. Especially for some of the demographic information in our
Table 5 shows the parameters for the global fit resulting from the analysis (e.g. household income), this claim is untenable as previous
CFA. The initial factor model contained the three indicators for en- studies found that the occurrence of missing data correlates with the
vironmental consciousness and the four indicators for techophilia from value of the respective variable (e.g. higher incomes are more likely to
Table 4. In reference to the thresholds for a good or acceptable global fit be not reported than lower incomes (Acock, 2005)).
(second and third line in Table 5), the outcomes of the CFA illustrate The third step of data preparation was to change the structure of the
that the co-variance structure postulated by the initial model did not data set. This was due to the necessity to analyse our data using a
adequately reflect the empirically observed co-variance structure hierarchical linear random effects model (cf. section 3) with the re-
(fourth line in Table 5). Modification indices suggested a problem with spondents as macro-level units (level 2) and the EVSE offers as micro-
the indicators displayed in the third and fourth lines of the subsection level units (level 1). Accordingly, we changed the format of our dataset
“technophilia” in Table 4. As the item “I am always keen to use the so that the EVSE offers are the cases (lines) of the data matrix and the
latest technological devices” showed a slightly higher indicator relia- respondent is indicated by an ID variable.
bility (0.74 compared to .69; not displayed in Tables 5 and 6) than the In a fourth step we deleted all cases (now EVSE offers), which had
item “If I had the chance to do so, I would use the latest technical the attribute level “none” for either of the two attributes “Density of
products even more often than I do at present”, we excluded the latter charging spots in cities” and “Density of charging spots along the
from the analysis. This is reasonable as the latter refers to intention highway”. This was necessary as “none” corresponds to “N/A” for the
while the other three items refer to actual behaviour. charging duration in cities or along the highway respectively.
A factor model with only three indicators for technophilia shows an Therefore, the exclusion of these cases was inevitable due to the oc-
improved global fit (fifth line in Table 5): While the χ2-test indicates currence of perfect multi-collinearity between the two variables. The
that there are still significant deviations between the postulated and reason we included “none” in our questionnaire in the first place was
observed co-variance structure (p = .010), the other parameters in- that analysing the value placed on the very existence of public EVSE in
dicate a good fit of the model. CFI and RMSEA are above or below their cities and along the highway was a secondary objective of our study.
threshold values. The value for PClose indicates a close fit of the model However, this objective is beyond the scope of this paper. As a result of
as the hypothesis that RMSEA is higher than 0.050 can be rejected. In this data adjustment, our dataset contains evaluations of 8887 EVSE
addition, a poor fit of the model can be ruled out as the upper bound of offers from 1003 respondents, i.e. 1143 evaluations of EVSE offers were
the confidence interval of RMSEA is below 0.100 (Hi90 = 0.060). The excluded (between 4 and 10 evaluations of EVSE offers per respondent
χ2-test is disputed in the methodological literature because it tests for remained; on average each respondent evaluated 8.86 EVSE offers).
differences between postulated and observed co-variance structure. As In step five, we dichotomised all the non-metric variables included
the sensitivity of the χ2-test increases with sample size, and as no model in our analysis (e.g. gender) and defined a baseline value for each
fits the data perfectly, χ2 will always indicate insufficient global fit variable (e.g. female). Therefore, the effects for the dichotomised
variables reported in section 3 indicate how EVSE offers differ in at-
tractiveness depending on respondent characteristics compared to the
Table 5 baseline value (e.g. “men” compared to “women”). In addition, we
Results of CFA for global model fit. improved the interpretability of the results by mean centring
p CFI RMSEA PClose Hi90 (X Xsample ) all the non-dichotomous variables (all the attributes of the
EVSE offers and respondent characteristics like age, net household in-
Model n.a. > .95* < .08* > .05* ≤.10* come or the theoretical constructs depicted in Table 4). Therefore, the
> .05** > .97** < .05** n.a. n.a.
Initial model .000 .967 .102 .000 .117
effects of the EVSE offers' attributes in section 3 can be interpreted as
Revised model .010 .996 .039 .785 .060 the conditional effects for average EVSE offers (average monthly basic
fee, EVSE density and charging duration) evaluated by a baseline
Thresholds for acceptable* and good** model fit.
58
J. Globisch, et al. Transport Policy 81 (2019) 54–63
Table 6
Results of CFA for local model fit.
Factor reliability AVE Squared max. intercorrelation
a a
Factor > .6 > .5 < AVEa
Environmental consciousness .844 .643 .074
Technophilia .887 .724 .074
a
Thresholds for acceptable local fit.
Table 7
Summary of the hierarchical linear model.
Parameters Unconditional Level 1:fixed Level 2 & level 1: random Interaction
2
Pseudo-R .481 .529 .701 .643
Deviance statistic 28,734.842 28,073.576 27,586.353 27,627.860
Number of estimated parameters 3 13 41 38
respondent (e.g. female) with an average age, household income, the results of step 2 and step 4 provides information on the extent to
technophilia etc. (cf. Algina and Swarninathan, 2011). which the explanation of the dependent variable by level 1 variables is
improved by taking the nested structure of the data into account. In-
3. Results teractions between level 2 and level 1 variables were included in step 5
(column 5). This provides information about where the improvement in
The presentation and discussion of the results of the hierarchical step 4 originates from.
linear model is guided by the advice of Peugh (2010) and Hox (2010). Model evaluation. Table 7 provides a summary of the models. The
The results of our analysis are summarized in Tables 7–10 (variables model which contains cross-level interactions (“interaction”) entails
without significant effects are excluded from the analysis; non-sig- fewer parameters that need to be estimated than the “random” coeffi-
nificant results are omitted for the sake of clarity). cient model without cross-level interactions (38 vs. 41 parameters). The
As recommended by Hox (2010) we pursued a bottom-up strategy reason is that the “interaction” model does not contain random effects
for our analysis. This means (1) we first set up an unconditional model for the level 1 variable “monthly basic fee” in contrast to the “random”
before (2) we built a level 1 model with fixed effects, (3) included level coefficient model. After the inclusion of the cross-level interactions
2 variables, (4) added random effects for level 1 variables and (5) in- depicted in Table 7, there remains neither a significant variance of the
corporated cross-level interactions. These steps resulted in the different slopes for “monthly basic fee” nor a significant variance of the co-var-
models presented in the columns 2 to 5 of Table 7. The unconditional iances between “monthly basic fee” and the other level 1 variables. In
model (step 1) did not comprise any independent variables except for other words, if the cross-level interactions are taken into account, the
the ID of the respondents. This step assessed the share of variance of the strength of influence (i.e. the slope) of “monthly basic fee” on the
dependent variable explained by the clustered structure of the data (cf. evaluation of an EVSE offer does not differ significantly between re-
column 2). Step 2 estimated fixed effects for level 1 variables (attributes spondents.
of EVSE offers), i.e. no variations of the level 1 effects between the level As Peugh (2010) points out, there is no measure for effect strength
2 units (respondents) were allowed (cf. column 3). Results for steps 3 comparable to Cohen's d for hierarchical linear models. With regard to
and 4 are reported together in column 4. Step 3 comprised the inclusion explanatory power, we find a correlation of 0.802 between the real
of level 2 variables (respondent characteristics) to explain the variance evaluations of EVSE offers and the values predicted by the ‘interaction’
between respondents. Step 4 estimated the effects of level 1 variables model. The pseudo-R2 of this model is 0.643, i.e. the model explains
(EVSE offers) separately for the level 2 units (respondents). Comparing 64.3% of the variance in the evaluations of EVSE offers (Peugh, 2010).
Table 8
Effects and interactions of level 1 and level 2 variables.
Parameters Unconditional Level 1:fixed Level 2 & level 1: random Interaction
Intercept (γ0/0) 3.648*** (.032) 3.038*** (.376) 2.455*** (.370) 2.489*** (.367)
Monthly basic fee (γ1/0) – -.033*** (.002) -.032*** (.002) -.033*** (.002)
Density EVSE in cities (γ2/0) – -.014** (.004) -.016*** (.004) -.014** (.004)
Charging duration in cities (γ3/0) – -.037*** (.003) -.038*** (.004) -.037*** (.003)
Density EVSE along the highway (γ4/0) – -.016*** (.004) -.015** (.004) -.016*** (.004)
Charging duration along the highway (γ5/0) – -.041*** (.003) -.042*** (.004) -.042*** (.003)
Meanres: Monthly basic fee (γ0/1) – n.s. .024* (.011) .024* (.011)
Meanres: Charging duration in cities (γ0/2) – .063** (.023) .056** (.020) .057** (.020)
Meanres: Density EVSE along the highway (γ0/3) – n.s. n.s. .059* (.029)
Gender (γ0/4) – – -.168** (.057) -.169** (.056)
A-level (γ0/5) – – .097* (.041) .092* (.041)
University degree (γ0/6) – – -.640** (.230) -.612** (.229)
Age (γ0/7) – – -.008*** (.002) -.008*** (.002)
Environmental consciousness (γ0/8) – – .103** (.033) .099** (.033)
Technophilia (γ0/9) – – .129*** (.031) .129*** (.031)
General attitude towards EVs (γ0/10) – – .271*** (.021) .266*** (.021)
Charging duration in cities*age (γ3/7) – – – -.001* (.000)
Charging duration in cities*technophilia (γ3/9) – – – -.011** (.003)
Charging duration highway* Gen. attitude EVs (γ5/10) – – – -.006** (.002)
Significance of parameters indicated by *p < .050; **p < .010; ***p < .001.
59
J. Globisch, et al. Transport Policy 81 (2019) 54–63
Table 9
Variance components (random effects) of the hierarchical linear model.
Parameters Unconditional Level 1:fixed Level 2 & level 1: random Interaction
2
Residual (σ ) 1.186*** (.019) 1.075*** (.017) .870*** (.019) .921*** (.018)
Intercept (τ00) .851*** (.044) .864*** (.044) .617*** (.033) .608*** (.032)
Slope monthly basic fee (τ11) .003*** (.000) n.s.
Slope density EVSE in cities (τ22) n.s. .002* (.001)
Slope charging duration in cities (τ33) .005*** (.001) .004*** (.001)
Slope density EVSE along the highway (τ44) .002* (.001) .002* (.001)
Slope charging duration along the highway (τ55) .004*** (.001) .004*** (.001)
Covariance density EVSE in cities/density EVSE along the highway (τ24) .002** (.001) .002*** (.001)
Covariance charging duration in cities/density EVSE along the highway (τ34) .001* (.001) .002*** (.000)
Covariance charging duration in cities/charging duration along the highway (τ35) .002*** (.000) .001*** (.000)
Significance of parameters indicated by *p < .050; **p < .010; ***p < .001.
Table 10 offer is 2.5 (γ0/0 of the interaction model), i.e. between the answer
Proportional reduction of variance (PRV). options “I predominantly disagree” and “I rather disagree” regarding
Variable/interaction PRV the approval of the statement that such an offer is attractive.
To further improve the clarity of the information provided by the
regression coefficients, we divided the values of ‘density EVSE in cities’
2
Level 1 variables: PRV of residuals (σ )
Monthly basic fee (γ1/0) 5.5% by 100 (resulting unit = 100m) and the values of ‘charging duration in
Density EVSE in cities (γ2/0) 4.7%
Charging duration in cities (γ3/0) 7.7%
cities/along the highway’ as well as ‘density EVSE along the highway’
Density EVSE along the highway (γ4/0) 5.0% by 10 (resulting unit = 10 min/10 km). As a consequence of centring
Charging duration at the highway (γ5/0) 9.9% and harmonization the effects in Table 7 can be interpreted and sum-
Level 2 variables: PRV of intercepts (τ00) marized in the following way: An in all other respects average EVSE
Gender (γ0/4) 1.0%
offer is evaluated 0.033 points less positive for every euro increase in
A-level (γ0/5) 0.4%
University degree (γ0/6) 0.6% the basic monthly fee. Similarly, increasing the distance between
Age (γ0/7) 2.3% charging points in cities/along the highway by 100m/10 km results in a
Environmental consciousness (γ0/8) 1.0% 0.014/.016 point less favourable evaluation, respectively
Technophilia (γ0/9) 1.7% Conditional level 1 effects (effects of attributes of EVSE offers condi-
General attitude towards EVs (γ0/10) 15.3%
Cross-level-interaction: PRV of slopes (τXX)
tional on respondent characteristics). If charging for 100 km range takes
Slope charging duration in cities (τ33) by age (γ3/7) 3.2% 10 min longer in cities, the evaluation of an otherwise average EVSE
Slope charging duration in cities (τ33) by technophilia (γ3/9) 3.4% offer by a respondent of average age (the mean age is 48 years) and
Slope charging duration along the highway (τ44) by general attitude 2.3% technophilia (3.2 points of 5) worsens by .037 point. The effect is .001
towards EVs (γ5/10)
point more/less negative (γ3/7) for every year the person is older/
younger than the average. Furthermore, this effect is .011 point more/
less negative (γ3/9) for every point the person is above/below average
In contrast, the ‘random’ coefficient model yields a correlation of 0.843
on the technophilia scale. The effect for charging points along the
between predicted and observed values, which equates to a pseudo-R2
highway is in a similar range. Increasing charging time for 100 km by
of 0.710. However, despite the lower explanatory power of the ‘inter-
10 min results in a .042 point less favourable evaluation for persons
action’ model, it is more useful in the context of our research question:
with an average attitude towards EVs (4.4 points of 6). This effect is
We aim to describe potential target groups – thus the explanatory
.006 point more/less negative (γ5/10) for every point the person is
power and parsimony of the model is not an end in itself. The cross-
above/below average on the scale that measures the general attitude
level interactions are useful to describe potential target groups.
towards EVs.
The fixed effects reported in Table 8 are unstandardized regression
Level 2 effects (effects of respondent characteristics). In general, men
parameters for direct effects of level 1 variables (γX/0), level 2 vari-
evaluate an average EVSE offer 0.169 point less favourably than women
ables (γ0/X) and cross-level interactions between level 1 and level 2
(γ0/4). The level of formal education has an ambiguous effect:
variables (γX/X). Table 9 comprises the variance of the residuals of the
Compared to people with an educational qualification not higher than
dependent variable (σ2), the variance of the intercepts of level 2 units
GCSE (equivalent to EQF-Level 2), academics rate an average EVSE
(τ00), the variance of slopes of level 1 variables between level 2 units
offer worse by 0.612 point (γ0/6), while respondents with A-levels rate
and the covariance between level 1 variables among the level 2 units.
it more favorably by 0.092 point (γ0/5). Furthermore, the average
The last two variance components are only present if the model com-
EVSE offer is rated more negatively the older the respondent is – for
prises random effects, i.e. the slopes of level 1 variables can differ be-
every year a person is older, the rating decreases by 0.008 point (γ0/7).
tween level 2 units. The standard errors of the parameter estimates are
Attitudinal factors also influence the evaluations of EVSE offers. For
listed in parentheses.
each point increase on the environmental consciousness scale an
Level 1 effects (effects of attributes of EVSE offers). The model con-
average EVSE offer is rated 0.099 point better (γ0/8). Furthermore, the
tained those variables described in section 2.1, which significantly af-
evaluation of an average EVSE offer is 0.129 point more favourable for
fected the evaluations of EVSE offers. As described in section 2.2, we
a one-point increase on the technophilia scale (γ0/9). Also an increase
(grand-) mean centred our non-dichotomous independent variables as
of one point on the scale which measures the general attitude towards
this improves the interpretability of the regression parameters and in-
EVs results in a more positive evaluation by 0.266 point (γ0/10).
teraction effects. Consequently, level 1 effects are conditional effects for
Beside the variables described in section 2.1, we further included
an average EVSE offer by and baseline and average respondent. An
the respondent mean (X respondent) of each level 1 variable to control for
average EVSE offer has the following properties: €10 monthly basic fee,
effects of level 1 variables on level 2 as recommended by Algina and
charging spots every 624 m in cities and every 62 km along the
Swarninathan (2011). In Table 8, these variables are labeled by the
highway, 55 min to charge for 100 km range at charging points in cities
prefix “Meanres”. The effects of these variables imply that respondents
and along the highway. The mean evaluation of such an average EVSE
evaluate EVSE offers more positively that have on average a higher
60
J. Globisch, et al. Transport Policy 81 (2019) 54–63
monthly basic fee, longer charging duration in cities and fewer charging of charging duration. The basic fee and the density of the charging
spots along the highway. Although the significance levels of these ef- points have a weaker influence on the ratings.
fects are comparatively low, this counter-intuitive finding might in- With regard to the third research question concerning target groups
dicate a psychological effect known from prospect theory (cf. Thaler with different preference structures for public charging infrastructures,
et al., 1997): gains usually have a weaker effect than losses of the same it should first be noted that the overall evaluations show that EVSE
absolute magnitude. This means for respondents confronted with on offers are not very attractive. An average EVSE offer, i.e. with €10
average relatively unfavourable EVSE offers, there is less latitude for monthly basic fee, charging spots every 624 m in cities and every 62 km
(more) influential losses compared to respondents who evaluated EVSE along the highway, 55 min to charge for 100 km range at charging
offers that are more favourable on average. points in cities and along the highway, receives a mean evaluation of
Variance components (random effects). Furthermore, the results re- 2.5 (between 2 = “largely unattractive” and 3 = “rather unattractive”).
ported in Table 9 indicate that the effects of density of EVSE in cities Even if the charging duration in cities and along the highway would be
(τ22) and along the highway (τ44) as well as the effects of charging only 20 min for 100 km range and the density of charging point would
duration in cities (τ33) and along the highway (τ55) vary significantly be increase to every 300 m in cities and every 30 km along the highway
among respondents even after controlling for cross-level interaction the estimated evaluation would still be negative (2.9). This value is still
effects. The significant co-variance between the density of EVSE in ci- well below 3.5, the middle of the evaluation scale, from which one
ties and along the highway (τ24) indicates that respondents who assign could assume that such an offer would be taken up and that there would
more value to EVSE density in cities also assign more value to EVSE be a willingness to pay. To put the monthly basic fee of €10 into per-
density along the highway. Similarly, respondents who stronger ap- spective: e.g. Lin and Greene (2011) estimate (based on the assumption
preciate the charging duration in cities also stronger appreciate the of $15 average costs of compensating a household in the US for a day
density of EVSE (τ34) and the charging duration (τ35) along the without sufficient range of a BEV) that customers should have a will-
highway. ingness to pay for public EVSE of approximately between $200 - $800
Proportional reduction of variance (PRV). Due to the lack of a measure per year. In contrast, our results indicate that even very well equipped
for effect strength comparable to Cohen's d for hierarchical linear public charging networks would be valued well below this number.
models we subsequently report the proportional reduction of variance In the present study, survey respondents evaluated EVSE offers in-
for the variance components of the different variables (PRV = cluding certain costs for recharge. As most respondents do not own an
(VarNoPredictor-VarPredictor)/VarNoPredictor; for σ2, τ00 and τXX (Peugh, EV presently, their knowledge about the operating costs of EV is still
2010)). In Table 10, we report PRV on a percentage basis (PRV*100) for limited. One potential suggestion for future research is thus to include
the inclusion of level 1 variables (PRV of residuals of the dependent the cost for refuelling of a conventional internal combustion engine
variable at level 1), level 2 variables (PRV of intercept between level 2 vehicle as additional cost reference. The respondents could then more
units) and cross-level interactions between level 2 variables and level 1 directly relate to their own vehicle operating cost. However, in the
variables (PRV of slopes between level 2 units). present case, this information would have been irrelevant for the de-
For level 1 variables, i.e. the attributes of EVSE offers, charging cision and could have confused the respondents such that survey as
duration in cities (7.7%) and along the highway (9.9%) have the presented seems appropriate for our research questions and design.
highest PRV. The other three attributes provide a PRV of about five Further analysis into attitudinal variables suggests that there are
percent, in particular 5.5% for the monthly basic fee, 4.7% for the certain groups who assign more value to public EVSE than others. In
density of EVSE in cities and 5.0% for the density of EVSE on highways. particular, environmental consciousness, technophilia and the general
These numbers do not simply add up – in total, the inclusion of all level attitude towards EVs correspond to evaluations that are more favour-
1 variables results in a PRV of 22.3%. able. In the context of the effects of gender, education and age this
The variance of the intercept between the respondents can be ex- implies that younger non-academic women who are especially en-
plained by level 2 variables. Here, the general attitude towards EVs has vironmental conscious, technophile and open minded towards EVs are a
by far the highest PRV (15.3%). The remaining level 2 variables have group who value public EVSE more than others. Some of these variables
comparatively low PRVs between 2.3% (age) and 0.4% (A-level com- also affect the valuation of different aspects of public EVSE. For tech-
pared to no graduation/GCSE equivalent to EQF-Level 2). Again, these nophilia and the general attitude towards EVs we found significant
numbers do not simply add up – the PRV that results from including all cross-level interactions with charging duration in cities and along the
the level 2 variables is 28.1%. highway. These interaction effects indicate that more technophile and
In addition, level 2 variables can help to reduce the variance of EV-friendly persons are more sensitive to longer charging durations.
slopes of level 1 variables. The variance of the slope of charging Thus, not only does charging duration have the strongest overall in-
duration in cities between respondents can be reduced by 3.2% by fluence on the ratings of EVSE offers - the effect is also particularly
taking into account the interaction with age. A similar reduction in the strong among car drivers, who generally value public charging infra-
variance between respondents results from including the interaction structure more highly.
with technophilia in the model (3.4%). The reduction of slope variance The results discussed above illustrate that our model provides useful
of density of EVSE along the highway by including the interaction with insights into the preferences regarding public EVSE. However, there are
the general attitude towards EVs is slightly below these numbers (2.3%) also some limitations of our study. The result that the level 2 variables
explain only 28.1% of the respondent-related variance suggests that
4. Discussion relevant additional characteristics exist, which are not covered by the
model. In this regard, it seems noteworthy that several level 2 variables
Overall, it can be said that the variance in the evaluations of the turned out to be non-significant. In particular, this applies to the vari-
attractiveness of the EVSE offers by the 1003 car drivers in our sample ables that relate to mobility needs and mobility behaviour (cf. Table 3)
is well explained by our statistical model. In addition, our data provide as well as demographic variables that usually correlate with general
insightful results with regard to our three research questions. With re- mobility (net income of household and professional occupation) and the
gard to the research question of which factors influence the overall electric range of the hypothetical EV (cf. Fig. 1). In combination with
assessment of public charging infrastructure, our results show that a the very strong influence of the general attitude towards EVs, this might
basic charge as well as the charging duration and density of charging indicate that the hypothetical evaluation of EVSE offers is psychologi-
stations have a significant influence. With regard to the second research cally distant (cf. Rezvani et al. (2015) for the consequences of hy-
question, how strong the influence of these factors is on the evaluation pothetical evaluations of EVs). The evaluations of EVSE offers might be
of public charging infrastructure, our results underline the importance based more on basic beliefs towards (the usefulness of) EVSE rather
61
J. Globisch, et al. Transport Policy 81 (2019) 54–63
than on a deliberate assessment of the respondent's own mobility needs. Roberts, J.K. (Eds.), European Association of Methodology. Handbook Of Advanced
Similar to the suggestions of Rezvani et al. (2015) with regard to studies Multilevel Analysis. Routledge, New York, pp. 285–312.
Alhazmi, Y.A., Mostafa, H.A., Salama, M.M.A., 2017. Optimal allocation for electric ve-
of EV evaluation, future studies might benefit from analysing samples, hicle charging stations using Trip Success Ratio. Int. J. Electr. Power Energy Syst. 91,
which deliberately overrepresent respondents with actual experience of 101–116.
using EVs. Preferences might change after gaining hands-on experience Bunce, L., Harris, M., Burgess, M., 2014. Charge up then charge out? Drivers' perceptions
and experiences of electric vehicles in the UK. In: Transportation Research Part A:
(cf. Bunce et al. (2014) who finds that actual users of EVs even perceive Policy and Practice, vol. 59. pp. 278–287.
public recharging as more convenient than refuelling). Furthermore, Dong, J., Liu, C., Lin, Z., 2014. Charging infrastructure planning for promoting battery
the relevance of additional attitudinal characteristics like risk-tolerance electric vehicles: an activity-based approach using multiday travel data.
Transportation Research Part C: Emerging Technologies, vol. 38. pp. 44–55.
or the general willingness to adjust behavioural routines should be Dütschke, E., Paetz, A.-G., Wesche, J., 2013. Integration Erneuerbarer Energien durch
explored by future research. Elektromobilität – inwieweit sind Konsumenten bereit, einen Beitrag zu leisten?
UmweltWirtschaftsForum 21 (3-4), 233–242.
Gnann, T., 2015. Market Diffusion of Plug-In Electric Vehicles and Their Charging
5. Conclusions
Infrastructure. Ph.D. thesis. Fraunhofer Publishing, Karlsruhe Institute of
Technology (KIT), Karlsruhe, Germany.
On the basis of the results discussed above, several conclusions can Gnann, T., Funke, S., Jakobsson, N., Plötz, P., Sprei, F., Bennehag, A., 2018. Fast charging
be drawn with implications for possible business models and transport infrastructure for electric vehicles: today's situation and future needs. Transport. Res.
Transport Environ. 62, 314–329. https://doi.org/10.1016/j.trd.2018.03.004.
policy measures. Against the background that there is only very limited, Gnann, T., Plötz, P., 2015. A review of combined models for market diffusion of alter-
if any, willingness to pay for public charging infrastructure, the cost- native fuel vehicles and their refueling infrastructure. Renew. Sustain. Energy Rev.
efficiency of the overall system is of central importance. In this respect, 47, 783–793.
Gnann, T., Plötz, P., Kühn, A., Wietschel, M., 2014. Modelling Market Diffusion Of Electric
the strong influence of the charging duration is advantageous for two Vehicles With Real World Driving Data German Market And Policy Options (Working
reasons (cf. Gnann et al., 2018): (1) a high utilisation of public EVSE, Paper Sustainability and Innovation No. S 12/2014). Karlsruhe. Retrieved from
which can be fostered by a smaller number of charging spots, is im- Fraunhofer ISI website: http://www.isi.fraunhofer.de/isi-wAssets/docs/e-x/
working-papers-sustainability-and-innovation/WP12-2014_Modelling-Market-
portant to find a viable business case. (2) Installation costs due to ne- Diffusion-of-Electric-Vehicles_Gnann-et-al.pdf.
cessary groundwork etc. account for a large share of the total costs of Green, P.E., Rao, V.R., 1970. Rating scales and information recovery. How many scales
public EVSE. If car drivers can be satisfied with a smaller number of and response categories to use? J. Mark. 34 (3), 33.
Guo, Z., Deride, J., Fan, Y., 2016. Infrastructure planning for fast charging stations in a
charging points, this is would lead to lower investments. In addition, if competitive market. In: Transportation Research Part C: Emerging Technologies, vol.
detours do not bother car drivers too much, the charging spots can be 68. pp. 215–227.
placed at locations where installation cost are low (e.g. due to con- Hardman, S., Jenn, A., Tal, G., Axsen, J., Beard, G., Daina, N., Figenbaum, E., Jakobsson,
N., Jochem, P., Kinnear, N., Plötz, P., Pontes, J., Refa, N., Sprei, F., Turrentine, T.,
venient access to medium-voltage network).
Witkamp, B., 2018. A review of consumer preferences of and interactions with
Even if the construction of public charging infrastructure is very electric vehicle charging infrastructure. Transport. Res. Transport Environ. 62,
cost-efficient, it will be difficult to develop viable business models due 508–523.
to the low willingness to pay. Accordingly, there is likely to be a need Hox, J.J., 2010. In: Multilevel Analysis: Techniques And Applications, 2. ed. Routledge,
New York Quantitative methodology series.
for political support measures. Our results provide some indications Jakobsson, N., Gnann, T., Plötz, P., Sprei, F., Karlsson, S., 2016. Are multi-car households
that this could also contribute to the further spread of electric vehicles. better suited for battery electric vehicles? – driving patterns and economics in
First of all, the strong positive influence of the general attitude towards Sweden and Germany. Transport. Res. C Emerg. Technol. 65, 1–15. https://doi.org/
10.1016/j.trc.2016.01.018.
EVS on the evaluation of EVSE offers can be seen as an indication that Kline, R.B., 2011. Principles and practice of structural equation modeling. Methodology
public charging infrastructure is particularly important for people who in the Social Sciences, third ed. Guilford Press, New York.
are more likely to adopt EVs. The influencing factors relating to atti- Liberman, N., Trope, Y., Stephan, E., 2008. Psychological distance. In: Kruglanski, A.W.,
Higgins, E.T. (Eds.), Social Psychology: Handbook of Basic Principles. Recording for
tudes can also be regarded as a corresponding indicator. In particular, the Blind & Dyslexic, Princeton, N.J., pp. 353–384.
environmental awareness and technophilia were previously identified Lin, Z., Greene, D.L., 2011. Promoting the market for plug-in hybrid and battery electric
as characteristics of early adopters of EVs (Plötz et al., 2014a,b). In vehicles. Transport. Res. Rec.: J. Transport Res. Board 2252 (1), 49–56. https://doi.
org/10.3141/2252-07.
contrast, the demographic characteristics (gender, education, age) that
Nicholas, M.A., Tal, G., 2017. Survey And Data Observation On Consumer Motivations To DC
influence the basic evaluation of EVSE offers do not fit the classic early Fast Charge (Research Report – UCD-ITS-RR-17-21). (Sacramento).
adopter clientele who are described by Plötz et al. (2014a,b) as middle Nicholas, M.A., Tal, G., Davies-Shawhyde, J., Woodjack, J., 2012. DC fast as the only
public charging option? Scenario testing from GPS tracked vehicles. In: TRB 91st
aged males with a high socio-economic status. Taken together, this can
Annual Meeting Compendium of Papers DVD, Transportation Research Board Annual
be seen as an indication that public charging infrastructure is particu- Meeting.
larly important for potential buyers of electric vehicles that do not Neyer, F.J., Felber, J., Gebhardt, C., 2016. Kurzskala zur Erfassung von
correspond to the classic early adopted clientele. Public charging in- Technikbereitschaft (technology commitment). In: Zusammenstellung so-
zialwissenschaftlicher Items und Skalen, https://doi.org/10.6102/zis244.
frastructure could therefore help to win these target groups over to Norman, G., 2010. Likert scales, levels of measurement and the "laws" of statistics. Adv.
electric mobility. Health Sci. Educ.: Theor. Pract. 15 (5), 625–632.
Peugh, J.L., 2010. A practical guide to multilevel modeling. J. Sch. Psychol. 48 (1),
85–112.
Acknowledgement Philipsen, R., Schmidt, T., van Heek, J., Ziefle, M., 2016. Fast-charging station here,
please! User criteria for electric vehicle fast-charging locations. Transport. Res. F
This publication was written in the framework of the "Profilregion Traffic Psychol. Behav. 40, 119–129.
Philipsen, R., Schmidt, T., Ziefle, M., 2015. A charging place to Be - users' evaluation
Mobilitätssysteme Karlsruhe", which is funded by the Ministry of criteria for the positioning of fast-charging infrastructure for electro mobility. In:
Economic Affairs, Labour and Housing Baden-Württemberg and as a Procedia Manufacturing, vol. 3. pp. 2792–2799.
National High Performance Center by the Fraunhofer-Gesellschaft Plötz, P., Gnann, T., Wietschel, M., 2014a. Modelling market diffusion of electric vehicles
with real world driving data — Part I: model structure and validation. Ecol. Econ.
107, 411–421. https://doi.org/10.1016/j.ecolecon.2014.09.021.
Appendix A. Supplementary data Plötz, P., Schneider, U., Globisch, J., Dütschke, E., 2014b. Who will buy electric vehicles?
Identifying early adopters in Germany. Transport. Res. Pol. Pract. 67, 96–109.
https://doi.org/10.1016/j.tra.2014.06.006.
Supplementary data to this article can be found online at https://
Rezvani, Z., Jansson, J., Bodin, J., 2015. Advances in consumer electric vehicle adoption
doi.org/10.1016/j.tranpol.2019.05.017. research: a review and research agenda. Transport. Res. Transport Environ. 34,
122–136.
References Sadeghi-Barzani, P., Rajabi-Ghahnavieh, A., Kazemi-Karegar, H., 2014. Optimal fast
charging station placing and sizing. Appl. Energy 125, 289–299.
Schroeder, A., Traber, T., 2012. The economics of fast charging infrastructure for electric
Acock, A.C., 2005. Working with missing values. J. Marriage Fam. 67 (4), 1012–1028. vehicles. Energy Policy 43, 136–144.
Algina, J., Swarninathan, H., 2011. Centering in two-level nested designs. In: Hox, J.J., Shahraki, N., Cai, H., Turkay, M., Xu, M., 2015. Optimal locations of electric public
62
J. Globisch, et al. Transport Policy 81 (2019) 54–63
charging stations using real world vehicle travel patterns. Transport. Res. Transport aversion on risk taking: an experimental test. Q. J. Econ. 112 (2), 647–661.
Environ. 41, 165–176. Trommsdorff, V., 1975. Die Messung von Produktimages für das Marketing: grundlagen
Sierzchula, W., Bakker, S., Maat, K., van Wee, B., 2014. The influence of financial in- und Operationalisierung. In: Annales Universitatis Saraviensis, vol. 78 (Köln [etc.]:
centives and other socio-economic factors on electric vehicle adoption. Energy Policy Carl Heymanns).
68, 183–194. Wang, G., Xu, Z., Wen, F., Wong, K.P., 2013. Traffic-constrained multiobjective planning
Sovacool, B., Axsen, J., Kempton, W., 2017. The future promise of vehicle-to-grid in- of electric-vehicle charging stations. IEEE Trans. Power Deliv. 28 (4), 2363–2372.
tegration: a sociotechnical review and research agenda. Annu. Rev. Environ. Resour. Weiber, R., Mühlhaus, D., 2014. Strukturgleichungsmodellierung: Eine anwendungsor-
42, 377–406 ISSN 1543-5938. ientierte Einführung in die Kausalanalyse mit Hilfe von AMOS, SmartPLS und SPSS
Sun, L., Huang, Y., Liu, S., Chen, Y., Yao, L., Kashyap, A., 2017. A completive survey study (2., erw. und korr. Aufl. 2014). SpringerLink : Bücher. Springer Gabler, Berlin,
on the feasibility and adaptation of EVs in Beijing, China. Appl. Energy 187, 128–139. Heidelberg.
Sun, X.-H., Yamamoto, T., Morikawa, T., 2016. Fast-charging station choice behavior Wingerter, C., 2014. Allgemeines umweltbewusstsein. Zusammenstellung so-
among battery electric vehicle users. Transport. Res. Transport Environ. 46, 26–39. zialwissenschaftlicher Items und Skalen. https://doi.org/10.6102/zis208.
Thaler, R., Tversky, A., Kahneman, D., Schwartz, A., 1997. The effect of myopia and loss
63