Content uploaded by Sunbin Yoo
Author content
All content in this area was uploaded by Sunbin Yoo on Sep 19, 2022
Content may be subject to copyright.
1
Trust more, Fear less:
The Role of Social Support in Fully Automated Vehicle Choice
Sunbin Yoo, Junya Kumagai and Shunsuke Managi
We reveal how mental support from family,
friends, and local authorities—social support (SP)—
contributes to inclusivity in the context of fully
autonomous vehicles (FAVs). We focus on disaster
victims who have trouble driving and test how the
information on traumatic experiences relates to their
purchase decisions. To this end, we create a unique
survey that asks about respondents’ purchasing
decisions before/after we share information with
them on FAV traffic accidents that led to a driver’s
death. Our results show that SP is crucial for
encouraging people to choose FAVs, even after
fatal news concerning FAVs is shared, regardless of
disaster experience. We also find that other factors,
such as safety and convenience, can encourage
people to ‘substitute’ choosing FAVs in place of
not choosing FAVs. By adopting integrated choice
and latent variable (ICLV) models considering
individual heterogeneity over 60,000 respondents,
our results indicate possible policy paths through
which to utilize SP in creating FAVs as an inclusive
transportation mode.
JEL Classification: L62, Q48, Q55, Q58.
Keywords: Autonomous vehicle; Self-driving car;
Emerging transport modes; Mobility on-demand; De-
mand estimation; Integrated choice and latent vari-
able approach
Urban Institute & School of Engineering, Kyushu University,
yoo.sunbin.277@m.kyushu-u.ac.jp
I. I
We explore whether mental support encour-
ages people with traumatic experiences to choose
new technology, taking fully autonomous vehicles
(FAVs) as an example. Previous works have found
that FAVs can revolutionize the transportation
paradigm by reducing vehicle emissions ((1), (2)),
increasing road and travel efficiency ((3), (4), (5),
(6)), and substantially reducing the number of
traffic accidents due to human error ((7), (8), (9),
and (10)).Inspired by previous works ((11) and
(12)), we start from the viewpoint that these ben-
efits will be maximized if FAVs become inclusive
and embrace socially excluded vulnerable people.
We contribute to the literature on the impact of
social support (SP) on new technology adoption.
The previous works have already shown that SP
from family, friends, local authorities, and volun-
teers is critical in encouraging people to adapt to
new technology((13);(14)). In this study, we extend
the findings of these works to those who urgently
require SP—those who cannot acquire mobil-
ity due to physical/cognitive/mental challenges1.
Among the various types of shocks, we choose to
1In this study, while there are various types of ‘support’, such
as financial support, we focus on mental support that incor-
porates encouragement, trustful relationships and reliability.
2
consider individuals who are exposed to natural
disasters (NDs)—disaster victims—because disas-
ter experiences may induce the loss/changes of
mental, cognitive, and physical capability, which
may motivate such individuals to choose FAVs.
In this regard, taking Japan as our study area has
several advantages. Japan is particularly vulnera-
ble to NDs because of its climate and topography,
as evidenced by its experiences with countless
earthquakes, typhoons, and other types of disas-
ter. Our data provide a sample that includes over
10,000 disaster victim observations. This setting
allows our results to be more realistic because for
an individual to lose the ability to drive due to a
disaster, it may require multiple ND experiences,
which are likely to occur in Japan. Thus, our set-
ting eliminates the necessity of soliciting the time
and name of a specific disaster that caused phys-
ical and mental damage to respondents. While
we focus on disaster victims and FAVs, as our
study concerns how people with risk experiences
respond to new technologies, the implications
of our results can be extended beyond disaster
victims and FAVs.
We design and conduct a set of original surveys
that contain information on a fatal FAV accident,
which is likely to shift people’s preferences away
from such vehicles. We include a famous exam-
ple in our survey: the Tesla autopilot crushing
incident that resulted in a driver’s death, which
occurred in Delray Beach, Florida, in March 2016.
The incident raised concerns about Tesla’s autopi-
lot technology shortly after the National Trans-
portation Safety Board stated that the company’s
autopilot driver assistance system had been en-
gaged during that fatal crash. We share the acci-
dent information with the survey respondents and
ask them again about their preferences.
Taking advantage of the data of over 60,000
respondents from 2015 to 2017 in Japan, we ask
the respondents twice about their willingness to
buy (WTB; purchasing decision) FAVs: before/after
we share the accident information with them.
Simply, the survey is set up as follows:
Step 1. The respondents are asked to write down
whether they wish to purchase FAVs.
Step 2. We share information on the Tesla autopilot
accident, which resulted in the death of the
driver.
Step 3. The respondents are again asked to write
down whether they wish to purchase FAVs.
Steps 1 and 3 are written separately.
We include the following items that reflect the
respondents’ psychological intentions to adopt
FAVs—risk perceptions (RPs), traffic accidents,
convenience due to FAVs, SP, and environmen-
tal concerns. We identify these items, whenever
possible, based on statements that have previ-
ously been used and found to be effective in the
literature. Methodologically, we are interested in
capturing the utility associated with FAV choices
regarding the above factors. The goal is to esti-
mate a utility function (random utility function)
to model FAV choices. We additionally investigate
whether SP plays a pivotal role in substituting for
those individuals who are reluctant to choose FAVs
before the accident news is shared.
We consider two aspects: 1) the utility function
of socially excluded people may differ from those
who are not socially excluded, and 2) individ-
3
ual heterogeneity within/without a group may
exist. To this end, we choose integrated choice
and latent variable (ICLV) models, which have
recently become an increasingly popular exten-
sion of discrete choice models that attempt to
explicitly model the cognitive process underlying
the formation of any choice ((15)). We model our
utility functions to allow for one or more of the pa-
rameters in the model to be randomly distributed,
thus allowing for the coefficients in the model to
vary across decision makers. Figure 1shows our
study structure. We first estimate demand using
the answers from the 1st purchase decision, share
the accident news and then estimate the factors
closely correlated with the final decision.
The remainder of this paper is structured as fol-
lows. Section II provides background information
on the industry and policy. The data and model
are presented in Sections III and IV, respectively.
Section ?? reports the empirical results. Section
VI-B addresses our findings and provides policy
implications. Section VII concludes the paper.
II. B
This section first defines FAVs and then reviews
previous works on FAVs focusing on their potential
benefits, concerns, and demand patterns. Then,
we discuss the related literature, identify the gap
in the literature and present our contributions.
Then, we discuss why our study addresses FAVs in
achieving inclusive transport by focusing on social
exclusion, traffic accidents and ND experiences.
A. FAVs
FAVs have different levels of automatic opera-
tion, as presented in Figure 2. A system generally
unconditionally performs all driving operations at
level 5 (fully automated); however, as of 2021, a
fully automated system has not yet been realized
in Japan, and the Japanese government aims to
achieve this goal by 2025. Currently, consumers
cannot select the specific technology or devices
to integrate into their new cars when purchas-
ing level 1 and 2 AVs.2For high and full lev-
els of automation (levels 4 and 5, respectively),
cars are expected to pass driving qualification
tests that inspect whether obligatory automation-
related equipment and accessories are fully oper-
ational. As illustrated in Figure 2, the Society of
Automotive Engineers (SAE) has defined different
levels of automated functionality, ranging from no
automated features (level 0) to full automation
(level 5, commonly referred to as autonomous,
self-driving or driverless vehicles).
B. FAVs and Social Exclusion
FAVs can create mobility opportunities by pro-
viding access to socially excluded people who are
unable to drive, such as elderly individuals (with
noticeable problems in terms of driving capaci-
ties) and physically/mentally disabled people ((16)
and (17)). However, despite the soaring interest in
FAVs among researchers, whether FAVs can em-
brace socially excluded people remains relatively
unclear.
Recognizing the potential of FAVs, on March
29, 2019, the Japanese cabinet announced that
FAVs should pursue a transition to become more
flexible to change and respond as FAV technology
evolves, thereby allowing for the full realization
2We refer to 2021.
4
Fig. 1: Study Structure
of FAV benefits. This transition aimed not only
to improve efficiency based on existing socially
accepted objectives (e.g., improvement of conve-
nience and prevention of accidents) but also the
realization of new value created by the diversifi-
cation of the objectives themselves (by including
socially vulnerable people in the policy target)
and the resolution of problems such as transport
disparity, social exclusion, and transportation dis-
advantage.3
The Japanese government has emphasized the
necessity of building a transportation system ca-
pable of resolving social problems by listening to
the opinions of various stakeholders, including
those of stakeholders for whom it is challenging
to speak up in social situations. The Japanese gov-
ernment is urging the implementation of flexible
3We refer to the national report published by the Japanese
Government Cabinet Secretariat, "Social Principles of Human-
Centric AI," accessed August 17, 2021.
and effective legislative systems that can include
socially vulnerable people by inspiring the efforts
of automotive manufacturers to develop technolo-
gies that can assist these people.
C. Literature Review and Our Contributions
Our study provides a pathway through which to
create inclusive FAVs. Previous works have shown
how FAVs could be appreciated/adopted by the
general public ((18)) and have usually concluded
that younger, richer and better educated individ-
uals are more likely to choose FAVs ((19), (20),
and (21)). Moreover, we find limited numbers of
previous works that have dealt with FAV demand
in the social exclusion context. A notable excep-
tion is (22), which focused on the role of AVs for
elderly people in overcoming hindrances to travel
because older adults are likely to experience trans-
port disadvantages, transport poverty, and the risk
of social exclusion. Inspired by (22), we answer the
5
Fig. 2: Explanations of AV Technology by Level.
Source: SAE International (2021, Accessed August 4, 2021)
following questions. What is the likely outcome
for individuals experiencing physical or cognitive
challenges? People faced with cognitive challenges
may find it difficult to operate new technologies
and to override controls ((23)). Despite the impor-
tance of answering this question in creating inclu-
sive FAVs, to the best of our knowledge, no pre-
vious works have answered it. Are factors closely
correlated with public acceptance of FAVs still
crucial for the acceptance/appreciation of FAVs
among risk-experienced people? What are the dif-
ferences between risk-experienced people and the
general public? The answers to these questions
can be extended to how risk-experienced people
respond to new technologies.
D. ND Victims
We focus on disaster victims because NDs are
massive shocks that substantially reshape individ-
uals’ RPs, which may prevent them from driv-
ing (and therefore cause them to prefer FAVs)
or adopting new technology. Previous works have
usually shown that risk-averse individuals are un-
likely to adopt FAVs ((24) and (25)), which stems
from the findings of previous works that risk-
averse individuals do not prefer to adopt new
technologies ((26);(27);(28)). Understanding the
6
different realms of risk perspective illuminates the
path through which to provide FAVs as inclusive
transport modes by suggesting how to include
people with/without diverse types of shock, who
thus drive the change in risk preference. These
differences imply that the factors motivating those
who have experienced shocks require different
solutions. To address this issue, we include a latent
variable that reflects people’s risk perspectives and
shows whether they prefer higher risks.
Thus, we proceed by expecting ND experiences
to increase RPs ((29), (30), (31), (32)) and that
increased RPs discourage victims from accepting
new technologies ((33), (34)). Such trends can also
be found in the field of transportation (which
is usually represented by a new mode choice or
driving behavior) ((35), (36), (37)). Among these
studies, previous works on FAVs have reported
that increased risk hinders the use and adoption
of FAVs ((38), (39), (40)). Regarding travel behavior,
disaster victims who have experienced trauma are
likely to lose their driving ability due to the fear
of road accidents; some studies have found that if
such individuals drive, then their traffic behaviors
will be positively correlated with their likelihood
of having automobile accidents ((41), (42)).
III. D
In this section, we present and describe the
survey, present the data, and explain our empirical
model referring to the previous literature.
A. Survey Method
We first conduct three online surveys in Japan
in 2015, 2016, and 2017. Surveys are administered
to individuals aged 18 years and older. We ran-
domly select the respondents while maintaining
their gender and age distributions in a manner
similar to those of the Japanese population.4Con-
sequently, we have 69,647 respondents. To assess
the sample representativeness of the survey, we
present the distribution of socioeconomic vari-
ables of the survey and Japanese census data
in Appendix Table A1. We find that there are
slight differences in gender and education levels
between our survey respondents and Japanese
census data. Before starting the large-scale sur-
vey, a presurvey is administered to calibrate the
questionnaires.
In what follows, we explain our data by dividing
the explanation into the areas of 1) survey struc-
ture, 2) FAV-related questions, 3) identifying those
who have experienced risks, 4) socioeconomic
variables and 5) latent variables.
B. Survey Structure
As mentioned in Section I, we first ask the
respondents about their WTB on FAVs (Step 1).
Then, we share the information on the Tesla au-
topilot accident that resulted in the driver’s death
(Step 2). Next, the respondents are again required
to write down their WTB (Step 3). Steps 1 and 3
are written separately. Figure 3shows the survey
items and answers. We present the survey items to
the respondents in the order of the item numbers.
4This process is possible because we employ an internet
survey conducted by Nikkei Research Inc., the largest research
company in Japan. Several trap questions are included in the
survey to identify respondents who have not seriously an-
swered the questions. Those who have not correctly answered
such trap questions are excluded from the survey company’s
sample in the collection process.
7
Next, we elicit the perceived FAV value (here-
after, FAV value) for FAVs, which refers to the per-
ceived value of FAVs expressed in monetary terms.
Studies in the field of transportation have con-
cluded that perceived value works as a predictor of
purchase intention. ((43);(44)), and (45) and (46)
suggested the role of perceived value in terms of
autonomous vehicles and shuttles. Such analyses
are recommended because perceived value indi-
cates an influence on behavioral intentions, which
this study aims to consider by constructing latent
variables. Thus, in this study, FAV values work as
an index of appreciation for FAVs rather than the
costs of automation, which exhibit a negative re-
lationship with purchase intention. Therefore, it is
quite natural that a higher FAV value is positively
correlated with a higher chance of purchasing
FAVs.
For question number Q3-1, on accident aware-
ness, we do not sort our respondents according to
their answers; rather, we use accident awareness
as a dummy variable equal to 1 if a respondent
is aware of the accident and 0 otherwise. Because
we are interested in how people’s choices change
according to the accident information shared with
them, we take Question Q3-2 as the final WTB
decision and take the differences between Q3-2
and Q3-3 as one of our main dependent variables.
C. FAV-related questions
For the questions related to FAV purchase in-
tention, the respondents are asked the following
question: “Do you want to add a completely self-
driving option that allows you to move around
when you purchase a car in the future?" The re-
spondents are then offered the following response
options: “(1) Purchase for sure, (2) Purchase under
certain conditions, (3) Do not purchase, and (4) I
don’t know". Given that FAVs had not yet been
fully introduced into the market in 2017, we as-
sume that people who show an affinity for FAVs
can be potential future consumers.5Therefore, we
include those who respond (1) or (2) in a group of
‘potential consumers’ because they show affinity
for FAV use. Moreover, people who respond (3)
or (4) are reluctant to purchase FAVs, and thus,
we do not consider them potential consumers.
Therefore, we code WTB as equal to 1 if a respon-
dent belongs to the potential consumer group and
0 otherwise. Therefore, our analysis allows us to
identify which types of factors shift consumers
who belong to (3) or (4) to (1) or (2).
Note that we make a clear distinction be-
tween “adding” a completely self-driving option
and “purchasing” an FAV by asking the following:
“When you purchase a car in the future, do you
want to add a completely self-driving option that
allows you to move around?” Such a setting al-
lows us to clearly distinguish between ‘purchasing
a car’ and ‘adding a self-driving option’, which
allows us to exclude potential deviations in the
results due to factors associated with purchasing
a car, such as car price and vehicle attributes.
We want to clarify that we do not employ all
the information from the survey; that is, we do
not treat survey answers on WTB as ‘ordinal’, but
5We note that we do not believe that our data are outdated.
Given that FAVs were not completely introduced into the
market at the time and have yet to be introduced as of 2021, a
substantial change in the results, for example, a change in sign
or implications, is unlikely. Therefore, more attention should
be paid to the signs and relative comparisons of the coefficient
magnitudes of the latent constructs.
8
Fig. 3: Survey Structure
instead, we treat them as categorical variables.
For example, while it is possible to investigate the
result of ‘ordinal’ responses on WTB by treating 1
as not purchasing, 2 as not sure, 3 as considering
purchasing and 4 as purchasing, an increase from
1 to 2 does not necessarily indicate the probability
of a respondent purchasing an FAV. Thus, we ana-
lyze binary responses because we are interested in
whether a respondent wants to purchase an FAV.
Similarly, we do not investigate the multinomial
responses regarding WTB because we believe that
each response is independent; therefore, inves-
tigating how the decision on not to purchase
interacts with ‘no awareness’, for example, does
not fit our research.
D. Identifying Disaster Victims
We ask whether the respondents have experi-
enced physical/mental damage due to NDs. If so,
we then ask them to consult a list of different types
of disaster damage, allowing them to choose any
item that they have experienced, as illustrated in
Table I, which shows the number and proportion
of observations associated with each item in the
first column. Among the items listed in Table I,
we place particular emphasis on those that are
likely to be a ‘traumatic experience’, which covers
experiences that are likely to be closely related
to physical and mental problems such as anxi-
ety, major depression, nightmares, hypervigilance,
and panic attacks associated with trauma-related
stimuli according to previous works in the field of
medicine ((47), (48)).
9
The respondents are allowed to check multiple
items. For example, a respondent can check both
‘Severe Health Problems’ and ‘Severe Injury’. Sum-
ming up the findings from the aforementioned
works, we denote the following as ‘traumatic expe-
riences’: Entire Collapse of a Home (Item No. 1),
Home Destruction (Item No. 2), Severe Property
Damage (Item No. 3), Severe Furniture Damage
(Item No. 4.), Death of Family or Friends (Item No.
5), Severe Injury (Item No. 6), Posttraumatic Stress
Disorder (PTSD; Item No. 10), and Severe Health
Problems (Item No. 11). Consequently, we have
7,853 respondents, which is a sufficient number of
observations for us to conduct statistical analysis,
as presented in Table I.
E. Socioeconomic variables
To control for the respondents’ sociodemo-
graphic characteristics, we also include the follow-
ing sociodemographic variables: income, gender,
age, and number of family members. Moreover,
we include a car ownership dummy (=1 if the
respondent owns a car), commuting dummy (=1
if the respondent commutes) and license dummy
(=1 if the respondent has a driver’s license).
While we survey only those who are aware of
FAVs, it is possible that in 2017, the respondents
were less familiar with FAVs than their counter-
parts were in 2021. Thus, it is necessary to identify
the respondents who are not familiar with FAVs.
Thus, we exclude those who answer ‘I don’t know’
to all of the questions related to FAVs (5,987 ob-
servations). Finally, we exclude those who select ‘I
don’t know/I don’t want to answer’ for questions
about their individual income (30,156 observa-
tions). As a result, we have 69,391 respondents in
total.
Finally, we summarize our data in Table II,
which shows the descriptive statistics of the key
variables for the entire sample (the first column)
and each group in the following columns. The
summary statistics in Table II indicate that the
average perceived value gap among disaster vic-
tims is lower than those of respondents who are
not disaster victims. Such a trend indicates that
those who are disaster victims are more unlikely
to appreciate FAVs after the accident information
is shared. Moreover, we do not observe substantial
differences between socioeconomic variables (e.g.,
income, age, and commuting time) between those
who have experienced NDs and those who have
not.
We do not observe substantial differences be-
tween socioeconomic variables (e.g., income, age,
and commuting time) across groups. Then, we
proceed to the comparison between the latent
variables.
F. Explanatory Factor Analysis
We identify the latent variables that can be re-
lated to FAV choices based on, whenever possible,
statements previously used and found to be effec-
tive in the literature. Table III presents the names
of our latent variables, the items corresponding
to them, and a reference list. In the following two
paragraphs, we explain how we identify the latent
constructs. The survey consists of questionnaires
related to the following items:
1) RPs: Whether the respondent is risk-seeking
or risk-averse, which captures whether ex-
10
TABLE I: Criteria for Selecting Disaster Victims
No. Item Obs. Portion(%) Public Disaster Victims
1 Entire Home Collapse 1,474 1.47% X O
2 House Destruction 2,000 2.00% X O
3 Severe Property Damage 7,045 7.04% X O
4 Severe Furniture Damage 6,921 6.91% X O
5 Death of Family and Friends 1,137 1.14% X O
6 Severe Injury 1,019 1.02% X O
7 Evacuation 1,741 1.74% X X
8 Moving House 2,429 2.43% X X
9 Unemployment 1,369 1.37% X X
10 Posttraumatic Stress Disorder 879 0.88% X O
11 Severe Health Problems 836 0.84% X O
12 Others 3,591 3.59% X X
13 No Experience 69,647 70.92% O X
14 I Don’t Know 9,049 7.23% X X
Total 100,810 100% 69,647 21,915
Analytical Sample Total 57,105 7,853
TABLE II: Descriptive Statistics of Analytical Sample
Variable Mean Std. Dev. Min Max
Those without Disaster Experience (N=57,105)
Purchasing Intention (Before Accident) 0.593 0.687 0 2
Purchasing Intention (After Accident) 0.657 0.666 0 2
Perceived Value (10,000 JPY) 66.345 126.816 0 1,000
Gender (=1 if Female) 0.315 0.465 0 1
Individual Income (10,000 JPY) 470.094 401.723 100 3500
Age 50.747 11.17 18 99
Car Ownership (=1 if Owns a Car) 0.80 0.430 0 1
License (=1 if Owns a drivers license) 0.92 0.269 0 1
Accident Awareness 0.264 0.441 0 1
Those Disaster Victims Experience (N=7,853)
Purchasing Intention (Before Accident) 0.694 0.724 0 2
Purchasing Intention (After Accident) 0.742 0.687 0 2
Perceived Value (10,000 JPY) 76.434 143.846 0 1,000
Gender (=1 if Female) 0.272 0.445 0 1
Individual Income (10,000 JPY) 470.094 401.723 100 3500
Age 52.189 11.618 18 99
Car Ownership (=1 if Owns a Car) 0.832 0.374 0 1
License (=1 if Owns a drivers license) 0.925 0.268 0 1
Accident Awareness 0.242 0.428 0 1
posure to traffic accidents and NDs changes
RPs and whether such changed RPs are re-
flected in the respondents’ choices of new
technologies.
2) SP: The extent to which a respondent trusts
the government, local authorities, family
and friends during emergency situations. In
other words, this item refers to whether a
respondent can expect mental support from
the above actors.
3) FAV-related questions: Awareness of FAVs,
perceived value for FAVs, WTB on FAVs,
expected benefits of FAVs or merits (Tech-
savviness) and whether respondents fear
FAVs due to potential accidents (Accidents).
4) Environmental Concerns: Importance of
11
waste recycle (Recycle), resolving green-
house gas emissions (GHG Emission), alle-
viating air pollution (Air Pollution), and pre-
serving natural environment (Conservation).
Table III presents the list of questions and pre-
vious works to which we refer to create survey
questions according to the latent categories. First,
for RPs, using the examples that can be found
in daily routines, we select the questions that
can measure whether the respondents prefer high
levels of risks. If a respondent believes that he or
she is risk-seeking, then he or she chooses ‘Yes’
and ‘No’ otherwise.
For the SP variables, the respondents are asked
to answer questions about their beliefs in SP from
diverse authorities on a scale from 1 to 5, where
(1) indicates ‘strongly disagree’, (2) indicates ‘dis-
agree’, (3) indicates ‘neither agree nor disagree’, (4)
indicates ‘agree’ and (5) indicates ‘strongly agree’.
Next, we ask about the respondents’ concerns
regarding the environment in terms of its impor-
tance for policy. Based on previous studies, we
classify environmental policy topics into eight fac-
tors referring to the House of Councilors, National
Diet of Japan (2015). We have 11 questions in total,
and the topics cover air pollution, environmental
conservation, water pollution, endangered species
conservation (biodiversity), reuse and recycling,
waste disposal, and CO2emissions with questions
such as “How important is the policy to you?” The
scale of responses is as follows: (0) for no aware-
ness/interest at all—meaning that the difference
between those who answer (0) and others depends
on whether a person at least has an interest in a
certain policy/issue; (1) for very nonsignificant; (2)
for nonsignificant; (3) for neither important nor
nonsignificant; (4) for important; and (5) for very
important.
According to the context of the questions, we
divide the questions into importance of waste re-
cycling (Recycle), resolving greenhouse gas emis-
sions (GHG Emission), alleviating air pollution (Air
Pollution), and preserving the natural environ-
ment (Conservation) according to the relevance
of the survey item to the environmental concerns
context.
12
IV. M
A. Discrete-Choice Modelling
In our study, we separately identify two com-
ponents: a discrete choice model and a latent
variable model. Each of these submodels consists
of a structural component and a measurement
component. In the discrete choice component,
the alternatives’ utilities may depend on both the
observed and latent attributes of the alternatives
and characteristics of decision makers. Consistent
with the random utility maximization model, util-
ity as a theoretical construct is operationalized by
assuming that individuals choose the alternative
with the greatest utility.
For the discrete-choice part, we adopt mixed
logit referring to (60), which allows for unob-
served preference heterogeneity among respon-
dents. Mixed logit is a highly flexible model that
can approximate any random utility model ((61))
and overcomes the three limitations of standard
logit by allowing for random taste variation, un-
restricted substitution patterns and correlations
between unobserved factors over time.
In our model, Let Ynj represent a discrete choice
of a person namong Jalternatives. Suppose Unj is
the utility of the jth choice to ith individual. Then,
with the deterministic utility Vnj , and random
error term nj , the utility Unj can be written as;
Unj =β0
nxnj +nj (1)
where xnj are the vector of observed variables
that relate to the alternative and the decision-
maker, βnis the vector of coefficients of these
variables for person nrepresenting that person’s
tastes, and nj is a random term that is extremely
valuable.
In the mixed logit specification, the coefficient
varies over individuals in the population with
density f(β). This population density is a func-
tion of parameter θ, which represents the mean
and covariances of the vector βin the popula-
tion. In other words, this specification is identical
for standard logit except that the βcoefficient
varies across individuals rather than being fixed.
We allow for one or more of the parameters in
the model to be randomly distributed and thus
allow for the coefficients in the model to vary
across decision makers. In this specification, the
individual is aware of his or her own βiand nj
for all jand chooses alternative iif and only if
Uni > Unj ∀j6=n. If βnis observed, then the
choice probability is a standard logit; therefore,
the probability conditional on βnis
Lni(βn) = eβ0
nxni
Σjeβ0
nxni (2)
However, the researcher observes xnj but not βn
or nj . Therefore, the unconditional choice prob-
ability is the integral of Lni(βn)over all possible
variables of βn:
Pni(βn) = ZPni (βn)f(βn|θ)dβn(3)
which is the mixed logit probability. In this
study, we specify the distribution f(β), following
(62) and (63), to be normal: β∼N(b, W )with
estimated parameters band W.
13
TABLE III: Latent Variables, Explanation and Works Referenced
Explanation Source
Latents on Risk, Disaster and Accidents
Risk Perception (RP) (24), (25), (26)
Risk-Averse 1. Do you have a comprehensive medical examination on a regular basis?
Risk-Averse 2. Do you purchase travel insurance at your own expense
when travelling domestically or abroad?
Risk-Seeking 1. Do you gamble (pachinko, horse-racing,
bicycle racing, auto racing, etc.)?
Risk-Seeking 2. Do you prefer high risk and high return to
low risk and low return when purchasing financial products?
Social Support (SP)(49), (50), (51)
When faced with difficulties such as
supply, money, and housing in times of disaster,
choose the one that best describes your thoughts.
SP1. I expect physical/mental support from the government,
local authorities and public institutions.
SP2. I expect physical/mental support from family members and friends.
SP3. I expect physical/mental support from local volunteers and
members of local communities.
Latents on FAVs
Fear (FE) (52), (53), (54)
Accident 1. There is a possibility that children will be able to move it on their own.
Accident 2. There is a possibility that the software will be hacked (cybersecurity).
Accident 3. A malfunction may cause accidents.
Malfunction 1. It is unclear who is responsible for an accident due to FAV technology.
Malfunction 2. Traffic volume and congestion may increase because those without a license can drive.
Malfunction 3. A malfunction may take me to the wrong destination.
Merits (MR) (55), (56), (57)
Burden Reduction 1. People can drive without a license.
Burden Reduction 2. Burdens on driving will be decreased.
Burden Reduction 3. Able to do other work while driving (multitasking).
Safety 1. Children can drive a vehicle without a guardian.
Safety 2. Able to avoid responsibility for traffic accidents.
Environmental Concern (58), (59), (59)
GHG Emissions 1 Reducing annual greenhouse gas (GHG) emissions
GHG Emissions 2 Percentage of eco-cars out of total cars
GHG Emissions 3 The usage of renewable energy in power generation
Recycle 1 Final disposal volume of garbage and waste
Recycle 2 Cycle utilization rate (the percentage of the total amount of
reusable and recycled materials to be injected into society)
Air Pollution 1 Alleviating particulate matter (PM 2.5.) pollution is critical for our society.
Air Pollution 2 Resolving air pollution (particularly photochemical smog) is important
Conservation 1 Percentage of endangered species around
Conservation 2 Biodiversity
Conservation 3 The percentage of green area within 1,500 meters of one’s home
Conservation 4 Green purchasing: When purchasing goods and services,
I consider the environmental impact before purchasing.
B. ICLV Model Framework
In this study, we adopt the ICLV approach,
which incorporates discrete choices and latent
variables. For the discrete choice components,
the utility of the alternatives may be determined
by both their observed and latent attributes, as
well as the characteristics of the decision makers.
Utility, as a theoretical construct, is quantified by
assuming that individuals choose the alternative
with the greatest utility, which is consistent with
the random utility maximization model.
Conversely, the latent variable part is quite flexi-
ble, enabling simultaneous relationships between
latent variables. The latent variables themselves
are assumed to be measured by multiple indi-
cators such as responses to Likert-scale survey
questions.
Mathematically, the model is typically repre-
sented using the following set of four equations:
Unj =β0
nxnj +γ0
nx∗
n+nj (4)
14
in=Dx*n+ηn(5)
Ynj =
1,if Unj ≥Unj0forj0∈1, ..., J .
0,otherwise
(6)
where Unj refers to utilities of each of the J
alternatives, as perceived by decision maker n;
xnj is the (K×1) vector of the observed ex-
planatory variables; x∗
nis the (M×1) vector of
the latent explanatory variables; βnand γnare
the vectors of model parameters denoting sen-
sitivities to the observable and latent variables,
respectively; and nj is the stochastic component
of the utility specification. Moreover, inis the
(R×1) vector of the indicators used to measure
the latent variables; Dis the (R×M)matrix of
model parameters denoting the sensitivities of the
measurement indicators to the latent variables;
ηnis the (R×1) vector denoting the stochastic
component of the measurement equation; and Ynj
is the choice indicator, equal to one if decision
maker n chooses alternative jand zero otherwise.
The stochastic components nj and ηnare as-
sumed to be mutually independent.
The ICLV approach has a remarkable advantage
over the conventional choice model. In conven-
tional analyses of travel mode choice, the utility
is determined by tangible aspects such as travel
time and cost and socio-demographic variables
including age, gender, and income ((64)). In the
last two decades, the ICLV approach has been
proposed as a suitable way of including intangible
aspects by considering travelers’ perceptions
and attitudes as determinants of choice behavior
((65)). Using ICLV models, researchers have found
that subjective attitudes have statistically signif-
icant effects, and including them improves the
explanatory power of the choice model ((66), (67),
(68), (69)). Because of the advantage, ICLV models
have been frequently utilized in recent studies to
more precisely explain travel mode choice behav-
ior.
Figure 4describes the ICLV model framework.
In this setting, an individual’s choice is deter-
mined by the utility of using each travel mode,
and the utility is explained by both observed and
latent variables. The value of a latent variable
represents the strength of a certain subjective
perception of an individual (e.g., perception to-
wards social support), which is estimated from the
responses to several indicator variables connected
to the perception (e.g., how much one expects
physical/mental support from the government).
Since the latent variable is expressed as a variable
with a mean of 0 and a standard deviation of 1,
the coefficient of the latent variable indicates the
change in the probability of choice when the la-
tent variable increases by one standard deviation.
V. R
In this section, we first explain our results in
general in Subsection V-A and then demonstrate
the substitution pattern in Subsection V-B.
A. Main Results
Table IV reports the estimation results of FAV
choices before the sharing of accident-related
15
Fig. 4: Framework of ICLV model
news (Model (1) for those without disaster experi-
ences and Model (3) for disaster victims) and after
the sharing of accident-related news (Model (2)
for those without disaster experiences and Model
(4) for disaster victims). Panel (A) of the results
table reports the latent variables (interacted with
the disaster victims dummy); Panel (B) presents
the results for the socioeconomic variables; and
Panel (C) shows the results for the heterogeneity
coefficients, constants and model fit. We proceed
mainly by comparing the coefficient estimates of
Models (1) and (2). Overall, we find that SP is
pivotal for FAV choices and that such trends were
stronger for the disaster victims.
a) Notes on Interpretation.: Before interpret-
ing the results, we want to clarify that people
can have different combinations of latent vari-
ables. For instance, people can have high levels of
‘SP’ and ‘Risk-Averse.’ For example, a lower level
of ‘Risk-Averse’ is negatively correlated with FAV
choices, which does not indicate that SP for that
individual discourages his or her choice of FAV.
Potential consumers express affinity toward FAVs
if SP is given. Thus, our result shows the changes
in FAV choices following one-standard-deviation
increases in a latent variable, holding the other
latent variables fixed. All regressions include pre-
fecture, year and individual fixed effects. We also
calculate individual-level heterogeneity, which is
noted in the "Heterogeneity Coefficient" section
of Panel (C).
b) SP: SP (in Panel A-1) shows positive and
significant coefficients in all models and displays
an increasing trend after the accident news is
shared. Previous works in the field of medicine
((70)) have revealed that SP is likely to boost the
resilience of disaster victims and that resilience is
not only the ability to deal effectively with nega-
tive situations and instantly recover from negative
impacts but also mental preparedness for future
16
situations and vulnerabilities. Thus, our results in-
dicate that while SP is more effective when people
are exposed to accidents, those who choose FAVs
are likely to have higher levels of SP. This observa-
tion is consistent with anecdotes from documents
published by the Japanese government indicating
that the support of local authorities and family
is vital for achieving inclusive transportation by
encouraging people to choose FAVs.6
c) Risk Preferences: Panel A-2 displays the
results on risk preferences, accidents and merits.
First, risk-averse individuals are likely to adopt
FAVs, while risk-seekers are unlikely to do so. The
positive coefficients of risk aversion in Models (1)
and (3) indicate that risk aversion is positively
correlated with FAV adoption before the accident
news is shared, while it slightly decreases in Mod-
els (2) and (4).
The positive coefficient of risk aversion and
FAV adoption is consistent with previous works;
people mostly have positive perceptions of FAVs.
Driverless vehicles (without human driving con-
trols) are perceived as safer than vehicles with no
automation ((71)). Thus, it is quite natural to infer
that risk-averse individuals do not prefer fatal
accidents and therefore are likely to avoid such
risks. In this sense, the decline in the coefficient
in Models (2) and (4) is quite natural.
Likewise, FAVs are less preferred by those who
are risk seekers. The preferences of risk seekers
decrease even after the accident news is shared;
for the general public, the preference decreases
from -0.091 (Model (1)) to -0.230 (Model (2)), and
6We refer to the Japanese Cabinet’s guideline entitled “Social
Principles of Human-Centric AI" (2020).
for disaster victims, the preference decreases from
-0.084 (Model (3)) to -0.195 (Model (4)).
d) Accident and Merits: The general public
and disaster victims show indifference—a statisti-
cally insignificant coefficient—before the accident
news is shared, which then decreases significantly
after the accident news is shared. While the gen-
eral public indicates concern about such malfunc-
tion, leading to a reduction in the coefficients,
disaster victims do not.
Disaster victims care about safety more than the
public does, as seen from the positive coefficients
in Models (3) and (4). As expected, after the ac-
cident news is shared, a one-standard-deviation
increase in safety is likely to cause more disaster
victims to choose FAVs.
e) Environmental Perceptions: Panel A-
3 presents the estimated coefficients of
environmental perceptions. Two findings are
worth discussing. First, on the one hand,
those who are satisfied with greenhouse gas
(GHG) emissions and air pollution before
the accident are unlikely to choose FAVs; on
the other hand, conservation shows positive
coefficients across groups. Second, the statistical
significance of environmental perceptions after
the accident news is shared mostly disappears.
This trend is stronger for disaster victims, as
some of the parameters (i.e., GHG emissions and
conservation) remain significant for the public.
f) Socioeconomic Variables: Regarding the re-
maining parameters, as expected and consistent
with the previous works on FAV demand, socioe-
conomic attributes mostly exhibit significant co-
efficients, which implies that these characteristics
17
also play a role in FAV demand. For example,
perceived value, being male, being younger in age,
having higher income, being a commuter, and
having a license show higher coefficients. Those
who were already aware of the accident show
negative coefficients. One difference between the
public and disaster victims is that while the public
shows higher coefficients when they have a car,
disaster victims do not.
g) Heterogeneity Coefficients: The random co-
efficient results show a significant level of het-
erogeneity in the respondents’ preferences. Given
that the random coefficients are significant and
that these parameters help alleviate the well-
known problem of the independence of irrelevant
alternatives shown by traditional logit models, the
random coefficients play a critical role in defining
the substitution patterns, as explained by (72).
In summary, our demand estimation indicates
the importance of latent factors such as SP
and that disaster victims show different trends
concerning these factors, motivating their FAV
choices. We then take the next step: we consider
those who switch their decisions after the accident
news is shared. Specifically, we analyze those who
answer “do not purchase” before the accident
news is shared but switch to “purchase” after the
accident news is shared. Do the factors fostering
FAV choices change? Does SP still motivate the
public and disaster victims to choose FAVs to
overcome negative information? We answer these
questions by examining the substitution patterns
in the next subsection V-B.
B. Substitution Patterns
Table Vshows the estimation results. Model (1)
shows the estimated coefficients of those who are
not disaster victims, while Model (2) presents the
results of those disaster victims’ experiences. For
the substitution patterns, we estimate the demand
of those who answer “not purchase” before the
news regarding the accident is shared but decide
to “purchase” after the accident news is shared.
Model specifications such as fixed effects and
heterogeneity coefficients are the same as in the
model presented in Table IV. One difference from
the main models is that we employ the perceived
value gap, a product of perceived value before
accident news is shared minus that after accident
news is shared.
Focusing on SP and risks (Panel A-1), column
1 shows that SP supports FAV adoption among
the public by 4.92% (coefficient of SP), while the
impact of SP remains ambiguous to disaster vic-
tims (coefficient of -0.0003). Thus, shifting disaster
victims to choose FAVs through SP may not be an
effective approach. However, note that the coef-
ficient of SP is statistically significant for people
without disaster experience. Therefore, our result
still supports that SP is crucial when people shift
from non-FAVs to FAVs.
The difference between the public and disaster
victims may stem from risk preference, as dis-
played in Panel A-2. The coefficients of the risk
preferences of the public are nonsignificant, while
they show negative coefficients for disaster vic-
tims. Such results indicate that regardless of risk,
disaster victims’ preferences for FAVs are likely to
be discouraged if the accident news is shared.
18
The environmental perceptions and socioeco-
nomic variables listed in Panels A-3 and B imply
the trends of people who are likely to switch to
FAVs. For environmental perceptions, those who
are not satisfied with the conservation of the
natural environment and those who are satisfied
with the reduction in waste and GHG emissions
(public) are likely to switch to FAVs after the acci-
dent news is shared. The rest of the parameters are
consistent with the main result in Table IV, while
accident awareness shows a positive coefficient.
Thus, those who decide to switch to FAVs after the
accident news is shared are likely to be aware of
potential accident concerns. For heterogeneity, we
confirm the individual heterogeneities of 0.1354
from Model (1) and 0.1227 from Model (2), which
are significant.
VI. O D
FAVs are not yet introduced in the market in
any part of the world.7Therefore, people would
have a conceptual sketch of how FAVs work and
would consider whether they would purchase an
AV according to that impression. Therefore, com-
pared to the actual technical level, the perception
of FAVs would not vary significantly by country.
Then we must control for the differences in
how people perceive FAVs. Thus, we include the
possible factors that can affect people’s percep-
tions (latent) in the model. To do so, we identify
possible factors that can affect people’s percep-
tions (latent) by searching through the previous
works and found the following latent factors: risk
perceptions, concerns about accidents, merits, en-
vironmental concerns, and social support.
Thus, 1) because our study is based on how
people perceive FAVs rather than the actual tech-
nical level, and 2) because we control for people’s
perceptions through latent variables, substantial
changes in the results due to technological differ-
ences are less likely to happen across other coun-
tries. More attention can be given to the signs and
relative comparisons of coefficient magnitudes of
the latent constructs.
A. Policy Implications
Providing governmental (or local institutional)
guidelines on the prior knowledge of FAVs would
significantly alleviate the negative influence of
concerns associated with the locus of losing con-
trol. Most of the obstacles that hinder disaster
7This is not due to a lack of technological innovations but
more related to legal issues, particularly concerning responsi-
bility in times of accidents.
19
TABLE IV: Main Estimation Result
Model (1) Model (2) Model (3) Model (4)
Before Accident Accident Accident Before Accident Accident Accident
Without Disaster Disaster Victims Without Disaster Disaster Victims
Coefficient Std. Err Coefficient Std. Err Coefficient Std. Err Coefficient Std. Err
Panel (A): Latents
Panel (A-1): Social Support
Social Support (SP) 0.101*** 0.013 0.172*** 0.026 0.111*** .013 0.163*** .028
Panel (A-2): Risk, Accident and Merits
Risk-Averse 0.214*** 0.023 0.166*** 0.049 0.175*** 0.023 0.148*** 0.051
Risk-Seeking -0.091*** 0.027 -0.230*** 0.053 -0.084*** 0.028 -0.195*** 0.056
Accident -0.006 0.056 -0.427*** 0.121 0.015 0.058 -0.216* 0.127
Malfunction -0.038 0.056 -0.417*** 0.122 -0.002 0.058 -0.171 0.128
Safety -0.015 0.025 0.019 0.053 0.057** 0.027 0.128** 0.057
Burden Reduction 0.377*** 0.029 0.308*** 0.062 0.298*** 0.030 0.248*** 0.065
Panel (A-3): Environmental Perceptions
Recycle 0.020 0.021 0.004 0.041 0.010 0.022 -0.001 0.043
GHG Emission -0.059*** 0.018 -0.071** 0.035 -0.037** 0.018 -0.029 0.037
Air Pollution -0.109*** 0.017 -0.026 0.034 -0.090*** 0.018 0.001 0.036
Conservation 0.037* 0.020 0.077* 0.040 0.037* 0.021 0.060 0.042
Panel (B): Socio-Economic Variables
ln Perceived Value 0.305*** 0.051 0.385*** 0.052 0.328*** 0.054 0.360*** 0.061
Female -0.257*** 0.026 -0.252*** 0.056 -0.125*** 0.027 -0.125** 0.058
ln Age -0.362*** 0.047 -0.634*** 0.100 -0.237*** 0.048 -0.437*** 0.106
ln Income 0.162*** 0.015 0.079** 0.031 0.117*** 0.016 0.045 0.033
Commuter 0.055** 0.024 0.019 0.051 0.037 0.025 0.045 0.053
Car Ownership 0.233*** 0.027 0.133** 0.060 -0.102*** 0.028 -0.143** 0.064
Accident Awareness -0.401*** 0.022 -0.347*** 0.048 -0.295*** 0.022 -0.225*** 0.050
License 0.572*** 0.040 0.665*** 0.086 0.234*** 0.040 0.225** 0.088
Panel (C): Others
Heterogeneity Coefficient 0.361*** 0.090 0.262*** 0.083 0.470*** 0.099 0.423*** 0.119
Constant -1.198*** 0.326 0.947 0.579 -0.424 0.341 1.150* 0.621
Log-Likelihood -40,467.62 -38,410.474
Note: Standard errors in parentheses. * p<0.1, ** p<0.05, *** p<0.01.
N=69,391
N of Without Disaster=57,105
N of Disaster Victims=12,286
Prefecture Fixed Effects are included.
victims from choosing FAVs derive from the fear of
losing control, particularly in times of accidents,
which thus renders disaster victims more fearful
than others of new technology ((73), (74)). In that
sense, our study reaffirms the findings of past
studies that have noted the impact of prior knowl-
edge on attitudes toward new types of vehicles
(e.g., electric cars) and on intentions to use them
((75)). Such a finding would apply to both disaster
victims and people without disaster experiences.
According to previous medical studies on trau-
matic symptoms ((76)), family and peer sup-
port can encourage disaster victims to adapt to
new technologies or return to driving. Rather
than hastily pushing disaster victims to adopt
and appreciate FAVs, gradually letting them adopt
FAV usage through repeated short-term travel or
simulation-based training with the support of
family and friends is desirable. Such a conclusion
also implies that social support can work as a
stimulus for disaster victims to overcome the fear
of disasters, which hinders them from choosing
FAVs.
B. Managerial Implications
a) SPs: Our results show that SP plays a criti-
cal role in persuading both the public and disaster
victims to choose FAVs. This finding is consistent
with previous works showing that family encour-
agement is a suitable stimulus for socially vulner-
able people to accept new technologies ((77)) and
extends the works of (78) and (79), which incorpo-
rate the effect of social interaction into the choice
20
TABLE V: Substitution Result
Model (1) Model (2)
Without Disaster Disaster Victims
Coefficient Std. Err Coefficient Std. Err
Panel (A): Latents
Panel (A-1): Social Support
Social Support (SP) 0.0492*** 0.0126 -0.0003 0.0262
Panel (A-2): Risk, Accident and Merits
Risk-Averse -0.0130 0.0226 -0.0947* 0.0489
Risk-Seeking 0.0282 0.0270 -0.1355*** 0.0526
Accident -0.2403*** 0.0563 -0.3588*** 0.1220
Malfunction 0.2039*** 0.0569 0.3113** 0.1229
Safety 0.0760*** 0.0253 0.1047** 0.0529
Burden Reduction -0.0978*** 0.0294 -0.1501** 0.0628
Panel (A-3): Environmental Perceptions
Recycle 0.0444** 0.0213 0.0716* 0.0415
GHG Emissions 0.0694*** 0.0180 0.0416 0.0358
Air Pollution -0.0168 0.0173 0.0282 0.0344
Conservation -0.0893*** 0.0203 -0.1198*** 0.0399
Panel (B): Socio-Economic Variables
ln Perceived Value 0.0146 0.0296 0.0224 0.0400
Female 0.3071*** 0.0253 0.3025*** 0.0547
ln Age 0.1314*** 0.0458 0.1800* 0.0978
ln Income -0.0785*** 0.0149 -0.0485 0.0311
Commuter -0.0778*** 0.0238 -0.1446*** 0.0500
Car Ownership -0.3311*** 0.0257 -0.2131*** 0.0587
Accident Awareness 0.3967*** 0.0211 0.4583*** 0.0465
License -0.2413*** 0.0360 -0.3630*** 0.0787
Panel (C): Others
Heterogeneity Coefficient 0.1354*** 0.0376 0.1227*** 0.0420
Constant -0.0668 0.2396 -0.8221 0.5357
Log-Likelihood -40,467.62 -38,410.474
Note: Standard errors in parentheses. * p<0.1, ** p<0.05, *** p<0.01.
N=69,391
Prefecture Fixed Effects are Included.
model. Specifically, this work contributes by 1)
examining disaster victims and 2) including other
latent variables associated with risk preferences,
accidents, merits, and environmental concerns.
While our results show that SP encourages people
without disaster experiences to shift to choose
FAVs, we find relatively weak evidence for such
a trend for disaster victims.
Thus, how can disaster victims be encouraged
to shift to FAVs? Promoting the practical benefits—
safety—would help with this shift. The coefficient
of safety for disaster victims is higher, positive,
and significant. At the same time, the coefficient
of accidents for disaster victims is negative, indi-
cating that disaster victims are afraid of potential
accidents. Therefore, increasing the reliability of
government entities and providing safety guide-
lines may attract both disaster victims and the
public because our results still show that SP plays
a role in public perception. Thus, our results do
21
not indicate that promoting SP is futile; neverthe-
less, SP is a critical factor in encouraging people—
those without disaster experience—to choose or
substitute with FAVs. Therefore, there is still a
need for policies that ensure SP as a viable so-
lution for boosting FAV adoption.
b) Risk Preferences: Another interesting find-
ing from this research is that we reveal that people
regard FAVs as a safe option. Risk-averse indi-
viduals prefer FAVs, and their preference declines
when faced with the news of the accident. There-
fore, from a policy implication perspective, in line
with SP, providing governmental support to design
and implement regulations on safety—which fos-
ters trust in FAVs—is recommended. These policy
measures fundamentally include the risk realloca-
tion process ((24)).
c) Environmental Perspective: With regard to
other types of parameters, while the previous
works have revealed that a higher level of environ-
mental concern motivates FAV adoption, interest-
ingly, our results reveal that different types of envi-
ronmental perceptions may have different impli-
cations. Our results emphasize the importance of
separately identifying environmental perceptions
according to context. Another interesting finding
from our results is that the coefficient magnitudes
of environmental perceptions are likely to disap-
pear after the accident news is shared, indicating
that people prioritize safety and practical benefits
over environmental perceptions. This result does
not imply that environmental perception is less
important; the concern of previous works and in-
dustry reports has been that FAVs might increase
the total vehicle miles traveled (VMT), as they
allow many people to travel freely (Center for
American Progress, 2016).
The impact of FAVs on energy use and emissions
largely depends on their effect on the total VMT,
their fuel efficiency and their fossil fuel consump-
tion. For example, Stephens et al. (2016) estimated
that the highest VMT increase and the smallest
efficiency increase could result in a 205 percent
increase in US transportation energy use. These
concerns again indicate the importance of envi-
ronmental perceptions, which make drivers aware
of their driving distances and the consequences of
their driving to the environment. As mentioned
in (16), another way to reduce emissions and
energy usage is to promote shared FAVs and elec-
tric FAVs and manage road infrastructures. Such
policies would be more effective if environmental
perceptions were promoted, as environmentally
concerned consumers are more likely to adopt
such policies.
VII. C
Achieving inclusive transportation requires ac-
commodating socially vulnerable people who can-
not drive unassisted. Moreover, because accidents
are inevitable, it is necessary to estimate demand
considering situations that occur after accidents.
Using a unique survey design and a large sample
size, our study discovers that SP is pivotal in mo-
tivating disaster victims to adopt and appreciate
FAVs. We find additional evidence on the relation-
ship between other psychometric and socioeco-
nomic factors and FAV demand. By considering
the individual-level demographic characteristics
and psychometric attributes of SP, risk prefer-
22
ence, environmental perceptions, accident con-
cerns, and merits, we construct a discrete-choice
model that considers individual heterogeneity. By
including such characteristics, our results have
implications for the design of effective policy in-
struments and information campaigns that appeal
to disaster victims. Therefore, our findings provide
clear, practical contributions.
Our results are straightforward. The social sup-
port of family, friends, and local authorities is
a crucial factor in motivating people to appre-
ciate and purchase FAVs. To prove this, we pro-
vide quantitative evidence of newly introduced
FAVs with an additional focus on natural disaster
victims–people who have become physically or
mentally challenged due to severe disaster dam-
age, including those with post-traumatic stress
disorder. Additionally, by comparing how respon-
dents react before/after accident news is shared
with them, we reconfirm the importance of social
support.
Our empirical framework can be extended to
a global context, as many countries are keen to
achieve inclusive transportation and encourage
people to choose FAVs. Ultimately, by illuminating
the importance of social support on FAV choices,
our study suggests the importance of providing
prior knowledge through institutional guidelines
and enticing disaster victims to choose FAVs by
increasing support from family and friends. Doing
so could allow FAVs to be inclusive in the future.
We conclude by suggesting the following policy
guidelines: increasing the reliability of govern-
ment entities and providing safety guidelines may
encourage both disaster victims and the public to
adopt FAVs because our results still show that SP
plays a role in the perceptions of the public.
Our study is based on short-term shock, and
the provided implications are short-term based.
A possible future study examining the long-term
effect of accident news would require panel data,
for example, of several years after such news is
shared. Unfortunately, we lack the data necessary
for such analysis. Thus, future research into the
long-term effects of accident news is worth pur-
suing. This is because even though the Tesla acci-
dent can have long-term negative effects such as
damages to the company’s reputation, we cannot
determine how long the fear of the accident would
last. For example, the fear due to the accident
could decrease (or increase) as time passes.
This paper calls for further studies related to
FAV adoption. As an example, future studies could
examine the different aspects of SP in the FAV con-
text. The findings of this study suggest practical
policy guidelines by illuminating the importance
of SP. However, it would be meaningful to investi-
gate the behavioral mechanisms of how different
types of SP influence FAV adoption. For example,
governments could provide SP in the form of
financial subsidies, thereby reducing RPs. Overall,
because FAVs will substantially impact the future
of transportation systems and it is necessary to
include those who are socially vulnerable, the
interplay between FAVs and SP is a fertile area for
future research.
23
Acknowledgements
Our special thanks go to Seilin Uhm, University
College London, Institute of Education,
Kyungwoong Koh, Faculty of Economics, Johns
Hopkins University, Kohei Kawasaki and Thierry
Coulibaly, Urban Institute, Kyushu University,
and the seminar participants at the University of
Tokyo ( Japan) and Kyushu University (Japan) for
their helpful comments. This work was supported
by JSPS KAKENHI Grant Number JP20H00648.
R
[1] M. A. Figliozzi, “Carbon emissions
reductions in last mile and grocery
deliveries utilizing air and ground au-
tonomous vehicles,” Transportation Research
Part D: Transport and Environment,
vol. 85, p. 102443, 2020. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S1361920920306301
[2] Z. Le Hong and N. Zimmerman,
“Air quality and greenhouse gas im-
plications of autonomous vehicles in
vancouver, canada,” Transportation Research
Part D: Transport and Environment,
vol. 90, p. 102676, 2021. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S1361920920308609
[3] S. M. Mousavi, O. A. Osman, D. Lord, K. K.
Dixon, and B. Dadashova, “Investigating the
safety and operational benefits of mixed traf-
fic environments with different automated
vehicle market penetration rates in the prox-
imity of a driveway on an urban arterial,”
Accident Analysis & Prevention, vol. 152, p.
105982, mar 2021.
[4] S. Tscharaktschiew and C. Evangelinos,
“Pigouvian road congestion pricing un-
der autonomous driving mode choice,”
Transportation Research Part C: Emerging
Technologies, vol. 101, pp. 79–95, 2019. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S0968090X18315328
[5] M. D. Simoni, K. M. Kockelman, K. M.
Gurumurthy, and J. Bischoff, “Congestion
pricing in a world of self-driving vehicles:
An analysis of different strategies in
alternative future scenarios,” Transportation
Research Part C: Emerging Technologies,
vol. 98, pp. 167–185, 2019. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S0968090X1830370X
24
[6] T. Li, F. Guo, R. Krishnan, A. Sivakumar, and
J. Polak, “Right-of-way reallocation for mixed
flow of autonomous vehicles and human
driven vehicles,” Transportation Research
Part C: Emerging Technologies, vol. 115, p.
102630, jun 2020.
[7] C. Laurell and S. Christian, “Social media
analytics as an enabler for external search
and open foresightx2014;the case of tes-
lax0027;s autopilot and regulatory scrutiny of
autonomous driving,” IEEE Transactions on
Engineering Management, vol. 69, no. 2, pp.
564–571, 2022.
[8] S. Karnouskos, “Self-driving car acceptance
and the role of ethics,” IEEE Transactions on
Engineering Management, vol. 67, no. 2, pp.
252–265, 2020.
[9] X. Zhao, Z. Wang, Z. Xu, Y. Wang, X. Li,
and X. Qu, “Field experiments on longitu-
dinal characteristics of human driver be-
havior following an autonomous vehicle,”
Transportation Research Part C: Emerging
Technologies, vol. 114, pp. 205–224, may
2020.
[10] M. Hagl and D. R. Kouabenan, “Safe
on the road – does advanced driver-
assistance systems use affect road risk
perception?” Transportation Research Part
F: Traffic Psychology and Behaviour,
vol. 73, pp. 488–498, 2020. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S1369847820304782
[11] S. YOO, K. Junya, K. Yuta, and M. Shunsuke,
“Achieving Inclusive Transportation: Fully
Automated Vehicles with Social Support,”
Research Institute of Economy, Trade and
Industry (RIETI), Discussion papers 22017,
Mar. 2022. [Online]. Available: https://ideas.
repec.org/p/eti/dpaper/22017.html
[12] M. Warschauer, The MIT Press. Technology
and Social Inclusion: Rethinking the Digital
Divide, 09 2004. [Online]. Available: https:
//doi.org/10.7551/mitpress/6699.001.0001
[13] Z. Han, X. Lu, E. I. Hörhager, and J. Yan,
“The effects of trust in government on earth-
quake survivors’ risk perception and pre-
paredness in china,” Natural Hazards, vol. 86,
p. 437–452, dec 2016.
[14] A. Joshi and M. Aoki, “The role of
social capital and public policy in disaster
recovery: A case study of tamil nadu state,
india,” International Journal of Disaster Risk
Reduction, vol. 7, pp. 100–108, 2014. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S2212420913000460
[15] A. Vij and J. L. Walker, “How, when and
why integrated choice and latent variable
models are latently useful,” Transportation
Research Part B: Methodological,
vol. 90, pp. 192–217, 2016. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S019126151630234X
[16] D. J. Fagnant and K. Kockelman, “Preparing a
nation for autonomous vehicles: opportuni-
ties, barriers and policy recommendations,”
Transportation Research Part A: Policy and
Practice, vol. 77, pp. 167–181, 2015. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S0965856415000804
[17] N. Adnan, S. Md Nordin, M. A. bin Bahrud-
din, and M. Ali, “How trust can drive forward
the user acceptance to the technology? in-
vehicle technology for autonomous vehicle,”
Transportation Research Part A: Policy and
Practice, vol. 118, pp. 819–836, 2018. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S0965856418309030
[18] J. Zhu and I. Tasic, “Safety analysis of free-
way on-ramp merging with the presence of
autonomous vehicles,” Accident Analysis &
Prevention, vol. 152, p. 105966, mar 2021.
[19] C. Mao, Y. Liu, and Z.-J. M. Shen, “Dispatch
of autonomous vehicles for taxi services:
A deep reinforcement learning approach,”
25
Transportation Research Part C: Emerging
Technologies, vol. 115, p. 102626, jun 2020.
[20] R. Krueger, T. H. Rashidi, and J. M. Rose,
“Preferences for shared autonomous ve-
hicles,” Transportation Research Part C:
Emerging Technologies, vol. 69, pp. 343–355,
aug 2016.
[21] M. A. Moore, P. S. Lavieri, F. F. Dias, and
C. R. Bhat, “On investigating the potential
effects of private autonomous vehicle use on
home/work relocations and commute times,”
Transportation Research Part C: Emerging
Technologies, vol. 110, pp. 166–185, jan 2020.
[22] K. Faber and D. van Lierop, “How
will older adults use automated vehicles?
assessing the role of avs in overcoming
perceived mobility barriers,” Transportation
Research Part A: Policy and Practice,
vol. 133, pp. 353–363, 2020. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S0965856419312091
[23] A. Nunes, B. Reimer, and J. F. Coughlin.,
“People must retain control of autonomous
vehicles,” Nature, vol. 556, pp. 169–171, 2018.
[24] S. Wang and J. Zhao, “Risk preference
and adoption of autonomous vehicles,”
Transportation Research Part A: Policy and
Practice, vol. 126, pp. 215–229, 2019. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S0965856417310315
[25] R. Shabanpour, N. Golshani,
A. Shamshiripour, and A. K. Mohammadian,
“Eliciting preferences for adoption of fully au-
tomated vehicles using best-worst analysis,”
Transportation Research Part C: Emerging
Technologies, vol. 93, pp. 463–478, 2018. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S0968090X18308957
[26] N. Huijts, E. Molin, and L. Steg,
“Psychological factors influencing sustain-
able energy technology acceptance: A
review-based comprehensive framework,”
Renewable and Sustainable Energy Reviews,
vol. 16, no. 1, pp. 525–531, 2012. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S136403211100428X
[27] P. Schweizer-Ries, “Energy sustainable
communities: Environmental psychological
investigations,” Energy Policy, vol. 36,
no. 11, pp. 4126–4135, 2008, transition
towards Sustainable Energy Systems. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S030142150800311X
[28] R. Wüstenhagen, M. Wolsink, and M. J.
Bürer, “Social acceptance of renewable
energy innovation: An introduction to
the concept,” Energy Policy, vol. 35,
no. 5, pp. 2683–2691, 2007. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S0301421506004824
[29] M.-C. Ho, D. Shaw, S. Lin, and Y.-
C. Chiu, “How do disaster characteristics
influence risk perception?” Risk Analysis,
vol. 28, no. 3, pp. 635–643, 2008. [Online].
Available: https://onlinelibrary.wiley.com/
doi/abs/10.1111/j.1539-6924.2008.01040.x
[30] M. M. Haddak, M. Lefèvre, and N. Havet,
“Willingness-to-pay for road safety improve-
ment,” Transportation Research Part A: Policy
and Practice, vol. 87, pp. 1–10, 2016. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S0965856416000197
[31] N. C. Bronfman, P. C. Cisternas, E. López-
Vázquez, and L. A. Cifuentes, “Trust and risk
perception of natural hazards: implications
for risk preparedness in chile,” Natural
Hazards, vol. 81, no. 1, pp. 307–327, Mar
2016. [Online]. Available: https://doi.org/10.
1007/s11069-015-2080-4
[32] T. Kato, “Sharing experience by residents
of past disaster sites: Effects of time
from disasters and types of experience,”
International Journal of Disaster Risk
Reduction, vol. 61, p. 102390, 2021. [On-
26
line]. Available: https://www.sciencedirect.
com/science/article/pii/S2212420921003514
[33] V. Mitchell, “Consumer perceived risk:
conceptualisations and models,” European
Journal of Marketing, vol. 33, no. 1/2,
pp. 163–195, Jan 1999. [Online]. Available:
https://doi.org/10.1108/03090569910249229
[34] C. Chen, H. Nelson, X. Xu, G. Bonilla, and
N. Jones, “Beyond technology adoption:
Examining home energy management
systems, energy burdens and climate change
perceptions during covid-19 pandemic,”
Renewable and Sustainable Energy Reviews,
vol. 145, p. 111066, 2021. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S1364032121003543
[35] P. Murray-Tuite, K. Wernstedt, and W. Yin,
“Behavioral shifts after a fatal rapid
transit accident: A multinomial logit
model,” Transportation Research Part
F: Traffic Psychology and Behaviour,
vol. 24, pp. 218–230, 2014. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S1369847814000527
[36] C. Măirean, “Fear and avoidance of
driving among drivers involved in a
road traffic crash. the role of traumatic
fear and driving cognitions,” Transportation
Research Part F: Traffic Psychology and
Behaviour, vol. 74, pp. 322–329, 2020. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S1369847820305143
[37] M. Aghababaei, S. B. Costello, and
P. Ranjitkar, “Transportation impact assess-
ment following a potential alpine fault
earthquake in new zealand,” Transportation
Research Part D: Transport and Environment,
vol. 87, p. 102511, 2020. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S1361920920306982
[38] X. Demeulenaere, “How challenges of
human reliability will hinder the de-
ployment of semi-autonomous vehicles,”
Technological Forecasting and Social Change,
vol. 157, p. 120093, 2020. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S0040162520309197
[39] T. Zhang, D. Tao, X. Qu, X. Zhang,
R. Lin, and W. Zhang, “The roles
of initial trust and perceived risk in
public’s acceptance of automated vehicles,”
Transportation Research Part C: Emerging
Technologies, vol. 98, pp. 207–220, 2019. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S0968090X18308398
[40] Z. Xu, K. Zhang, H. Min, Z. Wang,
X. Zhao, and P. Liu, “What drives people
to accept automated vehicles? findings
from a field experiment,” Transportation
Research Part C: Emerging Technologies,
vol. 95, pp. 320–334, 2018. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S0968090X18302316
[41] R. P. Monteiro, G. L. de Holanda Coelho,
P. H. Hanel, C. E. Pimentel, and V. V.
Gouveia, “Personality, dangerous driving,
and involvement in accidents: Testing a
contextual mediated model,” Transportation
Research Part F: Traffic Psychology and
Behaviour, vol. 58, pp. 106–114, 2018. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S1369847818300184
[42] D. Merecz, M. Waszkowska, and A. Wezyk,
“Psychological consequences of trauma
in mva perpetrators – relationship
between post-traumatic growth, ptsd
symptoms and individual characteristics,”
Transportation Research Part F: Traffic
Psychology and Behaviour, vol. 15,
no. 5, pp. 565–574, 2012. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S1369847812000526
[43] W.-T. Lai and C.-F. Chen, “Behavioral
intentions of public transit passengers—the
27
roles of service quality, perceived value,
satisfaction and involvement,” Transport
Policy, vol. 18, no. 2, pp. 318–325, 2011. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S0967070X10001125
[44] J. de Oña and R. de Oña, “Quality
of service in public transport based on
customer satisfaction surveys: A review and
assessment of methodological approaches,”
Transportation Science, vol. 49, no. 3,
pp. 605–622, 2015. [Online]. Available:
https://doi.org/10.1287/trsc.2014.0544
[45] P. N. E. Chee, Y. O. Susilo, and Y. D.
Wong, “Determinants of intention-to-use
first-/last-mile automated bus service,”
Transportation Research Part A: Policy and
Practice, vol. 139, pp. 350–375, 2020. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S0965856420306108
[46] C. Gkartzonikas and K. Gkritza, “What have
we learned? a review of stated preference
and choice studies on autonomous vehicles,”
Transportation Research Part C: Emerging
Technologies, vol. 98, pp. 323–337, jan 2019.
[47] C. Lonigan, M. Shannon, A. Finch,
T. Daugherty, and C. Taylor, “Children’s
reactions to a natural disaster: symptom
severity and degree of exposure,” Advances
in Behaviour Research and Therapy, vol. 13,
no. 3, pp. 135–154, 1991, impact of Natural
Disasters on children and families. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/014664029190002R
[48] C. S. North, A. Kawasaki, E. L. Spitznagel,
and B. A. Hong, “The course of ptsd,
major depression, substance abuse, and
somatization after a natural disaster,” The
Journal of Nervous and Mental Disease,
vol. 192, no. 12, 2004. [Online]. Available:
https://journals.lww.com/jonmd/Fulltext/
2004/12000/The_Course_of_PTSD,_Major_
Depression,_Substance.4.aspx
[49] M. Aklin, P. Bayer, S. Harish, and
J. Urpelainen, “Economics of household
technology adoption in developing countries:
Evidence from solar technology adop-
tion in rural india,” Energy Economics,
vol. 72, pp. 35–46, 2018. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S0140988318300690
[50] J. Axsen and K. S. Kurani, “Inter-
personal influence in the early plug-
in hybrid market: Observing social
interactions with an exploratory multi-
method approach,” Transportation Research
Part D: Transport and Environment,
vol. 16, no. 2, pp. 150–159, 2011. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S1361920910001525
[51] J. Jansson, T. Pettersson, A. Mannberg,
R. Brännlund, and U. Lindgren,
“Adoption of alternative fuel vehi-
cles: Influence from neighbors, family
and coworkers,” Transportation Research
Part D: Transport and Environment,
vol. 54, pp. 61–73, 2017. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S1361920915302534
[52] T. Liljamo, H. Liimatainen, and M. Pöllänen,
“Attitudes and concerns on automated
vehicles,” Transportation Research Part
F: Traffic Psychology and Behaviour,
vol. 59, pp. 24–44, 2018. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S1369847818303292
[53] S.-A. Kaye, I. Lewis, S. Forward, and
P. Delhomme, “A priori acceptance of
highly automated cars in australia,
france, and sweden: A theoretically-
informed investigation guided by the
tpb and utaut,” Accident Analysis &
Prevention, vol. 137, p. 105441, 2020. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S0001457519307390
28
[54] T. Zhang, D. Tao, X. Qu, X. Zhang, J. Zeng,
H. Zhu, and H. Zhu, “Automated vehicle ac-
ceptance in china: Social influence and initial
trust are key determinants,” Transportation
Research Part C: Emerging Technologies,
vol. 112, pp. 220–233, 2020. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S0968090X19312537
[55] F. Zhou, Z. Zheng, J. Whitehead,
S. Washington, R. K. Perrons, and L. Page,
“Preference heterogeneity in mode choice for
car-sharing and shared automated vehicles,”
Transportation Research Part A: Policy and
Practice, vol. 132, pp. 633–650, 2020. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S0965856419307402
[56] E. Molin, K. Adjenughwure, M. de
Bruyn, O. Cats, and P. Warffemius,
“Does conducting activities while traveling
reduce the value of time? evidence
from a within-subjects choice experiment,”
Transportation Research Part A: Policy and
Practice, vol. 132, pp. 18–29, 2020. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S0965856419307451
[57] Z. Wadud, D. MacKenzie, and P. Leiby, “Help
or hindrance? the travel, energy and carbon
impacts of highly automated vehicles,”
Transportation Research Part A: Policy and
Practice, vol. 86, pp. 1–18, 2016. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S0965856415002694
[58] J. Wu, H. Liao, J.-W. Wang, and T. Chen, “The
role of environmental concern in the public
acceptance of autonomous electric vehicles:
A survey from china,” Transportation
Research Part F: Traffic Psychology and
Behaviour, vol. 60, pp. 37–46, 2019. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S1369847818303425
[59] P. Kopelias, E. Demiridi, K. Vogiatzis, A. Sk-
abardonis, and V. Zafiropoulou, “Connected
and autonomous vehicles-environmental
impacts-a review,” Science of The Total
Environment, vol. 712, p. 135237, 2020. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S0048969719352295
[60] K. E. Train, “Recreation demand models
with taste differences over people,” Land
Economics, vol. 74, no. 2, pp. 230–239,
1998. [Online]. Available: http://www.jstor.
org/stable/3147053
[61] D. McFadden and K. Train, “Mixed mnl mod-
els for discrete response,” Journal of Applied
Econometrics, vol. 15, no. 5, pp. 447–470,
2000.
[62] D. Revelt and K. Train, “Mixed Logit with
Repeated Choices: Households’ Choices of
Appliance Efficiency Level,” The Review of
Economics and Statistics, vol. 80, no. 4,
pp. 647–657, 11 1998. [Online]. Available:
https://doi.org/10.1162/003465398557735
[63] M. Ben-Akiva, D. Mcfadden, M. Abe,
U. Böckenholt, D. Bolduc, D. Gopinath,
T. Morikawa, V. Ramaswamy, V. Rao,
D. Revelt, and D. Steinberg, “Modeling
methods for discrete choice analysis,”
Marketing Letters, vol. 8, no. 3, pp.
273–286, Jul 1997. [Online]. Available:
https://doi.org/10.1023/A:1007956429024
[64] M. E. Ben-Akiva, S. R. Lerman, S. R. Lerman
et al., Discrete choice analysis: theory and
application to travel demand. MIT press,
1985, vol. 9.
[65] M. Ben-Akiva, D. McFadden, K. Train,
J. Walker, C. Bhat, M. Bierlaire, D. Bolduc,
A. Boersch-Supan, D. Brownstone, D. S.
Bunch et al., “Hybrid choice models: Progress
and challenges,” Marketing Letters, vol. 13,
no. 3, pp. 163–175, 2002.
[66] M. V. Johansson, T. Heldt, and P. Johansson,
“The effects of attitudes and personality traits
on mode choice,” Transportation Research
Part A: Policy and Practice, vol. 40, no. 6, pp.
29
507–525, 2006.
[67] Á. Fernández-Heredia, S. Jara-Díaz, and
A. Monzón, “Modelling bicycle use inten-
tion: the role of perceptions,” Transportation,
vol. 43, no. 1, pp. 1–23, 2016.
[68] M. F. Yáñez, S. Raveau, and J. d. D. Or-
túzar, “Inclusion of latent variables in mixed
logit models: modelling and forecasting,”
Transportation Research Part A: Policy and
Practice, vol. 44, no. 9, pp. 744–753, 2010.
[69] H. Link, “Is car drivers’ response to conges-
tion charging schemes based on the correct
perception of price signals?” Transportation
Research Part A: Policy and Practice, vol. 71,
pp. 96–109, 2015.
[70] N. Makwana, “Disaster and its impact
on mental health: A narrative review,”
Journal of family medicine and primary care,
vol. 8, no. 10, pp. 3090–3095,
Oct 2019. [Online]. Available:
https://pubmed.ncbi.nlm.nih.gov/31742125
[71] S. Hardman, R. Berliner, and G. Tal, “Who will
be the early adopters of automated vehicles?
insights from a survey of electric vehicle
owners in the united states,” Transportation
Research Part D: Transport and Environment,
vol. 71, pp. 248–264, 2019, the roles of users in
low-carbon transport innovations: Electrified,
automated, and shared mobility. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S1361920918303043
[72] S. Yoo, K. W. Koh, and Y. Yoshida, “Are con-
sumers abandoning diesel automobiles be-
cause of contrasting diesel policies? evidence
from the korean automobile market,” Energy
Economics, vol. 92, p. 104969, 2020.
[73] R. Bennett, R. Vijaygopal, and R. Kottasz,
“Willingness of people with mental
health disabilities to travel in driver-
less vehicles,” Journal of Transport &
Health, vol. 12, pp. 1–12, 2019. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S2214140518304080
[74] R. W. Novaco, D. Stokols, J. Campbell,
and J. Stokols, “Transportation, stress, and
community psychology,” American Journal
of Community Psychology, vol. 7, no. 4,
pp. 361–380, Aug 1979. [Online]. Available:
https://doi.org/10.1007/BF00894380
[75] Z. Rezvani, J. Jansson, and J. Bodin,
“Advances in consumer electric vehicle
adoption research: A review and re-
search agenda,” Transportation Research
Part D: Transport and Environment,
vol. 34, pp. 122–136, 2015. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S1361920914001515
[76] T. A. Novack, D. Labbe, M. Grote, N. Carlson,
M. Sherer, J. C. Arango-Lasprilla, T. Bushnik,
D. Cifu, J. M. Powell, D. Ripley, and R. T.
Seel, “Return to driving within 5 years
of moderate–severe traumatic brain injury,”
Brain Injury, vol. 24, no. 3, pp. 464–471,
2010, pMID: 20184403. [Online]. Available:
https://doi.org/10.3109/02699051003601713
[77] S. Kenyon, G. Lyons, and J. Raf-
ferty, “Transport and social exclusion:
investigating the possibility of promot-
ing inclusion through virtual mobility,”
Journal of Transport Geography, vol. 10,
no. 3, pp. 207–219, 2002. [Online].
Available: https://www.sciencedirect.com/
science/article/pii/S0966692302000121
[78] M. Kamargianni, M. Ben-Akiva, and
A. Polydoropoulou, “Incorporating social
interaction into hybrid choice models,”
Transportation, vol. 41, no. 6, pp.
1263–1285, Nov 2014. [Online]. Available:
https://doi.org/10.1007/s11116-014-9550-5
[79] F. Manca, A. Sivakumar, N. Daina,
J. Axsen, and J. W. Polak, “Modelling
the influence of peers’ attitudes on
choice behaviour: Theory and empirical
application on electric vehicle preferences,”
30
Transportation Research Part A: Policy and
Practice, vol. 140, pp. 278–298, 2020. [On-
line]. Available: https://www.sciencedirect.
com/science/article/pii/S0965856420306984
A A: A T D
Table A1 presents the distribution of the
socioeconomic variables in our sample
and government statistics.
Table A3 shows the average score (when
we ask the respondents to respond on
1-5 point scales) and the proportion of
respondents’ evaluations of benefits
and concerns regarding FAVs (for multiple
choice questions). We calculate the pro-
portions as follows: the number of peo-
ple who choose the option/sample size
(N=100,803).
A B: R C
Although we use subsample analysis,
we conduct additional analyses by using
the interaction terms on the following
variables: social support, risk preferences,
accident and malfunction. The results are
shown in Tables A4 and A5. We find
that the estimation results do not change,
quantitatively and qualitatively, from our
main results, suggesting that our results
in Table IV and Vare robust.
31
TABLE A1: Socioeconomic Distribution of the Respondents and the Japanese Population
Sample (%) Government Statistics (%)
(n =100,803)
Gender Female 41 51.3
Male 59 48.7
Education level Junior high school or less 2.1 9.5
High school 26.9 42.3
Some college 22.6 15.6
Bachelor / Master / Doctor 45.9 23.9
Other 1.9 8.6
Age 18–19 0.2 2.3
20–29 5.4 11.7
30–39 18.1 13.3
40–49 31.9 17.2
50–64 25.8 22.1
Over 65 10.7 33.4
Household income <2 million JPY 7.8 18.3
2–3 million JPY 8.9 17.2
3–4 million JPY 11.9 15.3
4–5 million JPY 12.3 12.2
5–6 million JPY 11.9 9
6–7 million JPY 9.6 6.9
7–8 million JPY 9.1 5.8
8–9 million JPY 6.9 4.1
9–10 million JPY 6.7 3.4
10–15 million JPY 10.5 6
15–20 million JPY 2.7 1.1
≥20 million JPY 1.7 0.7
Don’t know / don’t want to answer - –
Region Hokkaido 4.6 4.2
Tohoku 5.9 6.9
Kanto 38.2 34.4
Chubu 16.6 16.8
Kinki 20.1 17.7
Chugoku 5.1 5.8
Shikoku 2.5 2.9
Kyushu/Okinawa 7.1 11.3
Household size 1 15.6 34.5
2 30.1 27.9
3 23.6 17.6
4 and above 30.1 20
Sources: MIC (2017, 2019a, 2019b)
TABLE A2: Cronbach’s alpha score of each latents
Latent Factors Cronbach’s Alpha
Social Support 0.66
Risk-Averse 0.33
Risk-Seeking 0.29
Accident 0.62
Malfunction 0.65
Burden Reduction 0.55
Safety 0.50
Recycle 0.83
GHG emissions 0.84
Air Pollution 0.87
Biodiversity 0.74
32
TABLE A3: Proportion and Mean Value of Respondents’ Evaluations for Latent Construct. The mean value of
the latent construct for the whole population would be zero, with a standard deviation of 1.
Without Disaster Disaster Victims
Mean Std.dev Mean Std.dev
Social Support -0.200 0.761 0.093 0.819
Risk-Averse 0.110 0.516 -0.520 0.517
Risk-Seeking -0.140 0.429 0.660 0.491
Accident -0.007 0.801 0.030 0.815
Malfunction -0.007 0.784 0.030 0.797
Burden Reduction -0.007 0.717 0.030 0.756
Safety -0.007 0.662 0.030 0.681
Recycle -0.017 0.954 0.080 0.899
GHG emissions -0.017 0.966 0.080 0.911
Air Pollution -0.018 0.948 0.080 0.903
Conservation -0.017 0.914 0.080 0.868
33
TABLE A4: Main Estimation Result with Interactions
Model (1) Model (2)
Before Accident After Accident
Coefficient Std. Err Coefficient Std. Err
Panel (A): Latents
Panel (A-1): Social Support
Social Support 0.100*** 0.013 0.109*** 0.013
Social Support (×Disaster Victims) 0.084*** 0.029 0.068** 0.030
Panel (A-2): Risk, Accident and Merits
Risk-Averse -0.221*** 0.023 -0.181*** 0.023
Risk-Averse (×Disaster Victims) 0.094* 0.053 0.068 0.055
Risk-Seeking 0.093*** 0.027 0.086*** 0.028
Risk-Seeking (×Disaster Victims) 0.161*** 0.058 0.134** 0.061
Accident 0.011 0.055 0.026 0.057
Accident (×Disaster Victims) -0.544*** 0.129 -0.308** 0.136
Malfunction 0.024 0.056 -0.008 0.058
Malfunction (×Disaster Victims) 0.475*** 0.132 0.235* 0.139
Safety -0.008 0.023 0.072*** 0.024
Burden Reduction 0.363*** 0.026 0.288*** 0.027
Panel (A-3): Environmental Perceptions
Recycle 0.018 0.019 0.009 0.020
GHG Emissions -0.061*** 0.016 -0.035** 0.017
Air Pollution -0.092*** 0.015 -0.072*** 0.016
Conservation 0.045** 0.018 0.041** 0.019
Panel (B): Socio-Economic Variables
ln Perceived Value 0.309*** 0.045 0.316*** 0.051
Female -0.259*** 0.024 -0.128*** 0.024
ln Age -0.420*** 0.042 -0.276*** 0.044
ln Income 0.146*** 0.014 0.104*** 0.014
Disaster Experience 0.192*** 0.024 0.185*** 0.025
Commuter 0.048** 0.022 0.038* 0.023
Car Ownership 0.217*** 0.024 -0.108*** 0.025
Accident Awareness -0.392*** 0.020 -0.284*** 0.020
License 0.588*** 0.036 0.232*** 0.036
Panel (C): Others
Heterogeneity Coefficient 0.295*** 0.071 0.457*** 0.095
Constant -0.578** 0.278 0.094 0.303
Log-Likelihood -40,467.62 -38,410.474
Note: Standard errors in parentheses. * p<0.1, ** p<0.05, *** p<0.01.
N=69,391
Prefecture Fixed Effects are Included.
34
TABLE A5: Substitution Patterns with Interactions
Coefficient Std. Err
Panel (A): Latents
Panel (A-1): Social Support and Risk
Social Support 0.049*** 0.012
Social Support (×Disaster Victims) -0.047* 0.028
Panel (A-2): Risk, Accident and Merits
Risk-Averse -0.014 0.022
Risk-Averse (×Disaster Victims) -0.070 0.052
Risk-Seeking -0.028 0.027
Risk-Seeking (×Disaster Victims) 0.099* 0.057
Accident -0.447*** 0.056
Accident (×Disaster Victims) -0.098 0.133
Malfunction -0.210*** 0.057
Malfunction (×Disaster Victims) 0.078 0.130
Safety 0.081*** 0.023
Burden Reduction -0.108*** 0.027
Panel (A-3): Environmental Perceptions
Waste Reduction 0.049*** 0.018
GHG Emissions 0.063*** 0.020
Air Pollution -0.008 0.178
Conservation -0.094*** 0.190
Panel (B): Socio-Economic Variables
ln Perceived Value Gap 0.014 0.026
Female 0.305*** 0.023
ln Age 0.134*** 0.041
ln Income -0.073*** 0.013
Disaster Experience 0.007 0.023
Commuter -0.090*** 0.021
Car Ownership -0.312*** 0.023
Accident Awareness 0.407*** 0.019
License -0.262*** 0.033
Panel (C): Others
Heterogeneity Construct 0.123*** 0.032
Constant -0.173 0.194
Log-Likelihood -40,809.132
Note: Standard errors in parentheses. * p<0.1, ** p<0.05, *** p<0.01.
N=69,391
Prefecture Fixed Effects are Included.