1 s2.0 S1361920916306058 Main PDF

Transportation Research Part D 58 (2018) 208–224
Contents lists available at ScienceDirect
Transportation Research Part D

journal homepage: www.elsevier.com/locate/trd
How open are Canadian households to electric vehicles? A national

T
latent class choice analysis with willingness-to-pay and
metropolitan characterization
⁎
Mark Fergusona, Moataz Mohamedb, , Christopher D. Higginsc, Elnaz Abotalebia,
Pavlos Kanarogloud
a
McMaster Institute for Transportation and Logistics, McMaster University, Canada
b
Department of Civil Engineering, McMaster University, Canada
c
Department of Land Surveying and Geo-Informatics/Building and Real Estate, The Hong Kong Polytechnic University, Hong Kong
d
School of Geography and Earth Science, Founding Director of McMaster Institute for Transportation and Logistics, McMaster University, Canada
AR TI CLE I NF O AB S T R A CT
Keywords: This paper reports on results developed from a 2015 national survey of Canadian consumers that
Electric vehicles sought to assess attitudes and preferences towards consumer electric vehicles. A latent class
Stated preferences discrete choice model was developed based on stated preferences choices. Four classes emerged
Latent class choice model with each being oriented to one of the primary vehicle technologies considered. The dominant
Attitude
characteristics of the Internal Combustion Engine (ICE)-oriented class are purchase price sensi-
Willingness-to-pay
tivity, EV scepticism and an apparent resistance to change; for the Hybrid Electric Vehicle (HEV)-
oriented class it is reluctance to plug-in and an unusual combination of high environmental
concern and an acceptance to burn gasoline; for the suburban-oriented Plug-in Hybrid Electric
Vehicle (PHEV)-oriented class it is measured optimism about plugging-in combined with an
orientation to a replacement vehicle for the next purchase; and for the younger and most urban
Battery Electric Vehicle (BEV)-oriented class it is the highest optimism about electric vehicles and
a focus on positive aspects such as rapid acceleration and minimized maintenance costs. By or-
ientation of household mindset, approximately 40% are ICE, 30% are PHEV, 20% are HEV and
10% are BEV. These results suggest considerable openness to electric vehicles. Willingness-to-pay
for vehicle and charging attributes and incentives were calculated and are highly useful in in-
terpreting the latent classes. The results feature interesting geographical variation which is
captured at the level of Canadian metropolitan areas.
1. Introduction
Despite apparent economic, energy, and environmental benefits, Electric Vehicles (EVs) are only very slowly beginning to gain a
foothold in the global auto market (Al-Alawi and Bradley, 2013; Klöckner et al., 2013). As of mid-2017, a cumulative total of
approximately 35,000 new EVs “Battery electric vehicles (BEV) and Plug-in Hybrid Electric Vehicles (PHEV)” have been sold in Canada in
contrast to approximately 650,000 EVs in the United States. On a per capita basis, the U.S. market is approximately 2.5 times more
advanced than the Canadian one. A majority of the Canadian sales have been supported by purchase price incentives that have been
administered in the most populated provinces: Quebec, Ontario and British Columbia. Lagging EV sales in Canada are more
⁎
Corresponding author at: Department of Civil Engineering, McMaster University, JHE-301, 1280 Main Street West, ON L8S 4L8, Canada.
E-mail address: mmohame@mcmaster.ca (M. Mohamed).
https://doi.org/10.1016/j.trd.2017.12.006
Available online 19 December 2017

1361-9209/ © 2017 Elsevier Ltd. All rights reserved.
M. Ferguson et al. Transportation Research Part D 58 (2018) 208–224
noteworthy when it is considered that the national electricity generation profile is one of the cleanest in the world. Certainly there is
evidence that dirty electricity generation will, to a varying extent, defeat the purpose of EV adoption (Holland et al., 2015) but almost
all developed countries are below the accepted 600 TCO2e/GWh threshold (Kennedy, 2015). Canada is far below this threshold
(167 tCO2e/GWh). Canada makes for an interesting laboratory to study the consumer landscape for EVs. From coast-to-coast Canada
covers 4 ½ time zones. It has two official languages, a diversity of ethnic groups, nearly forty metropolitan areas of varying sizes,
regionally distinct weather patterns and an array of different terrain.
There have been extensive efforts to understand the adoption of EVs in varying contexts and using different methods. A clear
pattern of research activities has emerged in the literature whereby consumer adoption/rejection of EVs is assessed on the basis of
two broad theoretical foundations: economic theories of preference utilitarianism and behavioural theories (Axsen et al., 2015;
Rezvani et al., 2015; Schuitema et al., 2013). The more frequent stream of research is rooted in preference utilitarianism theory and
deals with the adoption of EVs as a rational choice process (Dumortier et al., 2015; Rezvani et al., 2015; Schuitema et al., 2013)
linked to utility maximization (Buskens, 2015).
On the other hand, there is a growing stream of research that relates the adoption of EVs to behavioural aspects (Anable et al.,
2011; Mohamed et al., 2016; Wang et al., 2014). Here, there is a more pronounced focus on consumers’ personal beliefs, personality,
perception and emotion as they relate to EV adoption. Other approaches have included the theory of planned behaviour (Egbue and
Long, 2012; Mohamed et al., 2016; Moons and De Pelsmacker, 2012; Wang et al., 2014), normative theories (Moons and De
Pelsmacker, 2012), consumer innovativeness (Schuitema et al., 2013), Diffusion of Innovation (DOI) theory (Morton et al., 2016) life-
style theory (Axsen et al., 2015) and grounded theory (Caperello and Kurani, 2012; Graham-Rowe et al., 2012) to link EVs adoption
behaviour to environmental, attitudinal, symbolic, emotional, and societal factors.
Evidence from previous studies provides a sketch of the characteristics of EV adopters, although it has been argued that a rather
fragmented picture in the understanding of EV adoption has emerged (Rezvani et al., 2015). EV adopters have been revealed as
middle-to-high income (Anable et al., 2011), environmentally concerned middle-aged households (Hidrue et al., 2011; Moons and De
Pelsmacker, 2012; Wang et al., 2014; Ziegler, 2012), and with relatively higher education (Hidrue et al., 2011) and full time em-
ployment (Plötz et al., 2014). It has been also argued that males are showing more interest in EV technology (Anable et al., 2011;
Plötz et al., 2014). A clear distinction is identified in the profile of consumers interested in PHEV and BEV as well (Anable et al., 2011;
Plötz et al., 2014). Although these assessments hold part of the truth, it can be argued that there are some fundamental concerns.
Firstly, the adopted theoretical perspective, and its procedural tools, can influence the identification of consumers’ reaction to EVs
(Axsen et al., 2015). The output from a quantitative choice model will not necessarily be confirmed by a qualitative modelling of
behavioural aspects for the same users. Although both are aimed at depicting the heterogeneity of consumer adoption of EVs, they
represent polar ends of a continuum in approach (Mohamed et al., 2016; Schuitema et al., 2013). Secondly, the circumstances of any
behavioural/choice decision vary significantly across contexts. This justifies the imperative need for approaches that capture such
variation, and incorporate both preferences and beliefs in a single model.
While not pretending to offer the final answer on such broader methodological questions, this study aligns with preference
utilitarianism and uses a particular implementation of the latent class choice model, which is sensitive to how attitudes shape EV
preferences. This paper’s approach enhances the latent class choice approach of prior EV applications (Axsen et al., 2015; Hidrue
et al., 2011) by following the method of Beck et al. (2013) that was applied in assessing the impacts of emissions charges on vehicle
choice. In that study, specific Likert-based statements, intended to measure pertinent beliefs, enter directly into the class probability
model along with respondent demographics to help allocate choice makers into latent classes. As such, Likert-based indicator
variables do not enter into the class-specific sub-models in which utilities of the choice alternatives are estimated.
In comparing the approach of Beck et al. (2013) to EV studies that utilize the latent class choice model, some differences are worth
noting. Axsen et al. (2015) uses a sequential approach to form constructs that relate to technology orientation, environmental lifestyle
and an indicator of openness to new lifestyles and these constructs enter into the class membership model. Hidrue et al. (2011) avoid
the issue of forming attitudinal constructs by asking respondents only about past observable changes in behaviours that relate to the
environment. Also, their focus is essentially on BEVs versus conventional vehicles which ultimately results in only two latent classes.
Though not focused on attitudes or EVs, Train (2008) carries out a methodologically oriented study where a latent class choice model
assesses alternative vehicles powered by hydrogen.
It is worth highlighting other advanced discrete choice approaches that are applicable in the EV context. The mixed logit model
has been applied frequently in the EV literature (Mabit and Fosgerau (2011), Hackbarth and Madlener (2013), Tanaka et al. (2014)
but these implementations have not offered a mechanism to assess attitudes and beliefs and the general approach has been criticized
as harder to interpret (Axsen et al., 2015). Beck et al. (2013) offer a good discussion of the advantages of the latent class choice model
over the mixed logit and Shen (2009) has undertaken a comparison between the mixed logit model and the latent class choice model
that reflects favourably on the latter. The latent class choice model also does not suffer from the Independence of Irrelevant Alter-
natives (IIA) property that the multinomial logit model possesses and which the nested multinomial logit model seeks to address
(Potoglou and Kanaroglou, 2007).
The hybrid choice model (Bolduc et al., 2008), which has been less frequently used to this point, seeks to generate representative
latent attitudinal constructs from a series of indicator variables that directly enter into the utility functions of choice alternatives. The
hybrid choice model is conceptually more complex than the latent class model and unlike the latter, does not segment choice makers
into discrete classes. In the latent class context, these offer rich potential for interpretation and comparison.
For the current study of Canadian households, the purpose is to understand the preferences for the two primary types of electric
vehicles; Plug-in Hybrid Electric Vehicles (PHEVs) and Battery Electric Vehicles (BEVs) relative to each other and against Hybrid
Electric Vehicles (HEVs) and conventional Internal Combustion Engine vehicles (ICE) and in a way that is sensitive to household
209
demographics, attitudes; household vehicle ownership traits; vehicle, charging and government incentive attributes; and the geo-
graphic context in which households shape their preferences.
To this point, geographical variation in preferences has not featured prominently in past choice models developed to assess
prospects for electric vehicles and we are not aware of a study where a representative national sample is assessed for sub-national
geographical variation as is done in this study. Most samples that have been collected focus on being broadly representative of the
associated country in demographic terms (e.g. Hidrue et al. (2011) for the U.S., Hackbarth and Madlener (2013) for Germany) and
there have been studies that compare countries (Helveston et al., 2015; Tanaka et al., 2014). Some are more focused on particular
regions within a country (e.g. Lebeau et al. (2012) for Flanders a region within Belgium. Potoglou and Kanaroglou (2007) study a
medium-sized urban area. In this study, a household’s locational position in the urban-rural hierarchy has been explicitly taken into
account in assessing choice of powertrain and the effects isolated against natural demographic variation across geographic space.
Also related to geography, the study illustrates sub-national variation in powertrain mindsets with a comparison across approxi-
mately 40 Canadian metropolitan and urban areas.
This case study in Canada utilizes a comprehensive survey of 20,520 Canadian households (17,953 were validated and used) to
offer a detailed picture of how Canadian consumers in mid-2015 are evaluating their vehicle powertrain options and how open they
are to electric vehicles. There are three main contributions of this paper: (1) A set of useful and interesting results for Canada are
generated in a national modelling context; (2) this study offers more in-depth coverage of geographical variation in EV preferences
than has been seen in prior studies; (3) the particular latent class choice modelling approach seen in Beck et al. (2013) is being
utilized here in the EV choice context. In addition, a novel aspect with the experimental design of the present study is that households
are asked to evaluate vehicle attributes that are appropriate for the specific vehicle body type that they would prefer for their next
vehicle. As such, luxury car buyers, for example, are not asked to evaluate attribute levels that might be more appropriate for an
economy car buyer (Higgins et al., 2017) and the choice scenarios that are presented to respondents are more relevant to individual
household contexts than in most past studies.
The paper is separated into six major sections. After this introduction, a second section on methods and data describes further
details on the survey, data collected, the stated preference exercise, and the choice model to be implemented. A third section on
model results provides a summary of the important insights emerging from the choice model and describes the nature of the four
latent classes that are derived. A fourth section illustrates how latent class membership varies by metropolitan area in Canada. The
paper ends with a discussion of results as the fifth section and finally a concluding section.
2. Methods and data
2.1. Survey data
The data for this analysis were captured from an on-line survey that was offered to members of a Canadian survey panel. A sample
of 20,520 households was captured. The survey was offered in both English and French (Canada’s two official languages) to achieve
national representation. The design was stratified in the sense that proportionally more observations were sampled from the less-
populated provinces. The aim was to ensure adequate counts of household observations in as many of the ten provinces as possible
given the constraints of the survey panel. Canada’s lightly populated northern territories were excluded from the final data collection
due to inadequate representation in the survey panel. Fig. 1 summarizes the final observation counts by province and highlights the
map pattern.
The survey was aimed at characterizing households as opposed to individual respondents and this was done through surveying a
representative of the household. While Hensher et al. (2010) provide evidence that the best approach is probably to consult with
multiple decision-makers in the household, we considered that our most promising option was to highlight to the respondent their
role as household representative and to re-emphasize that concept during the survey. The household perspective is important con-
sidering the high expense to acquire an automobile and also empirical evidence for early adopters that EVs (hereafter, EV is used for
PHEV and BEV) are acquired largely in multiple vehicle households (Klöckner et al., 2013). Respondents were screened on the basis
of whether they were aged 18 or over and whether they saw themselves as playing a role in important household financial decisions
(such as purchasing a vehicle). One implication is that there are relatively few 18–24 years old respondents in the sample because
there are few household heads in this age group.
Apart from a focus on households, the survey was general in its sampling frame on the grounds that a research objective was to
characterize a wide array of Canadians. Households were not excluded on the basis of urgency or recency of a vehicle purchase. Mabit
and Fosgerau (2011) and Hackbarth and Madlener (2013), for example, screened on such criteria. The former screened also on new
car buyers as did Axsen et al. (2015). Conversely, Tanaka et al. (2014) and Qian and Soopramanien (2011), as examples, focused on
general consumers and the latter included households presently without a vehicle as is the case with our sample. The only households
excluded from this sample were those identified as “Not at all likely” to purchase or lease another vehicle in the foreseeable future.
Overall, a final validated sample of 17,953 participants were used in the analysis.
In Table 1 the households surveyed are compared to census distributions across a range of variables and we see that the sample
has less representation from one-person households and those less educated. Partly such households are not found in the sample
because they are associated with the screened-out group having little intention to acquire a vehicle and partly the highly-educated are
more likely to participate in survey panels. We see the sample as reasonably representative of prospective vehicle-acquiring
households in Canada.
Apart from the initial screening section, the survey was structured into six other sections: household vehicle inventory; household
210
Fig. 1. Distribution of respondents by province.
vehicle purchase plan; educational material on vehicle technologies and their attributes; stated preference experiments for next
household vehicle; a suite of Likert based statements on attitudes and beliefs; and finally, household spatial, demographic, and
socioeconomic characteristics. As such, a long list of potential explanatory variables was developed to assess preferences towards
vehicle powertrains.
The Likert-based statements used in the survey were included to facilitate different analyses. They support the structural equation
models developed in (Mohamed et al., 2016) and they also support the latent class choice implementation of this study. The
statements were chosen to assess the Theory of Planned Behaviour (Ajzen, 1991) in the Canadian electric vehicle context. As such,
and as seen in Mohamed et al. (2016), there are statements that deal with environmental concern, attitudes towards EV adoption,
subjective norm, personal moral norm, perceived behavioural control and EV adoption intention.
The survey went through a feedback phase where two pilot phases were executed with respondents from the survey panel and this
permitted adjustments to wording of questions and also suggested opportunities for further questions. One other important ob-
servation about the household vehicle inventory collected as part of the survey of Canadian households was that it captured only a
handful of current EV owners. As such, the modelling exercise associated with this study is necessarily stated preference in nature
with no opportunity to employ a model that utilized both stated and revealed preference data.
2.2. Stated preference exercise
The centrepiece of the survey was a stated preference (SP) exercise which consisted of an educational section, so that respondents
could be made familiar with the characteristics of each powertrain and associated attributes, and then a series of four scenarios of the
type shown in Fig. 2. A given scenario offered versions of four main vehicle types (ICE, HEV, PHEV, and BEV). This was a “labelled”
choice experiment (Rose and Bliemer, 2009), meaning that the identities of alternatives were clear to the respondents and possibly
even more so subsequent to their study of the educational section. Indeed, previous research has shown that vehicle buyers may be
unfamiliar with EVs and their attributes (Anable et al., 2011; Caperello and Kurani, 2012; Kurani et al., 1994) and this reinforces the
need for an educational section to be prominent in the SP exercise.
In Table 2, some important aspects of past relevant SP experiments are outlined. Several of the studies feature samples that are
national in scope, though the sample sizes that seek to achieve national coverage have typically been relatively small. A trend toward
larger samples seems apparent in the more recent studies. Choice-making entities are often seen as households but in equally as many
cases the samples represent individual consumers. Ewing and Sarigöllü (1998) quite specifically target suburban commuters in a large
metro. A considerable range of powertrain/fuel types are utilized across studies with as few as 3 and up to 7. Experiments in Europe
have tended to feature more powertrains likely because more of these types are observed extensively in practice. With the exception
of Potoglou and Kanaroglou (2007), EVs make up at least one vehicle type in each study.
211
Table 1
Surveyed households compared to Canadian households as derived from census.
Variable Value Frequency Percentage Canadian census
Age of Household Head (Years) 18–24 815 4% 4%

25–34 2952 14% 16%
35–44 3642 18% 20%
45–54 4625 23% 25%
55–64 4537 22% 21%
65+ 3949 19% 14%
Language Spoken Most Often at Home English 15,786 77% 65%

French 3748 18% 23%
Chinese & other 986 5% 12%
Marital status of Household Head Single 5068 25% 18%s

Married or Common Law 14,210 69% 58%
Other 1242 6% 23%
Household Size 1 3720 18% 28%

2 9280 45% 34%
3 3464 17% 16%
4 2782 14% 14%
5 956 5% 5%
6 or more 318 2% 3%
Dwelling Type Single Detached House 13,438 65% 56%

Townhouse/Semi-Detached 2442 12% 11%
Apartment or Condo 4361 21% 33%
Other 279 21% 33%
Dwelling Tenure Owner 16,103 78% 69%

Renter 4417 22% 31%
Education Level of Household Head No certificate, diploma or degree 624 3% 17%

High school diploma or equivalent 3790 18% 22%
College, CEGEP, Apprenticeship trades certificate or diploma 6960 34% 37%
Univ. certificate, diploma or degree at bachelor level or above 9146 45% 23%
Household gross income Less than $25,000 943 5% 18%

$25,000–$49,999 3292 19% 23%
$50,000–$99,999 7498 43% 33%
$100,000 and more 5906 33% 26%
Prefer not to answer 2881 – –
While having EVs in the choice set has been common in such experiments for some time, the table illustrates that PHEVs have
been emerging frequently as explicit members in choice sets within the past five years. In prior years, BEVs were typically the
prototype electric vehicle.
With a choice set that is focused on HEV, PHEV and BEV, a tailored series of stated preference attributes can be identified to test
differences in perceptions about vehicle charging context or driving range based on how that range is powered. Certainly, there are
significant differences in the behaviours required from consumers to operate the available types of vehicles and associated power-
trains (Rezvani et al., 2015; Schuitema et al., 2013); hence, attributes associated with these vehicles will ideally be carefully con-
ceptualized.
It was stressed to respondents that they needed to take their time with respect to the SP exercise and, after Hidrue et al. (2011),
choose as though their own “hard-earned money were on the line.” Time spent on the total survey, the educational section, and the
actual SP exercise were tracked to help assess attentiveness. In the event that a respondent rushed through the educational section
while taking the survey, a cautionary message was shown urging again for them to proceed deliberately. The main purpose of the
educational section was to make sure that respondents understood the nature of the four vehicle types and all associated attributes.
The hybrid (HEV) option was positioned as a vehicle that would run on gasoline, never be plugged in and emit less. The battery
electric (BEV) option was discussed as being totally dependent on plugging in and electricity and with zero tailpipe emissions. The
plug-in hybrid (PHEV) option was portrayed as having the flexibility to run on electricity or gasoline and to thus have the potential to
emit much less.
Each SP screen included four main attribute categories: cost, operational, charging, and non-cash incentives for a total of 11
attributes. The categories were for visual effect and also it was seen as helpful for the respondents. The educational section presented
attributes and their meanings in these same four groups so that the entire presentation would be integrated. The four categories of
attributes were shown as in Fig. 2 for half of respondents and in reversed order (with cost attributes at the top and charging attributes
at the bottom) for the remaining half. Ordering of attributes within categories was not altered. This approach was motivated mostly
by a concern about having respondents focus excessively on purchase price in choosing their vehicles. Models estimated separately on
the two groups of respondents yielded generally insignificant differences in parameters. Some choice experiments vary ordering of all
attributes randomly but this approach was incompatible with our implementation.
212
Fig. 2. A typical stated preference scenario.
Our overall philosophy for the SP exercise was that the choice among vehicle powertrains is a complex decision with many
influencing factors. Simpler scenarios with fewer attributes could easily have been implemented. We opted for fewer scenarios (4 per
respondent) of a more involved type that best captured the choice criteria. Good visual design, reminders to pay close attention, and
tracking of time spent on survey sections were used in conjunction to best ensure that our sample was supported by careful eva-
luations. All 18 graphic icons on the SP screen featured informative pop-up text that was sensitive to “mousing over” and could be
reviewed in the event that respondents needed to refresh their memory about some aspect of a scenario.
One important note about the implementation of the experiment is that vehicle body type played a significant role in the con-
ceptualization. In particular, respondents were asked to select a preferred vehicle body type for their next vehicle (i.e. one of:
economy, intermediate sedan, full sedan, luxury sedan, minivan/crossover, SUV, pick-up truck) at the beginning of the experiment
and this selection applied to all scenarios. This approach contrasts with some studies (e.g. Maness and Cirillo (2012) and Potoglou
and Kanaroglou (2007)) where body type is included endogenously as a stated preference attribute or with other studies where body
type is not considered at all. For those households interested in luxury sedans, for example, it makes sense to present a scenario that
achieves some degree of realism on vehicle price and other attributes. For each of the seven vehicle body types, a base value for each
stated preference attribute was researched and identified using a widely-known ICE vehicle prototype.
Quantitative values displayed on the SP screen for some attributes were context sensitive to the respondent’s circumstances.
Values attached to fueling/charging costs, maintenance costs, and annual emissions were calculated on the fly, based on the annual
mileage reported by the respondent, and with consideration of a participant household’s preferred vehicle body type. In the event
that the there was no current household vehicle, an annual mileage of 15,000 km was assumed. This assumed mileage proved close to
the average annual mileage of vehicles actually reported through this survey.
Table 3 outlines assumed levels associated with the SP attributes and how these were implemented across powertrains. Most of
these are expressed in relative terms and realistic “base” values were associated with each of the seven body types by attribute. The
end-result is that SP scenarios were specific to each body type. In general, fewer levels were implemented for the ICE and HEV
alternatives as vehicles based on these powertrains have been in wide use for some time. Ultimately, the modelling exercise un-
dertaken below employed a data set that spans the seven vehicle body types. For each body type, 48 distinct scenarios were derived
213
M. Ferguson et al.
Table 2
Important aspects of selected past choice experiments on electric and alternative fuel vehicles.
Study Geography Target group Respondents Vehicle Types Explicit Model-type

PHEV?
Bunch et al. (1993) Southern California – South General Households 692 3 NO Nested Logit
Coast Air Basin
Ewing and Sarigöllü (1998) Montreal Suburbs, Canada Commuters 881 3 NO Multinomial Logit
Dagsvik et al. (2002) Norway General Consumers 922 4 NO Ranked and Serially Dependent
Random Utility
Potoglou and Kanaroglou (2007) Hamilton, Ontario, Canada Car buyers within next five years 482 3 and none EV NO Nested Logit
Mabit and Fosgerau (2011) Denmark Very recent new car buyers 2146 5 NO Mixed Logit
Hidrue et al. (2011) United States Future car buyers who expect to spend more 3029 2 – BEV and ICE NO Latent Class Choice Model
than $10K
214
Qian and Soopramanien (2011) China General Households including non-car owners 527 3 NO Nested and Multinomial Logit
Ziegler (2012) Germany Potential car buyers (at dealer/inspection sites) 598 7 NO Multinomial Probit
Maness and Cirillo (2012) U.S. State of Maryland General Households 1906 5 YES Mixed and Multinomial Logit
Hackbarth and Madlener (2013) Germany Car buyers who purchased within past year or 711 7 YES Mixed Logit
intend to purchase within a year
Ito et al. (2013) Japan General Consumers 1531 4 NO Nested and Multinomial Logit
Hoen and Koetse (2014) Netherlands Existing private car owners 1903 6 YES Mixed Logit
Tanaka et al. (2014) Japan and 4 U.S. states General Consumers 4202 in U.S. 4000 in 3 but excluding YES Mixed Logit
Japan HEV
Axsen et al. (2015) Multiple Canadian Provinces Households that acquired a vehicle new in past 1754 4 YES Latent Class Choice Model
5 years
Present Study Canada Households with interest to acquire a vehicle in 20,520 4 YES Latent Class Choice Model
future
Transportation Research Part D 58 (2018) 208–224
Table 3
Relative attribute levels used in choice scenarios.
Attributes ICE HEV PHEV BEV
Purchase Price ($) (1) Base (1) +50% of base (1) +100% of base (1) +100% of base
(2) −25% of base (2) Base (2) +50% of base (2) +50% of base
(3) Base (3) Base
Annual Maintenance Cost ($/km) (1) Base (1) Base (1) Base (1) −25% base
(2) −25% base (2) −25% base (2) −50% base
(3) −50% base (3) −75% base
Annual Fuel/Charging Cost ($) (1) +40% base (1) Base (1) −20% base (1) −60% base
(2) +20% base (2) −20% base (2) −40% base (2) −80% base
(3) Base (3) −60% base
Acceleration (0–100 km/h sec) (1) Base (1) +50% base (1) +50% base (1) +50% base
(2) Base (2) Base (2) Base
(3) −50% base (3) −50% base (3) −50% base
Tailpipe Emissions (tonnes/annum) (1) Base (1) −25% base (1) −50% base (1) −100% base
(2) −25% base (2) −50% base (2) −75% base
Cash Incentive ($) – – (1) $0 (1) $0
(2) $5000 (2) $5000
(3) $10,000 (3) $10,000
Gasoline Range (km) (1) Base (1) +25% base (1) +25% base –
(2) −25% base (2) −25% base
E-Range (km) – – (1) 30 (1) 150
(2) 60 (2) 250
(3) 90 (3) 350
Public Charging Time (h) – – (1) +80% base (1) +80% BEV base
(2) Base (2) +100% base (BEV base)
(3) −80% base (3) −80% BEV base
Home/Work Charging Time (h) – – (1) Base (1) +100% base
(2) −25% (2) +50% base
(3) −50% (3) Base
Battery Warranty – – (1) 3 yr/58,000 km (1) 3 yr/58,000 km
(2) 5 yr/96,000 km (2) 5 yr/96,000 km
(3) 8 yr/160,000 km (3) 8 yr/160,000 km
Fuel/Charging Availability (1) Base (1) Base (1) +100% base (1) +100% base
(2) Base (2) Base
(3) −90% base (3) −90% base
that were implemented as 12 blocks of four scenarios. A given respondent was presented with scenarios from one of those blocks
taken from the scenarios of their preferred body type. An efficient design approach was pursued via Ngene Software to derive an
appropriate array of scenarios for each respondent.
In terms of how attributes were conceived and implemented for this experiment, there are a few important points to highlight:
• While purchase price and governmental cash incentives appear on the SP screen as a single line item, they were implemented in
the experimental design as two distinct attributes. Non-zero cash incentives applied only to PHEV and BEV.
• In past studies, vehicle range has been implemented as a single attribute. Here, we viewed it important to distinguish between
gasoline range (the distance on a full tank) and electric range (the distance on a full charge). BEVs show up as having zero gasoline
range, ICE and HEV are shown as having zero battery range and PHEV is the only option with a non-zero range on both.
• Estimated annual fuel costs are combined with annual electricity costs in a single item. This attribute is portrayed as purely
financial and is affected by gasoline and electricity prices. The range attributes are most closely aligned to the actual behaviour of
travel and generating emissions. Dagsvik et al. (2002) have noted the distinction between fuel consumption/mileage and fuel cost
for powertrain choice experiments. Fuel costs are captured explicitly here and mileage is largely captured through the range
variables.
• Emissions are shown in absolute tonnes emitted based on annual mileage and they are tailpipe emissions which effectively implies
zero emissions for BEV. Clearly, well-to-wheel emissions are not zero in many jurisdictions for BEV, but the concept of zero
tailpipe emissions is central for the consumer to understand the core of the BEV proposition.
• Public charging times reflect the charging time using Level 2 and Level 3 chargers (assuming 80% and 20% utilization for each
type respectively).
• With regard to the locational availability of public charging infrastructure, this is being measured relative to existing gasoline
infrastructure as has been common in the past (Brownstone et al., 2000; Tanaka et al., 2014). Hoen and Koetse (2014) chose to
evaluate required detour times to access electric charging.
• The scenarios consider three types of non-cash incentives that were made specific to PHEV and BEV. For a given vehicle in a
scenario, all three incentives could be active, none could be active or both possibilities in between. When incentives have been
included in past studies, implementations (e.g. Qian and Soopramanien (2011)) have tended to rotate such incentives on a single
attribute line.
215
After omitting respondents who spent less than 3 min on the SP section (educational section and choice scenarios), 17,953
households were retained from the original set of 20,520 for the choice modelling exercise to follow.
2.3. Latent class choice model
The latent class choice model can be expressed as:
S
Pij = ∑ Pij | s His
s=1
where Pij denotes the unconditional probability that household i selects powertrain j. This unconditional probability is a discrete
mixture of S distinct conditional probabilities Pij|s, where S denotes the total number of classes, each of which is weighted by a class
membership probability His. The S conditional probabilities Pij|s, relating to household i and powertrain j, are multinomial logit
probabilities for each class while the single class probability His is also typically implemented in a multinomial logit form, though this
is not a necessity. The interested reader is encouraged to consult Greene and Hensher (2003) for more details on the model including
the composition of the log-likelihood function and how statistical parameters are derived to vary across households.
Note that there is no observed choice behaviour relating specifically to the class probability model. Prior probabilities of per
household class membership are derived as part of the maximizing the log-likelihood function. After the fact, posterior probabilities
can be derived and these produce class probabilities that more closely reflect powertrain choices made by the respondent across their
several choice trials. For some households, posterior class probabilities suggest a very strong association with a single class, while for
others no single class may stand out.
Like the mixed logit model, the latent class version allows statistical parameters to vary across choice makers, which is essential to
capturing preference heterogeneity, but the latent class model blends discrete segments to achieve this outcome. Each discrete
segment is associated with its own bundling of unique statistical parameters and a unique set of preference tendencies that can be
characterized and described. Each segment can be described in isolation and also in terms of how it relates to other derived segments.
A given segment is not some abstract construct. Some choice makers in a sample can be very heavily oriented to one segment and
their choice preferences and associated utility parameters essentially described by that corresponding segment. In the case of the
mixed logit model, variation of parameters across choice-makers is achieved through the use of continuous distributions. Beck et al.
(2013) note that in many cases the assumption that preferences vary continuously over the sample may cause distortions and that it is
much more difficult to assess how specific choice makers relate to distributional results per explanatory variable.
The latent class choice approach produces distinct parameter vectors on a class-by-class basis. Each parameter vector contains
insight about how associated households derive utility in evaluating the powertrain alternatives. Households are not deterministically
associated with any one class but rather are probabilistically assigned across classes.
3. Model results
The best-performing model that could be developed consisted of four latent classes. Note that the number of classes is not
determined endogenously and it is required to run a series of models to determine a best number of classes. Each class leans to one of
the four powertrains in the choice model. As such, there is a class oriented to ICE, HEV, PHEV, and BEV, and each is associated with a
unique evaluation of the powertrain options. All components of the latent class choice model are estimated simultaneously.
In Table 4 results for the Bayesian Information Criterion (BIC) and the Akaike Information Criterion (AIC) are reported. Both show
improvement as the number of classes increases indicating that each incremental class is adding value. The improving log-likelihood
value and increased number of statistical parameters are shown also. It did not prove possible to achieve a stable calibrated result
with five or more classes. A four-class solution was chosen over three primarily because all four classes, when interpreted, appeared
to make a great deal of sense. Note that the model described is based on a weighted estimation. Weights were applied to correct for
the fact that some smaller provinces were over-sampled in the survey.
The further discussion on model results is separated into three sub-sections. Following the approach of Hidrue et al. (2011), we
initially discuss the results of the class probability sub-model followed by those of the class-specific utility sub-models. The discussion
covers the important results across explanatory variables. The nature of the specific variables employed are generally self-ex-
planatory, but to the extent they are not, descriptions are provided. The final results sub-section offers and overview description of
the latent classes that have been developed.
Table 4
Statistics to assess best number of latent classes.
Classes k LL N AIC BIC
2 50 −70,349 17,953 140,798 141,188

3 84 −64,151 17,953 128,470 129,125
4 118 −61,841 17,953 123,918 124,838
216
Table 5
Class probability model results.
Latent class
Reference: HEV-oriented class ICE-oriented PHEV-oriented BEV-oriented

Variable Coefficient Coefficient Coefficient
Constant 5.5366*** −0.9714*** −5.6769***
Socioeconomic and Demographic

Age of Respondent −0.0027 −0.0186*** −0.0152***
Respondent is Female −0.2402*** −0.1426** −0.1371*
Household Education Score −0.0879*** 0.0921*** 0.0651**
French Primarily Spoken at Home 0.3966*** 0.1791*** 0.4807***
Vehicle Ownership
Time to Next Vehicle Purchase (yrs) −0.0411*** 0.0759*** 0.0612***
Next Vehicle is to be an Incremental Vehicle −0.0626 −0.0106 0.5445***
Proportion of Current Household Vehicles that are HEV −2.8706*** −0.4503** −0.5772**
Next Vehicle is to be Economy Car 0.3100*** 0.1361* 0.5855***
Attitudinal
Q1. I am willing to spend more money to buy an EV −0.5301*** −0.0920*** −0.0846**
Q2. In the long-term, I think owning an EV is more cost effective than owning a conventional vehicle −0.2541*** 0.1448*** 0.4545***
Q3. I am willing to tolerate some periodic battery charging inconvenience for the benefits of driving an EV 0.0393 0.4153*** 0.7286***
Q4. Plugging in an EV at my home is not practical −0.0325* −0.2548*** −0.2586***
Q5. With an excellent battery warranty, I would not worry about buying an EV −0.1025*** 0.1488*** 0.3576***
Q6. I feel that I ought to buy an EV to help replace uncertain foreign oil with made-in-Canada electricity −0.2234*** 0.0641** 0.1348***
Q7. Importance factors for next hhld vehicle “Reduced tailpipe emissions” −0.2283*** −0.0834*** −0.1288***
Q8. Importance factors for next hhld vehicle “Excellent fuel economy” −0.1340*** 0.0721* 0.1123**
Locational
Urban-Rural Index (0–10) −0.0065 0.0429** 0.1320***
* Statistical significance at the 10% level.

** Statistical significance at the 5% level.
*** Statistical significance at the 1% level or better.
3.1. Class probability sub-model
The specification of the class probability model, which is implemented in a multinomial logit form, is based on a wide range of
explanatory variables all of which vary over respondent households. Key components relate to demographics, vehicle ownership
characteristics, attitudinal variables and locational context. These elements have a significant influence on class membership as is
outlined in Table 5, note that the reference latent class is the HEV-oriented class.
Some of the most important results from the class probability sub-model are as follows. There is a clear outcome that younger and
better-educated households are more likely to be assigned to the BEV and PHEV classes. With regard to the education variable, note
that it was implemented as a household concept. An ordinal variable, based on six levels of educational achievement, was averaged
across results for adults in the household. Female respondents are more likely to possess an HEV-orientation and are least likely to
possess an ICE-orientation. With regard to language spoken at home, results suggest that French-speakers are less likely to possess an
HEV-orientation, which implies that the predominately French-speaking province of Quebec is less open to this means of vehicular
travel to reduce emissions.
A strong result emerges that those seeking an incremental vehicle are far more likely to be assigned to the BEV-oriented class.
Note that an acquisition that increases the household fleet from 0 to 1 vehicles is also considered incremental. The PHEV-oriented
class receives no such boost from the planned purchase of an incremental vehicle, indicating that PHEVs are seen as more viable as a
replacement vehicle. There is a strong result that those who currently possess one or more HEV vehicles are unlikely to have an ICE-
orientation. This would not mean that an ICE vehicle would never be acquired again but results do suggest greater affinity to green
vehicles. Another interesting outcome is that if the household prefers to acquire an economy vehicle next, there is a boost in the
likelihood of a BEV-orientation. There is also a lesser boost for the ICE-orientation. The results suggest that despite the best efforts of
BEV maker Tesla, with its larger vehicles, BEVs are still appealing more to those interested in small, economy vehicles. But many who
opt for economy vehicles are still seeing ICE vehicles as a good choice.
A 0–10 urbanization index that was developed for household locational context indicates that the BEV-orientation is strongly
urban while a PHEV-orientation is more moderately urban and is also oriented to suburban areas.
Attitudinal characteristics play an important role in defining and differentiating the classes. Several such statements are utilized
and there are some noteworthy outcomes. Q1 resonates most clearly with the ICE-oriented who appear averse to the concept of
spending more money to buy an EV. Those who are unwilling are far more likely to be ICE-oriented. Through other statements, these
households also appear less attuned to reducing emissions (Q7), getting good fuel economy (Q8) or showing concern with the
sourcing of foreign oil (Q6). Through Q2, the BEV-oriented are revealed to more likely accept the idea that EVs (BEV and PHEV) are
cost-effective over time while the ICE-oriented are sceptical. Through Q3 there is an important result that having a tolerance for some
217
Table 6
Results from MNL model, class utility models, and WTP.
MNL ICE-oriented HEV-oriented PHEV-oriented BEV-oriented
Cost
Purchase Price ($000’s) −0.0417*** −0.1282*** $1000 −0.0379*** $1000 −0.0650*** $1000 −0.0496*** $1000
Cash Incentive ($000’s) 0.0373*** 0.0778*** $607 0.0217** $573 0.0586*** $902 0.0505*** $1019
Maintenance Costs ($100’s) −0.0136*** −0.0204*** −$159 −0.0259** −$684 −0.0205*** −$315 −0.0557*** −$1124
Operational
Accel (0–100 km/h) (sec) −0.0114*** −0.0286*** −$223 −0.0017 −$44 −0.0153*** −$235 −0.0217*** −$437
Gasoline Range (km) 0.0004*** 0.0008*** $6.24 0.0008*** $21.91 0.0006*** $9.69 0.0002 $4.44
Electric Range (km) 0.0008*** 0.0017*** $13.18 0.0006 $16.89 0.0020*** $31.22 0.0015*** $29.86
Non-Cash Incentives
Free Municipal Parking 0.0724*** 0.1765*** $1376 −0.1628** – 0.1192*** $1832 0.0970*** $1958
No Tolls −0.0020 0.1567*** $1222 0.1132 $2987 −0.0386 – 0.0532 $1074
HOV access 0.0587*** 0.0652 $509 −0.1177 – 0.1473*** $2265 0.0556 $1121
Battery Warranty 0.0717*** 0.1190*** $929 0.1195*** $3153 0.1497*** $2302 0.0397* $800
Charging
Charge Station Availability 0.0311*** −0.0012 – 0.0120 $317 0.0451** $694 0.0556** $1122
Public Charging Time (h) −0.0374*** −0.0681*** −$531 −0.0747*** −$1971 −0.0266*** −$409 −0.0366*** −$738
Home/Work Charging (h) −0.0343*** −0.0223 −$174 −0.0386 −$1020 −0.0052 −$80 0.0664*** $1341
ICE constant 0.7020*** 3.0160*** $23,528 1.0842*** $28,614 −0.0133 −$205 −2.2124*** −$44,642
HEV constant 0.4025*** 1.1443*** $8927 3.2092*** $84,698 0.8663*** $13,322 −1.9140*** −$38,620
PHEV constant 0.4303*** 0.4453*** $3474 1.0982*** $28,983 2.0936*** $32,194 −0.9374*** −$18,914
Log-Likelihood at Convergence −88,371 −61,841

McFadden’s rho-squared 0.11 0.38
* Statistical significance at the 10% level.

** Statistical significance at the 5% level.
*** Statistical significance at the 1% level or better.
battery charging inconvenience is most associated with the BEV class and to a lesser extent with the PHEV class. The ICE and HEV
classes are intolerant in this regard. While the HEV-oriented are not attuned to plugging in, results for Q7 suggest that they are most
concerned with reducing tail pipe emissions. Through Q5 it is evident that deriving comfort from a good battery warranty is more
associated with membership in the BEV class and to a lesser extent in the PHEV class. Statements on the importance of environmental
responsibility and attitudes toward climate change were tested as well but had minimal power in allocating households to the latent
classes.
3.2. Utility sub-models by latent class
Each latent class has its own unique character as to how changes in vehicle attributes affect utility and these aspects are captured
in a utility sub-model per class (Table 6). This table shows modelled parameters and willingness-to-pay results as well as a multi-
nomial logit result for the sake of comparison. Differing willingness-to-pay (WTP) patterns are evident within and between the four
latent classes. The critical attribute on which WTP calculations are based is vehicle purchase price. WTP results are in dollars and
ultimately it becomes possible to assess what increases in purchase price would be tolerated by households for an improvement in a
vehicle attribute.
The alternative-specific constants associated with each latent class play a significant role in defining the classes. These capture
components of systematic utility that cannot be specifically linked to changes in the other model attributes. Since these are all part of
the same model, their relative magnitude can be compared within and between latent classes. In Table 6 the constants are converted
into WTP terms. Results are indicative of the base level of attractiveness that each class associates with the four powertrain options,
other things being equal. As such, the ICE and HEV classes appear least enthusiastic about BEVs. The PHEV class may prefer HEVs to
BEVs. Interestingly, the BEV class appears almost as negative on HEVs as they are for ICE.
We focus initially on cost and incentive attributes. Purchase price is strongly negative and significant for each latent class sug-
gesting, not surprisingly, that utility decreases as prices increase. The ICE class is by far the most purchase price sensitive. A related
observation is that high purchase price sensitivity for the ICE-oriented will translate into this class having a lower WTP for other
attributes. Annual vehicle maintenance costs are negatively associated with utility as expected but the results are strongest for the
BEV-oriented. A member of the BEV class would accept an $1124 increase in vehicle purchase price to save $100 per year on
maintenance costs. Results for the cash incentive variable, which was notably made specific to the PHEV and BEV choice alternatives,
are strong but the WTP results suggest, for example, that ICE and HEV-oriented households mentally discount the value of a $1000
cash incentive down to about $600. Results for non-cash incentives are generally significant but not as strong as cost factors. There
are several cases where a class would pay between $1000 and $2000 for exemptions on municipal parking fees or tolls or to have
access to an HOV lane. For the latter, the PHEV class, which is quite suburban, will pay $2265 extra for a vehicle to obtain access to
faster highway travel. For the battery warranty attribute, which combines three combinations of duration and mileage, there is the
218
interesting result that the BEV-oriented will pay the least ($800) to improve the warranty by one level while the HEV-oriented will
pay the most ($3153).
For the operational and charging attributes, there are a few key outcomes. Good performance, as measured by rapid acceleration
from 0 to 100 km/h, is well-received and in particular the BEV-oriented would pay $437 in increased purchase price per second saved
while the HEV class appears least concerned with this aspect. Vehicle range for the BEV and PHEV classes is most valued when
powered by electricity and the averaged result is about $30 per km of extra range. The BEV class values gasoline range as not different
from zero while the HEV-oriented class highly values gasoline-powered range at about $22 per km. Overall, the classes think dif-
ferently about range depending on how it is fueled. Time spent charging at public stations is viewed much more negatively than the
time spent charging at home or work and this is particularly true of the HEV class which would pay $1971 per hour of public charging
time saved. Increased charging time at home/work is not perceived as a problem across classes and the BEV class actually appears to
value it positively. Public charging station availability appears to be receiving little consideration from the ICE and HEV classes who
are focused on public charging time. The BEV class are more attuned and will pay $1122 for a “step” improvement in the availability
of public charging stations.
Two vehicle attributes from the stated preference scenarios provided insignificant results and were excluded from the final model.
Fueling/charging cost was expressed as an estimated annual dollar expenditure to unite the costs of gasoline and electricity in one
attribute. Respondents appeared to react more to the kilometre range variables which address the important fuel efficiency aspect as
distinct from fuel cost. Some further exploratory analysis revealed more attention being paid to this cost when an economy vehicle is
the preferred body type. The second insignificant vehicle attribute was the estimated annual emitted tonnage of tailpipe CO2. This
quantity was always set to zero for BEVs in the scenarios. The use instead of well-to-wheel emission estimates would have resulted in
non-zero quantities but would have reduced clarity for respondents about functional differences between PHEVs and BEVs. Note that
an importance attitude about vehicular emissions was significant in the class probability model.
3.3. Description of latent classes
The nature of the latent classes is partly defined by the class membership sub-model, and choice-maker attributes that guide the
allocation, and secondly by the class-specific utility models. The important contributing variables have been described in Sections 3.1
and 3.2. At this point, each of the four derived latent classes is described in a comprehensive and integrated fashion based on the
results outlined from the prior two sub-sections.
The first latent class is ICE-oriented. It is clearly older, less educated, less urban and is pre-occupied with financial considerations.
There is a strong reluctance to pay more for a car, whether an EV (PHEV and BEV) or not, and there is scepticism with the idea that an
EV will pay for itself over time. Cash incentives for BEV and PHEV are discounted by this class. Less concern is shown for good
mileage and lower emissions than other classes but this class is nevertheless more likely to lean toward an economy vehicle for the
next vehicle purchase. With respect to charging behaviour, this class shows little enthusiasm for the idea of plugging in at present but
showed responsiveness to potential improvements in public charging time and is somewhat swayed by a strong battery warranty. It is
noteworthy that the ICE class is by far less likely to be currently owning an HEV, which potentially means that they are even further
from owning a BEV or a PHEV.
The HEV-oriented class is like the ICE class in that it is older but it combines arguably the highest levels of environmental concern
with a real aversion to potential inconveniences of the required charging for BEV and PHEV. It is the least purchase price sensitive of
all the classes but heavily discounts cash incentives for BEV and PHEV. In demographic terms, the HEV class is least likely to be
French-speaking, and thus least oriented to Quebec, and is more likely to be associated with female respondents. HEV class members
prefer not to lose time with public charging. Accordingly, with respect to charging, this class is highly focused on improvements in
public charging time and in the battery warranty. Of all the classes, this one feels strongest about the importance of reducing their
vehicular emissions but, in contrast to other classes, they choose to do so by valuing gasoline range very highly. Not surprisingly,
hybrid vehicles are much more likely to be currently owned by this class. For both the ICE and HEV classes, results suggest that
members have not yet gotten to the point where they worry about charging station availability. Relative to availability, they are
paying more attention to electric range, public charging time and battery concerns.
The PHEV and BEV oriented classes share several similarities. Both classes are younger, better educated and more urban though
the PHEV class is the more educated of the two and is also more suburban in character. Both classes show less urgency to pursue their
next vehicle purchase but are equally receptive to cash incentives. Both classes are paying the most attention of all to evaluating
electric range and less attention to gasoline range and public charging time. Both are focused on the availability of charging stations.
In terms of differences, the BEV-oriented class is more focused on positive aspects such as rapid acceleration, relative mechanical
simplicity and the potential for a significant electric travel range. The PHEV-oriented class has much more concern with batteries and
associated warranties than the BEV class and is more sensitive to purchase price. The PHEV class does not share the same indifference
for gasoline range that the BEV class does. The BEV class is more likely to be pursuing an economy vehicle for the next vehicle
purchase and is much more associated with an incremental, as opposed to replacement vehicle, for the next purchase than the PHEV
class. Relative to other classes, the BEV class is most pre-occupied with aspects such as minimal maintenance costs and rapid vehicle
acceleration. Finally, the BEV class is far more likely to speak French at home indicating greater acceptance of BEVs in Quebec.
4. Metropolitan variations in openness to electric vehicles
In Fig. 3, posterior class membership probabilities (Greene and Hensher, 2003), derived from the latent class model on a per
219
Quebec Agglomerations, QUE (228) 0.164 0.348 0.119 0.370

Montreal, QUE (4061) 0.174 0.304 0.165 0.357
Victoria, BC (365) 0.138 0.338 0.170 0.353
Vancouver, BC (2504) 0.146 0.324 0.168 0.362
Saguenay, QUE (160) 0.110 0.357 0.206 0.327
Oshawa, ON (389) 0.129 0.314 0.175 0.382
Sherbrooke, QUE (215) 0.189 0.248 0.150 0.412
Toronto, ON (6130) 0.132 0.301 0.191 0.376
Quebec City, QUE (806) 0.106 0.309 0.164 0.421
Ottawa - Gatineau ON/QUE (1332) 0.118 0.296 0.167 0.419
Kelowna, BC (197) 0.092 0.321 0.172 0.415
Fredericton, NB (56) 0.070 0.343 0.253 0.334
Kitchener-Cambridge-Waterloo, ON (511) 0.086 0.320 0.193 0.400
Barrie, ON (203) 0.137 0.266 0.279 0.317
Hamilton, ON (772) 0.112 0.288 0.237 0.363
Abbotsford-Mission, BC (184) 0.095 0.300 0.181 0.425
St. Catharines - Niagara, ON (408) 0.116 0.279 0.244 0.361
London, ON (506) 0.099 0.295 0.183 0.423
Edmonton, ALTA (1363) 0.078 0.313 0.206 0.403
Thunder Bay, ON (125) 0.053 0.338 0.199 0.410
Trois-Rivieres, QUE (156) 0.168 0.221 0.136 0.475
Greater Sudbury, ON (165) 0.095 0.284 0.134 0.487
Halifax, NS (418) 0.097 0.274 0.234 0.396
Saint John, NB (127) 0.095 0.272 0.181 0.452
Kingston, ON (170) 0.044 0.320 0.217 0.419
Moncton, NB (148) 0.080 0.281 0.203 0.435
Calgary, ALTA (1440) 0.086 0.273 0.213 0.428
Winnipeg, MAN (793) 0.081 0.277 0.201 0.441
Regina, SASK (241) 0.086 0.259 0.211 0.444
Saskatoon, SASK (305) 0.065 0.266 0.233 0.436
Guelph, ON (153) 0.059 0.270 0.158 0.512
Non-urban 0.077 0.250 0.203 0.470
Peterborough, ON (123) 0.088 0.230 0.158 0.524
St. John's, NFLD (214) 0.075 0.239 0.194 0.493
Brantford, ON (144) 0.067 0.211 0.238 0.483
Windsor, ON (336) 0.062 0.189 0.157 0.592
National Sample 0.097 0.295 0.192 0.416
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
BEV-Oriented PHEV-Oriented HEV-Oriented ICE-Oriented

Fig. 3. Prominence of each latent class by Census Metropolitan area.
household basis, have been aggregated and put on the unit interval to yield an allocation of sampled households by metropolitan
areas across the four classes. A similar exercise has been done using prior class membership probabilities but the results are generally
not materially different at the level of metropolitan areas. The results illustrate vividly that there are very strong variations in class
membership by regions and urban areas.
For each line item, the descriptive name of the metropolitan area, the province in which the metro resides and the population of
the metro in thousands of people are shown. The list is sorted in descending order of the sum of the EV oriented class proportions
(BEV oriented + PHEV oriented). With a handful of exceptions, the sampled households per metropolitan area number over 100 with
samples as large as approximately 2000 in the cases of Montreal and Toronto. Fig. 3 contains a line item called “Quebec Agglom-
erations” which is an aggregation of three smaller cities located east of Montreal. Another item, associated with 24% of the sample, is
named “non-urban” and contains all small census agglomerations, towns and rural areas that have not been otherwise captured in
Fig. 3.
The aggregated posterior class membership probabilities for these metros are influenced primarily by prevailing household at-
titudes (Table 5), through the share of households that fit powertrain profiles in terms of age, education and preferences for small
vehicles and through the position of metros on the urban hierarchy which affects flow of information about EVs. Though it has not
been modelled directly, the actual presence of incentives in some provinces and their absence in others has likely had an indirect
220
influence on the attitudes that have been sampled from Canadians and which partly shape the results in Fig. 3. However, the direct
impact of existing incentives is beyond the scope of this study.
The list is dominated by metros that are located in the three most populated provinces of Quebec, British Columbia and Ontario.
These provinces also have the cleanest electricity generation profiles and are the only provinces that have pursued monetary in-
centives to boost the adoption of electric vehicles. All of Canada’s three large metros (Toronto, Montreal and Vancouver) are near the
top of the list.
Urban areas from Quebec are most prominent near the top of the list and many of the high BEV scores are associated with Quebec.
This is true as well of smaller metros that might internalize less travel and thus appear less compatible with limited BEV range. In
Quebec cases, the HEV-orientation gets “squeezed out” by a stronger electric vehicle orientation. BC metros are also strongly as-
sociated with the top of the list. Overall, results suggest that Quebec is the province most open to electric mobility and certainly to
BEVs.
The results for Ontario are more diverse with several of its smaller metros appearing near the bottom of the list in terms of being
oriented to plug-in technologies. The Windsor, Ontario metro, which is directly adjacent to Detroit and is tightly integrated with
Detroit in automotive manufacturing supply chains, is highly ICE-oriented. Though further investigation would be required, this
result could be hypothesized to suggest discomfort among Windsor residents about status quo disruptions caused by the rise of
electric mobility.
The other strong theme that emerges is that metros in the interior western provinces of Alberta, Saskatchewan and Manitoba are
all in the lower half of the list. In these cases, scores for BEV are all well under 0.1 and scores for PHEV are generally nearer to 0.25
than 0.3. Alberta is at the heart of fossil fuel production in Canada and natural resources are important to Saskatchewan and
Manitoba as well. Results for the three main east coast metros are similar. Fig. 3 shows that results across the metros of the less
populated provinces (i.e. the eastern and prairie provinces) are somewhat more EV oriented than the result for non-urban areas in
Canada.
With reference mostly to future work, it is worth noting that Saskatchewan and Manitoba are two of the coldest significantly
populated regions of Canada and the negative impact of cold weather on EV performance and electric range (Jaguemont et al. (2016)
could be a factor among others working against higher interest in EVs. Location-specific average annual or winter temperatures, for
example, were not included as variables in this analysis. Some would contend that the ongoing absence of government cash incentives
in these provinces or other factors may be more important.
5. Discussion
A wide combination of choice-maker attributes, including specific attitudinal statements, vehicle ownership characteristics, de-
mographic and locational factors have proven important in defining the latent classes of this analysis and a much larger array were
tested. The latent class choice approach adapted from Beck et al. (2013) for this study, as described in the introduction, has helped to
generate additional insight on how personal attitudes towards specific aspects of EVs shape choice of powertrain. Such personal
attitudes also play a significant role to influence overall latent class memberships in this study. In the national sample the probability
of membership in the plug-in oriented classes (BEV + PHEV) is 0.392 which is far in excess of current market shares. This result
suggests considerable openness to electric vehicles in Canada and the possibility of higher market shares over time.
The types of attitudes and beliefs proving significant in this study deal with practical realities of utilizing each powertrain in daily
life. Certain important factors have been less directly linked in past discrete choice implementations. Aspects include: attitudes
toward the concept of required charging behaviours, and especially in public settings; paying higher upfront costs for longer term
benefits; comfort level on having to rely so much on an expensive battery; and whether plugging in a vehicle at home is seen as
practical. Thoughts about having to pay a purchase price premium are relevant also. Generally, those who are PHEV or BEV-oriented
have a higher comfort level with these aspects.
Consider that each of these factors is distinct from an emphasis on environmental concern or general attitudes about technology
which are themselves important to some extent. But the results here suggest that these aspects operate more in the background as was
confirmed by Mohamed et al. (2016) using a structural equation modelling approach. For example, results for the HEV-oriented class
suggest that a concern with the behaviours required of operating a BEV or PHEV seem to overwhelm their stated considerable desire
to minimize vehicular emissions. Results from the latent class model also showed that the HEV-oriented are willing to burn gasoline
to fuel travel while the BEV-oriented are much less so.
With regard to demographics, household income was not observed in the current analysis to be a significant variable in shaping
the choice of powertrains and this result echoes the findings of Axsen et al. (2015) in Canada. Partly this is because younger classes,
who expressed more optimism for EVs, have not reached their peak earning years and older classes, who have reached higher income
levels, in many cases are more resistant to the changes required of operating an EV. Though not in the context of a latent class choice
model, Hackbarth and Madlener (2013) found strong age and education patterns in the German context, which are confirmed in the
current analysis for Canada. Results showed that the sex of the respondent had a bearing on powertrain choice with women being
more oriented to the HEV class. Pragmatism combined with risk aversion seems to be at play where an ICE is seen as less en-
vironmentally responsible and an HEV is seen as the best way to achieve better environmental outcomes while not taking risks with
new technologies. A similar outcome has been noted in prior studies (Dagsvik et al., 2002; Morton et al., 2015).
Geography is an important theme in this analysis and one that has not been emphasized in prior research in this area. Regionally,
it has been shown that powertrain orientations in French-speaking Quebec, differ from the rest of Canada. At regional levels and
below, there is significant variation in how BEV and PHEV preferences spatially manifest themselves. Partly these differences are
221
explained by how households of different types organize themselves over space but the analysis also revealed a distinct urban-rural
effect based on the application of a highly localized urbanization index across all of Canada. Variables used to capture related effects
instead through a respondent’s estimates of annual or daily mileage driven, for example, proved not significant. The PHEV and
especially the BEV orientation were found to be urban. In many urban settings, an appropriate matching of household profiles, daily
travel circumstances, and more mobility options combine to contain perceptions about range anxiety among the BEV class. Rural
areas differ on these aspects and in addition residents prefer larger vehicles such as pickups and SUVs. The electric vehicle market for
such body types is immature from both the supply and demand perspectives at this point. More work remains to be done regionally
and also with intra-metropolitan and urban variations in mindsets toward EVs.
Also on vehicle body type, having the intent to purchase an economy vehicle as the next household vehicle significantly affects
choices among powertrains while this same variable for several other body types covered by the survey is not significant. This is a
relevant result to consider in tandem with outcomes seen in Higgins et al. (2017) which developed separate models based on the body
type of the next household vehicle. The current result suggests that some of the key demographics, attitudes and geographical aspects
that are captured in this paper are more directly linked to the choice of powertrain than the vehicle body types that households
prefer.
The topic of range anxiety has been linked to the composition of vehicle household fleets in past work. Empirically, early-adopter
results for Norwegian EV owners (Klöckner et al., 2013) and also for EV owners in this sample show that PHEVs and BEVs appear
more often in households with more vehicles. In the current sample, 69% of existing EV households are multi-vehicle households
compared to 53% which are multi-vehicle in non-EV households. However, in the current model, the number of household vehicles
was found not to influence powertrain choice. In addition, a dummy variable on “multicar” households did not perform well in
contrast to results of Tamor and Milacic (2015). A more relevant variable for the current study is whether the household expects an
incremental or replacement vehicle for their next acquisition. This appears to strongly influence membership in the BEV class. Based
on results of the current model, other means that the BEV-class uses to temper range anxiety include seeking out an economy-sized
BEV so that long electric range can be achieved at a lesser purchase price, and simply by exhibiting greater tolerance for the
possibility of charging inconvenience.
On the topic of vehicle attributes, purchase price sensitivity stands out as an important element of this study, with the ICE-
oriented class being by far the most price sensitive. This result is also evident in the results of Hidrue et al. (2011) where the
conventional vehicle class is much more purchase price sensitive than the EV class. Purchase price sensitivity alone does not suggest
any intrinsic aversion to the idea of acquiring an electric vehicle though the ICE-oriented class does exhibit an aversion based on
other aspects such as negative attitudes towards public charging and less concern with the environment. For the BEV class, annual
maintenance costs is given close attention. This class may be attracted to the relative simplicity of BEVs and the fact that there are
many fewer moving parts to maintain. This outcome resonates with the result of Morton et al. (2015) suggesting that BEV-oriented
consumers are more mechanically aware.
Comparing WTP results reveals more apparent sensitivity to changes in electric range than changes in public charging time among
the EV classes. For the BEV-oriented, a one hour reduction in public charging time is seen as equivalent to adding only 25 km of extra
range. For the PHEV-oriented, who may see themselves as less dependent on public charging, it is equivalent to only 13 km of extra
range. Conversely, the HEV-oriented appear quite concerned about avoiding time spent with public charging and equate an hour of
charging time to 117 km of extra range.
6. Conclusions
This research has employed a comprehensive national survey data set collected from the Canadian population of households as
the basis for a set of vehicle choice experiments. Only households with little or no intent to purchase a vehicle in the future have been
excluded. On this basis, a national latent class choice model has been developed which has four distinct latent classes.
This work has shown that households can be well-classified based on a combination of their socio-demographics and their
attitudes towards EVs with the latter being more surgical in ability to differentiate. Certain demographics such as age and education
are indispensable though in understanding the latent classes. There does appear to be a clear separation between households that are
younger, more urban and dynamic and more open to change and those that are older, more established, less urban, less educated and
perhaps more set in their ways.
So how open are Canadian households to electric mobility? Results suggest that much of the population has an open mind. An ICE-
oriented mindset is not the majority in the survey data. In terms of how Canadian households are generally allocated to powertrain
mindsets, a 40–30–20–10 rule-of-thumb approximately captures the overall allocation as of 2015. These shares represent the ICE/
PHEV/HEV/BEV orientations respectively. The PHEV orientation is much more commonplace than a BEV orientation at present
though this outcome might be expected to evolve with improving batteries and electric range and with improving knowledge on
aspects such as mechanical simplicity of BEVs. In terms of current adoption patterns, Canadians are acquiring BEVs at a rate similar to
PHEVs. As to why existing market shares do not mirror the rule-of-thumb, important explanations include higher purchase prices of
BEVs and PHEVs, issues with vehicle supply and dealership bias in favour of ICE vehicles, and a remaining lack of general public
awareness about electric mobility.
The latter aspect is very important because measures to improve public awareness will work hand-in-hand with cash and non-cash
incentives to accelerate the adoption of EVs in Canada. ICE and HEV-oriented classes are shown in this research to heavily discount
government cash incentives for EVs. Partly this may stem from a general lack of awareness or misunderstanding about specific
aspects. The discounting by the HEV class, who appear quite apprehensive about public charging, could be reduced through
222
education that highlights the effectiveness of home charging. The ICE-class could be better trained on the long-term cost effectiveness
of EVs. The appreciation of the BEV-class for minimized maintenance costs and fewer moving parts is something that can be well-
emphasized in concept to the other classes. The current results also offer insights about how non-cash incentives are viewed. Access to
high occupancy vehicle lanes, for example, is something that can be stressed for suburban metropolitan areas with the highest levels
of automobile dependence and traffic congestion. These are all evidence-based examples supported by the current analysis and offer
strategic assistance in maximizing the impact of policy packages.
This research also offers substantial insight about geographical variation in EV openness that can highlight the most promising
metropolitan markets for growth in adoption and others that are further behind and perhaps need special attention.
Acknowledgements
This article is dedicated in memory of Professor Pavlos Kanaroglou (1948–2016) who has been most aptly described as a gen-
tleman and a scholar and who had a highly impactful academic career. Pavlos was influential in the development of each of his co-
authors on the present article and the work that has been done here would not have been possible without his contributions. The
authors gratefully acknowledge the support from the Social Sciences and Humanities Research Council of Canada (SSHRC) Grant No:
886-2013-000 and from our public and private sector partners whose insights we value highly.
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.trd.2017.12.
006.
References
Ajzen, I., 1991. The theory of planned behavior. Organ. Behav. Hum. Decis. Process. 50 (2), 179–211.
Al-Alawi, B.M., Bradley, T.H., 2013. Review of hybrid, plug-in hybrid, and electric vehicle market modeling studies. Renew. Sust. Energ. Rev. 21, 190–203.
Anable, J., Schuitema, G., Skippon, S., Kinnear, N., 2011. Who will adopt electric vehicles? A segmentation approach of UK consumers. In: Proceedings from ECEEE
2011 Summer Study, Belambra Presqu’Ile de Giens, France.
Axsen, J., Bailey, J., Castro, M.A., 2015. Preference and lifestyle heterogeneity among potential plug-in electric vehicle buyers. Energy Econ. 50, 190–201.
Beck, M.J., Rose, J.M., Hensher, D.A., 2013. Environmental attitudes and emissions charging: an example of policy implications for vehicle choice. Transp. Res. Part A:
Pol. Pract. 50, 171–182.
Bolduc, D., Boucher, N., Alvarez-Daziano, R., 2008. Hybrid choice modeling of new technologies for car choice in Canada. Transp. Res. Rec.: J. Transp. Res. Board
2082, 63–71.
Brownstone, D., Bunch, D.S., Train, K., 2000. Joint mixed logit models of stated and revealed preferences for alternative-fuel vehicles. Transp. Res. Part B: Methodol.
34 (5), 315–338. http://dx.doi.org/10.1016/s0191-2615(99)00031-4.
Bunch, D.S., Bradley, M., Golob, T.F., Kitamura, R., Occhiuzzo, G.P., 1993. Demand for clean-fuel vehicles in California – a discrete-choice stated preference pilot
project. Transp. Res. Part A: Pol. Pract. 27 (3), 237–253.
Buskens, V., 2015. Rational choice theory. Sociology 901–906.
Caperello, N.D., Kurani, K.S., 2012. Households' stories of their encounters with a plug-in hybrid electric vehicle. Environ. Behav. 44 (4), 493–508.
Dagsvik, J.K., Wennemo, T., Wetterwald, D.G., Aaberge, R., 2002. Potential demand for alternative fuel vehicles. Transp. Res. Part B: Methodol. 36 (4), 361–384.
Dumortier, J., Siddiki, S., Carley, S., Cisney, J., Krause, R.M., Lane, B.W., Rupp, J.A., Graham, J.D., 2015. Effects of providing total cost of ownership information on
consumers’ intent to purchase a hybrid or plug-in electric vehicle. Transp. Res. Part A: Pol. Pract. 72, 71–86.
Egbue, O., Long, S., 2012. Barriers to widespread adoption of electric vehicles: an analysis of consumer attitudes and perceptions. Energ Policy 48, 717–729.
Ewing, G.O., Sarigöllü, E., 1998. Car fuel-type choice under travel demand management and economic incentives. Transp. Res. Part D: Transp. Environ. 3 (6),
429–444.
Graham-Rowe, E., Gardner, B., Abraham, C., Skippon, S., Dittmar, H., Hutchins, R., Stannard, J., 2012. Mainstream consumers driving plug-in battery-electric and
plug-in hybrid electric cars: a qualitative analysis of responses and evaluations. Transp. Res. Part A: Pol. Pract. 46 (1), 140–153.
Greene, W.H., Hensher, D.A., 2003. A latent class model for discrete choice analysis: contrasts with mixed logit. Transp. Res. Part B: Methodol. 37 (8), 681–698.
Hackbarth, A., Madlener, R., 2013. Consumer preferences for alternative fuel vehicles: a discrete choice analysis. Transp. Res. Part D: Transp. Environ. 25, 5–17.
Helveston, J.P., Liu, Y., Feit, E.M., Fuchs, E., Klampfl, E., Michalek, J.J., 2015. Will subsidies drive electric vehicle adoption? Measuring consumer preferences in the
U.S. and China. Transp. Res. Part A: Pol. Pract. 73, 96–112.
Hensher, D.A., Beck, M.J., Rose, J.M., 2010. Accounting for preference and scale heterogeneity in establishing whether it matters who is interviewed to reveal
household automobile purchase preferences. Environ. Resour. Econ. 49 (1), 1–22.
Hidrue, M.K., Parsons, G.R., Kempton, W., Gardner, M.P., 2011. Willingness to pay for electric vehicles and their attributes. Resour. Energy Econ. 33 (3), 686–705.
Higgins, C.D., Mohamed, M., Ferguson, M.R., 2017. Size matters: how vehicle body type affects consumer preferences for electric vehicles. Transp. Res. Part A: Pol.
Pract.
Hoen, A., Koetse, M.J., 2014. A choice experiment on alternative fuel vehicle preferences of private car owners in the Netherlands. Transp. Res. Part A: Pol. Pract. 61,
199–215.
Holland, S., Mansur, E., Muller, N., Yates, A., 2015. Environmental Benefits from Driving Electric Vehicles? National Bureau of Economic Research, Cambridge MA.
Ito, N., Takeuchi, K., Managi, S., 2013. Willingness-to-pay for infrastructure investments for alternative fuel vehicles. Transp. Res. Part D: Transp. Environ. 18, 1–8.
Jaguemont, J., Boulon, L., Dubé, Y., 2016. A comprehensive review of lithium-ion batteries used in hybrid and electric vehicles at cold temperatures. Appl. Energy 164,
99–114.
Kennedy, C., 2015. Key threshold for electricity emissions. Nat. Clim. Change 5 (3), 179–181.
Klöckner, C.A., Nayum, A., Mehmetoglu, M., 2013. Positive and negative spillover effects from electric car purchase to car use. Transp. Res. Part D: Transp. Environ.
21, 32–38.
Kurani, K.S., Turrentine, T., Sperling, D., 1994. Demand for electric vehicles in hybrid households: an exploratory analysis. Transp. Policy 1 (4), 244–256.
Lebeau, K., Van Mierlo, J., Lebeau, P., Mairesse, O., Macharis, C., 2012. The market potential for plug-in hybrid and battery electric vehicles in Flanders: a choice-
based conjoint analysis. Transp. Res. Part D: Transp. Environ. 17 (8), 592–597.
Mabit, S.L., Fosgerau, M., 2011. Demand for alternative-fuel vehicles when registration taxes are high. Transp. Res. Part D: Transp. Environ. 16 (3), 225–231.
Maness, M., Cirillo, C., 2012. Measuring future vehicle preferences. Transp. Res. Rec.: J. Transp. Res. Board 2285, 100–109.
Mohamed, M., Higgins, C., Ferguson, M., Kanaroglou, P., 2016. Identifying and characterizing potential electric vehicle adopters in Canada: a two-stage modelling
approach. Transp. Policy 52, 100–112.
223
Moons, I., De Pelsmacker, P., 2012. Emotions as determinants of electric car usage intention. J. Market. Manage. 28 (3–4), 195–237.
Morton, C., Anable, J., Nelson, J.D., 2015. Assessing the importance of car meanings and attitudes in consumer evaluations of electric vehicles. Energ. Effi. 9 (2),
495–509.
Morton, C., Anable, J., Nelson, J.D., 2016. Exploring consumer preferences towards electric vehicles: the influence of consumer innovativeness. Res. Transp. Bus.
Manage. 18, 18–28.
Plötz, P., Schneider, U., Globisch, J., Dütschke, E., 2014. Who will buy electric vehicles? Identifying early adopters in Germany. Transp. Res. Part A: Pol. Pract. 67,
96–109.
Potoglou, D., Kanaroglou, P.S., 2007. Household demand and willingness to pay for clean vehicles. Transp. Res. Part D: Transp. Environ. 12 (4), 264–274.
Qian, L.X., Soopramanien, D., 2011. Heterogeneous consumer preferences for alternative fuel cars in China. Transp. Res. Part D: Transp. Environ. 16 (8), 607–613.
Rezvani, Z., Jansson, J., Bodin, J., 2015. Advances in consumer electric vehicle adoption research: A review and research agenda. Transp. Res. Part D: Transp. Environ.
34, 122–136.
Rose, J.M., Bliemer, M.C.J., 2009. Constructing efficient stated choice experimental designs. Transp. Rev. 29 (5), 587–617.
Schuitema, G., Anable, J., Skippon, S., Kinnear, N., 2013. The role of instrumental, hedonic and symbolic attributes in the intention to adopt electric vehicles. Transp.
Res. Part A: Pol. Pract. 48, 39–49.
Shen, J., 2009. Latent class model or mixed logit model? A comparison by transport mode choice data. Appl. Econ. 41 (22), 2915–2924.
Tamor, M.A., Milacic, M., 2015. Electric vehicles in multi-vehicle households. Transp. Res. Part C: Emerg. Technol. 56, 52–60.
Tanaka, M., Ida, T., Murakami, K., Friedman, L., 2014. Consumers' willingness to pay for alternative fuel vehicles: a comparative discrete choice analysis between the
US and Japan. Transp. Res. Part A: Pol. Pract. 70, 194–209.
Train, K.E., 2008. EM Algorithms for nonparametric estimation of mixing distributions. J. Choice Model. 1 (1), 40–69.
Wang, S., Fan, J., Zhao, D., Yang, S., Fu, Y., 2014. Predicting consumers’ intention to adopt hybrid electric vehicles: using an extended version of the theory of planned
behavior model. Transportation 43 (1), 123–143.
Ziegler, A., 2012. Individual characteristics and stated preferences for alternative energy sources and propulsion technologies in vehicles: A discrete choice analysis for
Germany. Transp. Res. Part A: Pol. Pract. 46 (8), 1372–1385.
224

1 s2.0 S1361920916306058 Main PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1 s2.0 S1361920916306058 Main PDF

Uploaded by

Copyright:

Available Formats

Transportation Research Part D 58 (2018) 208–224

Contents lists available at ScienceDirect

Transportation Research Part D

How open are Canadian households to electric vehicles? A national

Available online 19 December 2017

2. Methods and data

2.1. Survey data

Fig. 1. Distribution of respondents by province.

2.2. Stated preference exercise

Variable Value Frequency Percentage Canadian census

Age of Household Head (Years) 18–24 815 4% 4%

Language Spoken Most Often at Home English 15,786 77% 65%

Marital status of Household Head Single 5068 25% 18%s

Household Size 1 3720 18% 28%

Dwelling Type Single Detached House 13,438 65% 56%

Dwelling Tenure Owner 16,103 78% 69%

Education Level of Household Head No certiﬁcate, diploma or degree 624 3% 17%

Household gross income Less than $25,000 943 5% 18%

Fig. 2. A typical stated preference scenario.

Study Geography Target group Respondents Vehicle Types Explicit Model-type

Attributes ICE HEV PHEV BEV

2.3. Latent class choice model

The latent class choice model can be expressed as:

Classes k LL N AIC BIC

2 50 −70,349 17,953 140,798 141,188

Reference: HEV-oriented class ICE-oriented PHEV-oriented BEV-oriented

Constant 5.5366*** −0.9714*** −5.6769***

Socioeconomic and Demographic

* Statistical signiﬁcance at the 10% level.

3.1. Class probability sub-model

MNL ICE-oriented HEV-oriented PHEV-oriented BEV-oriented

Log-Likelihood at Convergence −88,371 −61,841

* Statistical signiﬁcance at the 10% level.

3.2. Utility sub-models by latent class

3.3. Description of latent classes

4. Metropolitan variations in openness to electric vehicles

Quebec Agglomerations, QUE (228) 0.164 0.348 0.119 0.370

National Sample 0.097 0.295 0.192 0.416

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

BEV-Oriented PHEV-Oriented HEV-Oriented ICE-Oriented

Appendix A. Supplementary material

You might also like

Constant 5.5366* −0.9714* −5.6769***