Users’ Privacy Concerns and Attitudes towards Usage-based Insurance:
An Empirical Approach
Juan Quintero
1 a
and Alexandr Railean
2 b
1
Department of Computer Science, Friedrich-Alexander-Universit
¨
at Erlangen-N
¨
urnberg (FAU), Erlangen, Germany
2
Institute of Computer Science, Georg-August-Universit
¨
at G
¨
ottingen, G
¨
ottingen, Germany
Keywords:
Usage-based Insurance, Pay as You Drive, Telematics Insurance, Acceptance, Privacy Concerns, Driving
Style, Transparency.
Abstract:
Usage-based Insurance (UBI) is a car insurance model in which the insurance payment calculations are based
on driving data such as speed, acceleration, braking, location, etc. Driving data are collected and analysed by
the insurer to provide feedback on driving performance, help drivers improve their skills, and possibly apply
a discount on their next renewal. So far, UBI research has been focused more on its architecture, benefits, or
acceptance, while the users’ perception of such forms of insurance and their privacy concerns received less
attention. To fill this gap, we conducted an online survey with 281 participants and analysed their responses
using qualitative and quantitative methods. We found that data collection and sharing are the main privacy
concerns. Furthermore, we identified potential discounts as the most important feature in favor of adopting
UBI, while data collection and unfair ratings are the main reasons to avoid or quit UBI.
1 INTRODUCTION
Traditional insurance models are based on subsidized
systems, where customers who do not claim insurance
benefits subsidize others who file such claims. Thus,
it is common to find car insurance programs in which
the fees paid by a driver who does not report any ac-
cident do not differ much from others who usually re-
quire the assistance of the insurer. The fees in these
programs are calculated based on historical data av-
eraged across customers grouped by criteria such as
age, gender, or marriage status (Soleymanian et al.,
2019). UBI programs represent a trend, in which
the fee is based on a customer’s personalized driving
style. UBI is also known as Telematics Insurance, Pay
As You Drive (PAYD), or Pay How You Drive (PHYD).
As Figure 1 depicts, driving data are collected through
a telematics device, such as a dongle, a black box, a
smartphone app, or an embedded system, and anal-
ysed by the insurer. The user then gets feedback on
driving performance and a possible discount on the re-
newal payment in case of getting a good driving score.
Soleymanian et al. (2019) found that the main
benefits of UBI are the incentives to improve one’s
a
https://orcid.org/0000-0002-8205-7072
b
https://orcid.org/0000-0002-7472-2108
driving style through feedback, and the potential dis-
count on their insurance fees. On the other hand, the
main disadvantages are the privacy concerns (Arvids-
son et al., 2011; Derikx et al., 2016; Soleymanian
et al., 2019) and discrimination (most programs are
designed for novice drivers or young people).
Insurer
Telematics
device
User
acceleration, braking
GPS location, speed
time of day
in case of getting a
good driving score
Figure 1: General Usage-Based Insurance model.
Although there is research about the technical as-
pects of UBI (Troncoso et al., 2010; Iqbal and Lim,
2006; H
¨
andel et al., 2013), its benefits (Soleymanian
et al., 2019; Derikx et al., 2016; Litman, 2007), its
user acceptance (Mayer, 2012; Tian et al., 2020), and
usability issues (Quintero et al., 2020), less attention
has been given to the users’ perception of data collec-
tion and sharing, financial benefits, driving feedback,
and driving style. We conducted this online survey
290
Quintero, J. and Railean, A.
Users’ Privacy Concerns and Attitudes towards Usage-based Insurance: An Empirical Approach.
DOI: 10.5220/0011044000003191
In Proceedings of the 8th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2022), pages 290-299
ISBN: 978-989-758-573-9; ISSN: 2184-495X
Copyright
c
 2022 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved

to know more about these topics, and identify privacy
concerns related to UBI. The following research ques-
tions (RQ) are addressed in this paper:
• RQ-1. What are the privacy concerns identified by
users with UBI?
• RQ-2. What are the most important features of
UBI that influence a person’s decision for or
against UBI?
The contributions of this study are as follows: (1)
we determine that participants are aware of and con-
cerned about the sharing and storage of their driving
data, (2) we identify the discount on the next renewal
in case of getting a good driving score as the most
important factor in favor of UBI and collection of my
driving data (GPS location, acceleration, etc.) and
driving rating is unfair as the most important factors
against deciding to use UBI. (3) Based on our find-
ings, we draw recommendations that can help insur-
ers increase the adoption of UBI. (4) We provide the
source code and data to facilitate the replication of our
analysis.
2 BACKGROUND AND
TERMINOLOGY
Based on prior exploratory research (Derikx et al.,
2016; Quintero et al., 2020), we identified several key
factors as potential candidates for influencing UBI
perception. In what follows, we define and describe
each factor, and formulate hypotheses about the rela-
tionships between these factors.
Intention to Use UBI (IU) is the degree to which a
person has formulated a conscious intention to be or
not to be covered by UBI (adapted from (Warshaw
and Davis, 1985, p. 214)).
Privacy Concerns (PC) follow a scale devised by
Dinev and Hart (2006), and relate to opportunistic be-
havior with respect to the personal information shared
with companies by the respondent in particular.
My Perceived Driving Style (MS) is the degree
to which drivers believe that they drive carefully
and cautiously, obeying traffic rules (adapted from
(Mayer, 2012)).
Others’ Perceived Driving Style (OS) is the percep-
tion of how carefully others drive and follow traffic
rules Quintero et al. (2020).
Discount in UBI (DI) is how much money a driver
can save by participating in UBI.
Driving Feedback (DF) relates to how often feed-
back is given to drivers based on their driving perfor-
mance.
We then formulate the following hypotheses based
on the aforementioned factors:
H1. People with low privacy concerns (PC
↓
) are more
willing to be covered by UBI (IU) than others with
high privacy concerns (PC
↑
).
H2. Magnitude of discount (DI) plays a greater role
than the driving style feedback (DF) in UBI percep-
tion by users with high privacy concerns (PC
↑
).
H3. People with high privacy concerns (PC
↑
) prefer
to avoid using smartphones as a telematics device.
H4. People with high privacy concerns (PC
↑
) pre-
fer to avoid sharing telemetry with entities other than
their insurer.
3 METHOD
In July 2021, we conducted an online survey with UBI
users to determine their: (1) privacy concerns with
UBI; (2) the most important UBI features that influ-
ence a person’s decision for or against UBI; (3) pref-
erences for sharing and storage of driving data. The
selection criteria for our study were (1) to have a driv-
ing license, (2) to have experience being covered by a
car insurance program. Our study got approval from
the data protection officer of our university. The sur-
vey was conducted in English.
3.1 Recruitment
To recruit participants and manage the study we de-
cided to use Prolific, an online platform for research
studies, and hosted the survey on a LimeSurvey server
in our university. Participants were screened asking
about their experience with car insurance and specif-
ically with UBI, as well as when they got their first
driving license. Based on the number of UBI pro-
grams in the market and the participants in Prolific
who met the selection criteria, only participants from
Germany, Ireland, United Kingdom, and the United
States were considered in this study. The screen-
ing survey took approximately 3 minutes to complete.
791 participants out of 807 met the criteria of having
a driving license and car insurance. In UBI, we de-
fined Current users (c) as people covered by a UBI
program. Former users ( f ) are people who had been
covered by UBI and for some reason are no longer
covered by it. Potential users (p) are people with a
driving license who never enrolled in a UBI program.
In this paper, we noted current, former, and potential
users with the subscripts c, f , p respectively. Thus,
we identified 30 current, 77 former, and 684 potential
users of UBI. We invited 30
c
, 77
f
, and 250
p
to take
Users’ Privacy Concerns and Attitudes towards Usage-based Insurance: An Empirical Approach
291

part in our study, from which 15
c
, 41
f
, and 225
p
par-
ticipated. The survey took approximately 10 minutes
to complete. We got a total of 281 completed sur-
veys. Participants were rewarded with 0.30 GBP for
the pre-screen survey and 1 GBP for the final survey.
3.2 Survey Structure
We developed the online survey questions based on
existing literature and brainstorming sessions. We
conducted a pilot with 5 researchers to identify con-
cerns related to wording, and the connection between
the research and survey questions. After including
their feedback, we conducted another pilot with 6 par-
ticipants (2
c
, 2
f
, 2
p
) to test completion time and the
survey flow. The final version of the survey has a
55s explanatory video and contains 68 questions, of
which 3 are used for checking the understanding of
UBI concepts from the video.
In what follows, we summarize the questionnaire,
while the full text and other materials are available at
zenodo.org/record/6114141:
Part 1: Explanatory video, including 3 quiz questions
to validate the understanding of UBI concepts.
Part 2: Questions of UBI coverage, name of insurer
and UBI program, and how long participants have
been covered by this insurance. For former users, we
asked about the reasons to quit.
Part 3: Questions related to privacy concerns (PC)
and driving style (MS, OS).
Part 4: UBI scenario depending on each user group
(
c
,
f
,
p
). We provided a scenario for each kind of par-
ticipant to contextualize them about using UBI before
starting the questionnaire. For p users we presented
a hypothetical situation in which their insurer is of-
fering a UBI program and they have the option to be
covered by UBI. For c and f users we suggested re-
plying to the survey based on their current and former
experience using UBI, respectively.
Part 5: Intention to use (IU) and preferences on UBI.
We formulated questions related to which information
participants are willing to share for evaluating their
driving score, which telematics device they would
prefer to use, with whom they prefer to share their
driving data, and their preferred frequency and com-
munication channel to get driving feedback. For c and
f users we asked about the discount on their insurance
programs.
Part 6: Demographics, such as age, gender, country
of residence, education level, occupation, and driving
experience (annual distance traveled).
3.3 Data Analysis
After collecting the survey data, we conducted a qual-
itative and quantitative analysis. We used an induc-
tive thematic analysis (Braun and Clarke, 2006) going
from codes to themes to analyse the comments pro-
vided by users in the open-ended questions. For quan-
titative analysis we used Pandas and SciPy, which are
open-source data science tools. The statistical tech-
niques we applied are covered in detail in Section 4.2.
4 RESULTS
160 participants identified as female, 119 as male, and
2 as diverse. The most represented age category was
between 29 and 33 years old. Participants ranged
in age from 21 to 77 years old. Regarding educa-
tion, 118 had a bachelor degree, 74 had completed
a postgraduate degree, 85 had completed professional
or vocational education, and 4 preferred not to dis-
close. Most participants were potential users (225
p
,
41
f
, 15
c
) of UBI. The geographic distribution is as
follows: United Kingdom (255), followed by United
Stated (17), Germany (7), and Ireland (2).
4.1 Qualitative Analysis
Considering that some parts of our survey gave par-
ticipants the option to offer additional comments
about their responses, we performed thematic anal-
ysis (Braun and Clarke, 2006) to reveal common
themes among the 78 collected entries of unstructured
text. Two researchers did so by coding the comments
independently, then comparing notes and refining the
codebook until consensus was reached. The results of
thematic analysis are shown in Table 1. In what fol-
lows, we share the highlights, indicating which par-
ticipant group referred to this idea.
Distracting feedback can cause a decline in user
satisfaction: “It was annoying that they kept sending
me driving alerts, I felt like I was being tracked and
watched too much” (P36
f
). In addition, it can lead
to accidents caused by distracted driving: “On some
local roads that are technically limited to 30, virtually
every motorist drives at 40. When you’re using UBI
you have to stick to 30, and this causes a lot of irrita-
tion to the drivers stuck behind you, and as a driver it
caused me a great deal of stress at times” (P36
f
).
Model deficiency relates to statements about issues
in scoring algorithms, because they are incomplete
and cannot model the complex driving conditions that
occur sometimes. For example, P22
f
stated they got a
lower score for driving in the dark, even though it was
VEHITS 2022 - 8th International Conference on Vehicle Technology and Intelligent Transport Systems
292

Table 1: Summary of thematic analysis. The references indicate how many participants mentioned a theme, they are coded
by participant type: potential, current and former are marked as P, C, and F respectively.
Theme Subtheme
References
Code
P C F
Barriers
Distracting feedback 4 1 Users feel uncomfortable using UBI
UBI not convenient for users 4 3 Financially ineffective
Model deficiency
2 Inaccurate data
1 1 Danger awareness
5 2 4 Scoring process is incomplete or unfair
2 UBI tailored for young, inexperienced drivers
Decision
making
Decision making
3 Difficult to decide
5 Privacy concerns
Adoption
Increasing UBI adoption
1 1 Car sharing
2 Ownership
2 Self-sufficiency
2 Data portability
2 Critical mass
winter, and there are only a few hours of daylight dur-
ing the season. P34
f
summarized that as “conditions
on the road are not taken into account”. In the words
of P42
f
“those systems are not designed for real world
driving ... only the hypothetical driving that is taught
in lessons that doesn’t exist in real life”.
Danger awareness describes situations in which
drivers are forced to make exceptions and act in
ways that would be considered dangerous otherwise:
“sometimes it is necessary to overtake and one has to
speed up fast to do so and it [the system] can’t take
this into account. Also I have had to brake quickly
to avoid an animal/unexpected unindicated moving
across me and got marked down” (P2
c
). “If I have
to brake because a dog runs out in front of me I
shouldn’t have that go against me” (P63
p
). We hence-
forth group this issue, along with model deficiency,
under the term “big picture” problem.
Financially ineffective is how some participants
see UBI, because in their experience it never deliv-
ered the promise of a lower premium. On the contrary,
“[it is] absolutely useless and made my insurance
more expensive by default” (P42
f
), or “I have never
encountered anyone who has had their premium re-
duced” (P22
f
). This skepticism could be rooted in the
belief that insurance companies only seek increased
profits: “insurance companies are in the business of
making money, not making driving safer” (P190
p
).
Inaccurate data leads to erroneous scoring. Some
of our participants encountered loss of a GPS sig-
nal or had their telematics devices stop working for
a while: “I doubt its accuracy at times. We were told
that sometimes the link goes [down] and it can disrupt
the reading” (P63
f
).
Car sharing and ownership references were also
mentioned when pointing out limitations of UBI.
When a car is shared by multiple people, it is not al-
ways possible to attribute a score to a specific person:
“If you share a car with another person, how does UBI
know who to rate or do they just average it out. A lot
of people do share after all” (P82
p
). This can also
occur in families where the car is owned by a parent,
but driven by a child: “needs to be a system divorced
from driver ownership” (P249
p
).
Critical mass must be achieved for UBI to be ef-
fective, according to some of our participants. The
rationale is related to context, like in the case of dan-
ger awareness: “Data does not relate your driving to
that of others and how it impacts you. If all cars are
covered then UBI may be more conclusive” (P229
p
).
It can be difficult to decide whether to join UBI:
“Having never tried it or having had any experience
of it, I don’t know which I would prefer” (P114
p
).
Data portability is another potential acceptance
barrier, as tech-savvy users like P184
p
want to avoid
vendor lock-in: “portability of that data across multi-
ple UBI suppliers could also be an issue”.
Privacy concerns where explicitly named by some
participants, for example: “feels very big brother to
me, so not a fan” (P112
p
), or “do not know if my data
will be shared or mis-used, very off putting” (P88
p
).
Young drivers are seen as the group to benefit most
by some participants, because UBI can help them im-
prove their driving skills: “this is for the younger per-
son starting out on their driving life” (P143
p
).
Self-sufficiency can reduce the potential for pri-
vacy abuse, e.g.,: “a unit that can work to generate
scores without communicating” (P254
p
).
4.2 Quantitative Analysis
To test the hypotheses and answer the research ques-
tions, we applied statistical analysis. We first cleaned
the data set by discarding responses from participants
that did not provide consent, did not have a driving li-
cense at the time of participation, or have never been
Users’ Privacy Concerns and Attitudes towards Usage-based Insurance: An Empirical Approach
293

covered by car insurance. We then computed the cor-
relations between the following factors: privacy con-
cerns (PC), intention to use (IU), one’s driving style
(MS), the driving style of others (OS), the discount
in UBI (DI), the frequency with which one receives
feedback about their driving (DF), and preferences
related to collecting (CD) and sharing (SD) driving
data. Spearman and Pearson correlation was used for
ordinal and continuous factors respectively.
We set the significance level to p <= 0.05 and use
the following thresholds to define the correlations as
weak <= 0.35, moderate <= 0.67, high <= 0.89,
very high <= 0.99, or perfect 1 (Taylor, 1990). Note
that correlations can be positive or negative, thus a
correlation of -0.67 is “negative moderate”.
4.2.1 Hypotheses Testing
Prior to running our questionnaire, we formulated
several hypotheses about the participants’ attitudes to-
wards UBI. In what follows, we list these hypotheses
and explain how they were tested. Note that due to
space constraints, we do not include the detailed cal-
culations of how the latent variables below were com-
puted. However, all of the calculations are available
in the supplementary materials (see Section 3.2).
We considered several latent variables, which we
computed via 5-point Likert-scale questions. The
variables are: Intention to use (marked as IU, mea-
sured through 3 questions), Privacy concern (PC, 5
questions), preferences of sharing driving data (SD,
6 questions), preference of telematics device for col-
lecting driving data (CD, 1 question).
We then conducted an exploratory factor anal-
ysis to determine how many factors are needed to
represent each latent variable. Thus, we used the
Kaiser-Meyer-Olkin (KMO) test with a threshold of
KMO >= 0.6 (Kaiser and Rice, 1974).
Next, we computed the Pearson or Spearman cor-
relation for continuous and ordinal variables respec-
tively. Note that we treat a latent variable as continu-
ous if it can be represented by a single factor that was
computed as the mean of ordinal variables.
We now present the results for each hypothesis.
H1: Following the methodology above, we find that
the Pearson correlation between PC and IU is corr =
−0.129, p = 0.046. This indicates a weak negative re-
lationship, which suggests that participants with low
privacy concerns are more willing to be covered by
UBI. Therefore, our findings support H1.
H2: We considered the Discount, Driving feedback,
and Privacy concerns only for current and former
users, because only they have experience getting dis-
counts and driving feedback in UBI programs. Given
that these are ordinal variables, we calculated the
Spearman correlation between PC and DI (corr =
0.317, p = 0.044), and PC and DF (corr = 0.141, p =
0.379). We found a weak positive correlation between
PC and DI, while the correlation between PC and DF
is not significant. That means, people with high PC
expect to get a high discount in their UBI program.
Therefore, H2 is supported.
H3: We analyzed the users’ preference to collect their
driving data using a black box, dongle, embedded sys-
tem, or smartphone app as telematics device, as well
as the option to not collect any driving data at all. Fol-
lowing our methodology, we calculated the Spearman
correlation, resulting in the coefficients between vari-
ables depicted in Table 2
Table 2: Preference of collecting driving data using telemat-
ics device. Corr is the correlation between privacy concerns
(PC) and the preference of collecting driving data using a
particular telematics device. Note that entries in gray are
not statistically significant.
Data collection preference corr p
Do not collect data 0.283 1.407e-06
Embedded system 0.022 0.713
Dongle -0.025 0.682
Black box -0.030 0.611
Smartphone app -0.164 0.006
We found a weak positive correlation between PC
and the preference to avoid data collection, as well as
a weak negative correlation between PC and the usage
of a smartphone app as a telematics device. There-
fore, H3 is supported. Note that our results do not
show whether other types of telematics devices are
preferred, as other correlations were not significant.
Moreover, the data shows clearly that avoiding data
collection is preferred, if the option is available.
H4: We considered the users’ preference to share
their driving data with academia, a government
agency, a marketing agency, their insurer, other in-
surer, and traffic authorities. We computed the Pear-
son correlation between PC and the preference to
share driving data with the entities above, getting the
coefficients depicted in Table 3.
Table 3: Preference of sharing driving data. Corr is the
correlation between privacy concerns and the preference of
sharing driving data with a given entity.
Entity to share data with corr p
Road traffic authorities -0.197 0.001
My insurer -0.248 2.819e-05
Academic researchers -0.253 1.934e-05
Marketing companies -0.299 3.593e-07
Government agencies -0.305 2.116e-07
Other insurers -0.311 1.132e-07
We found a weak negative correlation between PC
and the willingness to share driving data with all en-
VEHITS 2022 - 8th International Conference on Vehicle Technology and Intelligent Transport Systems
294

tities above. Note that the strength of the correlation
varies, e.g., participants with high privacy concerns
are less reluctant to share data with road traffic au-
thorities, and more so when it comes to sharing with
other insurers or government agencies. For that rea-
son, H4 is not supported, because in retrospect it
is clear that the hypothesis was too optimistic. The
results show that participants with high privacy con-
cerns would rather not share data with anyone, even
their insurer.
4.2.2 Analysis of Research Questions
In this section we present our analysis for each re-
search question.
RQ-1: We considered the PC and questions related
to the preference of storage and sharing of driving
data, as well as the preference of telematics device
type. We calculated the Pearson correlation between
PC and each storage option for driving data, the re-
sults are shown in Table 4.
Table 4: Preference of driving data storage. Corr is the
correlation between privacy concerns and the preference of
storing driving data in a specific device or place. Entries in
gray are not statistically significant.
Data storage preference corr p
Dev. installed by me -0.026 0.663
Dev. installed by certified staff -0.092 0.122
Insurer -0.155 0.009
Other insurer -0.176 0.003
My phone -0.217 0.0002
The negative correlation shows that participants
with high PC are reluctant to accept the collection of
driving data regardless of storage method.
We also found that the Pearson correlation be-
tween PC and the option to participate in UBI without
storing driving data is corr = 0.215, p = 0.0003, thus
confirming the previous finding.
We also asked our participants about their pref-
erence to share driving data (e.g., speed, location,
mileage, dashcam footage, etc., see materials refer-
enced in Section 3.2 for a complete list). We found
that participants are more open to sharing their speed,
mileage, and braking behavior. Participants with high
privacy concerns are not willing to share their loca-
tion and dashcam recordings. More participants with
high privacy concerns prefer not to share their driv-
ing data (14.95%) than those with low (1.42%) and
neutral (3.56%) privacy concerns.
In addition, we computed the Pearson correlation
between PC and the preference of sharing driving data
with various entities. We found a weak negative cor-
relation between PC and all previously mentioned en-
tities (see Section 4.2.1), which suggests that partic-
ipants with low privacy concerns are more willing to
share their driving data.
We also asked our participants about their prefer-
ence to use UBI with a black box, dongle, embedded
system, or a smartphone app. We found that more
participants with high privacy concerns like to get the
benefits of UBI without any data collection.
If data collection is mandatory, participants with
low and neutral privacy concerns would prefer to use
a smartphone app as telematics device.
We conclude that participants identified as privacy
concerns the sharing and storage of their driving data,
especially when it comes to their position and dash-
cam recordings.
RQ-2: We asked participants about the most impor-
tant features of UBI in favor and against UBI. Based
on previous research and forums of UBI, we collected
features of UBI, elaborating two lists, one with the
arguments in favor (e.g., potential discount, driving
feedback, etc.), and another with arguments against
UBI (e.g., unfair ratings, age constraints, privacy con-
cerns, etc.).
We found that a discount on the next renewal in
case of getting a good driving score is the most im-
portant factor that influences a person’s decision in
favor of UBI. On the other hand, collection of my
driving data (GPS location, acceleration, etc.) and
driving rating is unfair are the most important factors
against deciding to use UBI.
Other Findings: In addition, we found a weak
positive correlation between PC and DI (corr =
0.334, p = 0.012), which suggests that participants
with high privacy concerns expect a high discount.
For driving feedback, we found that most participants
do not want to get any feedback. However, if it were
mandatory, they would like to receive it through email
or a smartphone app.
We found a weak positive correlation between the
MS-IU (corr = 0.181, p = 0.002), which suggests
that people who consider themselves good drivers are
willing to use UBI. We also found that only 8 partici-
pants out of 281 consider themselves bad drivers.
We found that participants aged between 29 and
38 have the highest privacy concerns. Also, partici-
pants under 33 are more willing to be covered by UBI.
Participants with a Master or Bachelor’s degree have
the highest privacy concerns in our sample. In terms
of occupations - employees, workers, and civil ser-
vants have the highest privacy concerns.
We did not find significant correlations between
the intention to use and the participants’ gender, the
mileage driven, the discount in UBI, the residence
country of our participants, or OS.
Users’ Privacy Concerns and Attitudes towards Usage-based Insurance: An Empirical Approach
295

5 DISCUSSION
Our findings indicate that participants with high PC
prefer not to use smartphone apps as telematics de-
vices. This could be because participants are in-
formed about potential vulnerabilities and tracking is-
sues associated with smartphones, therefore they do
not regard them as secure or private. However, this
question remains to be explored in future research.
The correlation between privacy concerns and the
expectation of a higher discount could be caused by
the fact that privacy conscious users want a fair com-
pensation for the data they share. In contrast, users
with low privacy awareness may not fully realize the
value of what they give up, hence they accept less fa-
vorable terms and have lower expectations.
Overall, our participants referred to the sharing
and collection of data as the main factors that make
them hesitate to join to UBI. The results show that
participants would be more open towards UBI if they
could take advantage of it without data collection and
sharing. It is thus possible, that a self-sufficient sys-
tem that relies on the resources of the car itself, and
does not involve third parties, could have a higher
acceptance. To the best of our knowledge, no such
solutions are available to customers at the moment.
Some researchers proposed systems that collect and
share aggregated data (H
¨
andel et al., 2013; Iqbal and
Lim, 2006; Troncoso et al., 2010), however they are
susceptible to the “big picture” problem discussed in
Section 4.1. Therefore, the design of an accurate
and auditable self-sufficient solution remains an open
question. We hypothesize that this can be achieved
by re-purposing self-driving car technology. Accord-
ing to the requirements defined by the Society of Au-
tomotive Engineers (SAE) in (SAE J3016, 2018), a
level-2 autonomous vehicle must be able to control
steering, braking and acceleration independently (i.e.,
without receiving instructions from a remote server).
Therefore, it must be aware of the environment, which
includes other cars, road markings, weather and road
conditions, etc. A re-purposed system could leverage
the same technology, but with the goal of evaluating
the driving style like an examination officer would,
rather than with the goal of driving the car. Such an
“examiner AI” could not only alleviate privacy con-
cerns, but also address the “big picture” problem.
This would be of a great benefit, because our re-
sults show that a common limitation perceived by par-
ticipants is the inability of UBI to understand the big
picture of the road conditions, thus leading to inac-
curate scores. Moreover, some users consider the
scores to be unfair, this can happen if they deviate
from the rules to prevent an accident. Such maneuvers
are penalized by the algorithm, because it is unaware
of what other agents (e.g, drivers, vehicles, pedestri-
ans, animals) were doing in the given circumstances.
Some of our participants express skepticism regarding
UBI, stating that it will be inefficient unless a critical
mass is achieved: either all vehicles must participate
in such insurance, or the algorithms must be sophis-
ticated enough to consider all relevant external fac-
tors. Since this is not feasible at the moment, a so-
lution would be to make the algorithms more tolerant
to such outlier behaviour. For example, they could
penalize drivers only if they consistently deviate from
the rules. Another possibility would be to provide an
option to dispute scores. However, this feature must
be implemented by taking into account the burden of
proof. A driver may not always have a video record-
ing to prove their innocence (e.g., in some countries
such cameras may be illegal), so it would be unfair to
penalize them if they failed to provide evidence dur-
ing the dispute. We believe that insurers are in a bet-
ter position to collect such evidence (e.g., by contact-
ing traffic safety authorities, retrieving public camera
footage, etc.), and thus the presumption of innocence
principle must be guaranteed.
The feedback collected from our participants in-
dicates that insurers are yet to make a compelling ar-
gument about why their service is unique and worth
the investment. This could make some potential users
hesitate to sign up, because UBI requires too much
commitment up-front (e.g., acquiring new hardware,
possibly modifying something in the car, etc.), with-
out the certainty that good driving will lead to a dis-
count. This is especially relevant when potential users
learn from others through word of mouth that they
have never met anyone who actually got a discount.
Therefore, we posit that there could be an entry bar-
rier that insurance companies have to bring down,
e.g., by offering the hardware for free or by leveraging
the hardware that modern cars are equipped with from
the factory, or by improving transparency and making
it easy to make accurate estimations of the discount
amount. Such transparent estimation tools could as-
sist potential users in making informed decisions.
Disagreement with the calculated driving scores
is a major concern raised by the participants. Al-
though none of them has explicitly stated they want
more transparency in this calculation, we believe that
the lack of such information could make users less
likely to perceive the scoring results as objective. Our
data show that most participants consider themselves
good drivers, so they could attribute errors to the algo-
rithm rather than to their own behaviour, thus reduc-
ing the level of satisfaction with the service. There-
fore insurers might remediate this by increasing the
VEHITS 2022 - 8th International Conference on Vehicle Technology and Intelligent Transport Systems
296

transparency of all processes, including data collec-
tion, data sharing and driving score calculation.
Data portability is another concern mentioned by
the participants. While Art. 20 of the General Data
Protection Regulation (GDPR) guarantees the right to
obtain one’s data in a “structured, commonly used and
machine-readable format and [..] to transmit those
data to another controller without hindrance”, cus-
tomers in some parts of the world do not necessarily
enjoy the same level of protection. It is therefore pos-
sible that adoption of UBI could be lower in regions
where such protections are not available.
Regarding the most important feature in favor of
UBI, the brochures of some UBI companies highlight
improving driving style and paying based on one’s
own driving style as key advantages. However, our
participants prioritized the discount on the next re-
newal, which can indicate that some users are focused
on saving money, rather than on improving their driv-
ing style. This raises a question: what is the users’
rationale for joining UBI? If the motivation is sav-
ing money, as participants remarked, this is included
in the traditional car insurance, where drivers pay
based on their historical claims. That means, even
though in traditional car insurance novice drivers have
a high insurance premium, they could pay less in the
next years if they report no incidents to the insurer.
Thus, good drivers can pay less even in traditional
insurance, hence UBI should attract users with argu-
ments other than reduced premiums. It is also possi-
ble that one’s cultural background determines the ex-
tent to which the calculated discount influences adop-
tion, however our sample is not diverse enough geo-
graphically to verify this hypothesis.
Regarding telematics devices, participants who do
not have high privacy concerns prefer to use a smart-
phone app as a telematics device if data collection
is unavoidable. This is despite usability issues that
Quintero et al. (2020) identified earlier, e.g., high bat-
tery consumption or inaccurate GPS location. This
suggests that users find smartphones appealing due to
some benefits that outweigh these issues. For exam-
ple, it could be due to (1) familiarity - they under-
stand the user interface and there is no need to learn
anything new, (2) privacy - they can turn it off when
they want (e.g., when driving to certain addresses),
(3) convenience - the smartphone is already used for
maps and navigating, (4) cost savings - there is no
need to purchase additional hardware, nor spend time
setting it up, (5) perceived trust - users are used to
storing sensitive data on their smartphones (e.g., pho-
tographs, correspondence), therefore they could also
entrust them with their driving data. However, the us-
age of smartphones can lead to security issues that
non-expert users are unaware of. Specifically, if a
device runs outdated software (e.g., no updates are
released by the manufacturer), then it could be an
easier target for attackers. Considering that only a
small portion of smartphones are running the latest
software (Quintero et al., 2020), insurers should care-
fully consider the implications of using smartphones
as telematics devices. This is especially important be-
cause driving data logs can be remotely manipulated
on a compromised smartphone, which could lead to
a higher rate of disputes. Moreover, malicious cus-
tomers could manipulate the logs themselves (e.g.,
delete data for the period when they drove aggres-
sively) in order to improve their score.
5.1 Recommendations
Based on the qualitative analysis of the feedback from
our participants (see Sec. 4.1), we recommend insur-
ers to work on improvements related to:
Car Sharing. If several family members use the same
car, or if drivers change during a road trip, UBI should
provide a way to indicate who is currently driving
the vehicle. The data from our participants shows
that such use cases are currently not handled well by
their insurers. Note that the user experience must take
into consideration the different types of telematics de-
vices. While it is possible to link several smartphones
to a car, or add a “user profiles” feature to a program
that runs on a single smartphone, this is not an option
when an embedded device is used for logging data,
unless each person can somehow “check in” before
they start driving.
Self-contained Systems that can evaluate the driving
style without sharing data with third parties or send-
ing it over a network would be a major step forward
in addressing users’ privacy concerns. Such systems
would also address the “big picture” issue, because
they provide fair scores even if not all vehicles on
the road participate in UBI. This also applies to cases
when no other vehicles are involved, e.g., a sudden
braking maneuver to avoid a collision with an animal
that crosses a street in a remote location. In such
circumstances there is no nearby infrastructure that
could provide recordings of the incident, nor are there
other cars that could have caught it on their dashcam.
A “discard trip” feature could be an alternative stop-
gap measure until self-contained systems are avail-
able. A driver could remove a trip from their history
if it involved an incident like the previously described
maneuver to avoid an animal on the road. Insurance
providers should also consider how this functionality
could be abused by drivers who use it to hide their vi-
olations. This can be addressed in different ways, e.g.,
Users’ Privacy Concerns and Attitudes towards Usage-based Insurance: An Empirical Approach
297

to limit the number of trips that can be discarded per
month, or to incur a progressively larger penalty for
each discarded trip, such that the accumulated penalty
would exceed the penalty of an incident one attempts
to conceal.
Participants’ Culture should be taken into account
at the moment of implementing UBI in a specific
country. Our findings indicate that most participants
are focused on saving money, which could be influ-
enced by the cost of living in the country of residence,
which might also influence acceptance of UBI.
Transparency. Providing users information about the
different stages in UBI (e.g., data collection, storage,
processing, etc.) could increase the transparency of
driving data handling and score calculation (Quintero
et al., 2020). Following the participants’ preference
about driving feedback, the insurers could provide
this information by Email or smartphone app. UBI
should provide clear information about how driving
data are transformed into one’s driving score.
5.2 Limitations
Our sample has a limited geographical distribution,
as 91% of the participants come from the UK and 6%
are from the USA, whereas the rest of the world adds
up to less than 3%. We are therefore unable to ob-
serve variations in attitudes towards UBI that might
be rooted in culture, or the way in which transport
infrastructure is managed in specific countries. Con-
sidering that road design influences driving behaviour
and leads to significant differences between countries
in terms of traffic safety and the severity of accidents
(Buehler and Pucher, 2017), we have reasons to be-
lieve that similar effects could apply to UBI and are
worthy of examining in future work.
In addition, we had few participants experienced
in UBI, most of them being potential users, rather
than current or former ones. This is because the rela-
tive novelty of UBI makes it difficult to recruit partic-
ipants with past experience.
6 RELATED WORKS
Aspects of UBI, such as acceptance, privacy, trans-
parency and disadvantages have been discussed by
researchers. In what follows, we summarize the find-
ings of related literature.
After comparing several distance-based insurance
programs, Litman (2007) concludes that these pro-
grams can be more beneficial for drivers and society
than for insurers. He argues that drivers could get
a lower premium by reducing the distance traveled,
thus benefiting society (e.g., less pollution, fewer ac-
cidents, etc.).
However, others did not find significant effects on
the mileage driven after using UBI for six months (So-
leymanian et al., 2019). They also found that drivers
improved their driving style during the observed pe-
riod, but it is not clear whether this is a long-lasting
effect.
Quintero et al. (2020) analyzed discussions in on-
line communities of current and former users of UBI,
and interviewed potential users. They found that
drivers may identify their mistakes based on feed-
back, and thus drive more cautiously, hence reduce
the frequency of accidents. In addition, insurers may
have a more accurate risk estimation by analysing col-
lected driving data.
The disadvantages of UBI discussed in literature
are usually related to privacy concerns, discrimina-
tion, dangerous driving, lack of transparency (e.g.,
data usage, unclear evaluation criteria), and the in-
vestment costs for the insurer (Arvidsson et al., 2011;
Derikx et al., 2016; Soleymanian et al., 2019; Quin-
tero et al., 2020). For example, drivers could pro-
voke dangerous situations while attempting to guess
how the scoring works, due to lack of algorithmic
transparency (Quintero et al., 2020). Discrimination
can occur because most UBI programs to date target
young drivers or people with little driving experience.
Privacy concerns in UBI are related to the collec-
tion of drivers’ location via GPS (Soleymanian et al.,
2019; Quintero et al., 2020). Although some UBI
programs do not use location to determine the driv-
ing score, several authors argued that privacy could
be compromised by inferring locations through com-
binations of factors like: distance traveled, speed and
time of driving, speed limits, start location or previ-
ous destinations (Dewri et al., 2013; Gao et al., 2014;
Wahlstr
¨
om et al., 2016; Bellatti et al., 2017).
Some academic solutions have been designed to
mitigate these privacy concerns, though at the time
of this writing they are not yet implemented in prac-
tice. In particular, they attempt to solve the problem
by performing all the calculations locally, and only
sending aggregated data to the insurer (H
¨
andel et al.,
2013; Iqbal and Lim, 2006; Troncoso et al., 2010).
Others examined the problem of UBI acceptance.
Mayer finds that privacy concerns do not play a
role, whereas the expected discount and hedonic
motivation on driving are important acceptance fac-
tors (Mayer, 2012). Derikx et al. (2016) find that
small financial rewards could motivate customers to
share their driving data. Tian et al. (2020) conclude
that perceived enjoyment and trust play an important
role for some age groups.
VEHITS 2022 - 8th International Conference on Vehicle Technology and Intelligent Transport Systems
298

7 CONCLUSIONS
We conducted an online survey with current, former,
and potential users of UBI, analysing their responses
through a qualitative and quantitative analysis.
The results indicate that privacy concerns arise if
driving data are stored and shared. Given the choice,
our participants would rather use self-sufficient UBI
implementations that perform all the analysis locally
and avoid sending data to the insurer or other parties.
We find that participants prioritize saving money
over improving their driving style, when it comes to
one’s perception of UBI utility. It should be noted that
no participants reported any actual savings they have
made via UBI, nor do they know anyone who has.
The issue is exacerbated by the lack of transparency of
the scoring algorithms, which made some participants
conclude that UBI is only meant to benefit insurers.
Therefore, we consider that improving transparency
should be a top priority, otherwise a growing share of
users might be disappointed, thus reducing adoption.
Based on the results, we propose several recom-
mendations for insurers, aimed at increasing UBI ac-
ceptance (see Section 5.1).
ACKNOWLEDGEMENTS
We thank Zinaida Benenson and Freya Gassmann
for their valuable comments on earlier versions of
this paper. This research has received funding from
the H2020 Marie Skłodowska-Curie EU project “Pri-
vacy&Us” under the grant agreement No 675730.
REFERENCES
Arvidsson, S. et al. (2011). Reducing asymmetric informa-
tion with usage-based automobile insurance. Swedish
National Road & Transport Research Institute (VTI),
(2010):2.
Bellatti, J., Brunner, A., Lewis, J., Annadata, P., Eltarjaman,
W., Dewri, R., and Thurimella, R. (2017). Driving
habits data: Location privacy implications and solu-
tions. IEEE Security & Privacy, 15(01):12–20.
Braun, V. and Clarke, V. (2006). Using thematic analysis in
psychology. 3(2):77–101.
Buehler, R. and Pucher, J. (2017). Trends in walking and cy-
cling safety: Recent evidence from high-income coun-
tries, with a focus on the united states and germany.
107(2):281–287.
Derikx, S., De Reuver, M., and Kroesen, M. (2016). Can
privacy concerns for insurance of connected cars be
compensated? Electronic markets, 26(1):73–81.
Dewri, R., Annadata, P., Eltarjaman, W., and Thurimella, R.
(2013). Inferring trip destinations from driving habits
data. In Proceedings of the 12th ACM workshop on
Workshop on privacy in the electronic society, pages
267–272.
Dinev, T. and Hart, P. (2006). An extended privacy calcu-
lus model for e-commerce transactions. Information
systems research, 17(1):61–80.
Gao, X., Firner, B., Sugrim, S., Kaiser-Pendergrast, V.,
Yang, Y., and Lindqvist, J. (2014). Elastic pathing:
Your speed is enough to track you. In Proceedings of
the 2014 ACM International Joint Conference on Per-
vasive and Ubiquitous Computing, pages 975–986.
H
¨
andel, P., Ohlsson, J., Ohlsson, M., Skog, I., and Nygren,
E. (2013). Smartphone-based measurement systems
for road vehicle traffic monitoring and usage-based in-
surance. IEEE systems journal, 8(4):1238–1248.
Iqbal, M. U. and Lim, S. (2006). A privacy preserving gps-
based pay-as-you-drive insurance scheme. In Sympo-
sium on GPS/GNSS (IGNSS2006), pages 17–21.
Kaiser, H. F. and Rice, J. (1974). Little jiffy, mark iv. Edu-
cational and psychological measurement, 34(1):111–
117.
Litman, T. (2007). Distance-based vehicle insurance feasi-
bility, costs and benefits. Victoria, 11.
Mayer, P. (2012). Empirical Investigations on User Percep-
tion and the Effectiveness of Persuasive Technologies.
PhD thesis, University of St. Gallen.
Quintero, J., Railean, A., and Benenson, Z. (2020). Accep-
tance factors of car insurance innovations: The case of
usage-based insurance. Journal of Traffic and Logis-
tics Engineering Vol, 8(2).
SAE J3016 (2018). Taxonomy and Definitions for Terms
Related to Driving Automation Systems for On-Road
Motor Vehicles. Standard, Society of Automotive En-
gineers (SAE).
Soleymanian, M., Weinberg, C. B., and Zhu, T. (2019). Sen-
sor data and behavioral tracking: Does usage-based
auto insurance benefit drivers? Marketing Science,
38(1):21–43.
Taylor, R. (1990). Interpretation of the correlation coeffi-
cient: a basic review. Journal of diagnostic medical
sonography, 6(1):35–39.
Tian, X., Prybutok, V. R., Mirzaei, F. H., and Dinulescu,
C. C. (2020). Millennials acceptance of insurance
telematics: An integrative empirical study. Tian, Xi-
aoguang, pages 156–181.
Troncoso, C., Danezis, G., Kosta, E., Balasch, J., and Pre-
neel, B. (2010). Pripayd: Privacy-friendly pay-as-
you-drive insurance. IEEE Transactions on Depend-
able and Secure Computing, 8(5):742–755.
Wahlstr
¨
om, J., Skog, I., Rodrigues, J. G., H
¨
andel, P., and
Aguiar, A. (2016). Map-aided dead-reckoning—a
study on locational privacy in insurance telematics.
arXiv preprint arXiv:1611.07910.
Warshaw, P. R. and Davis, F. D. (1985). Disentangling be-
havioral intention and behavioral expectation. Journal
of experimental social psychology, 21(3):213–228.
Users’ Privacy Concerns and Attitudes towards Usage-based Insurance: An Empirical Approach
299