UI Impact on Performance & Decisions CHI 2015, Crossings, Seoul, Korea

Displayed Uncertainty Improves Driving Experience and

Behavior: The Case of Range Anxiety in an Electric Car
Malte F. Jung David Sirkin Turgut M. Gür
Cornell University Stanford University Stanford University
Ithaca, New York, USA Stanford, California, USA Stanford, California, USA
mjung@cornell.edu sirkin@stanford.edu turgut@stanford.edu

Martin Steinert
Norwegian University of Science and Technology
Trondheim, Norway
martin.steinert@ntnu.no
ABSTRACT your washing machine starring at the “one minute left”
We explore the impact of the displayed precision of display only to find out that the washing machine’s one
instrumentation estimates of range and state-of-charge on minute was closer to 15? Have you ever been anxious about
drivers’ attitudes towards an all-electric vehicle (EV), on not having enough fuel in your car as you go up a hill and
their driving experience, and driving behavior under you suddenly notice that those “100 miles until your fuel
varying conditions of resource availability. Participants tank is empty” of the range display shrank to 20 miles?
(N=73) completed a 19-mile long drive through highway,
rural town and mountain road conditions in an EV that We encounter displays of estimates almost every time we
displayed high vs. low remaining range, and gave estimates interact with technical systems, and we hardly ever question
of that range with high and low information ambiguity. We the veracity of the displayed information, yet the outcomes
found that an ambiguous display of range preserved drivers’ of our decisions and actions are often highly dependent
feelings of trust towards the vehicle, despite encountering upon our understanding the system’s actual state.
situations intended to induce severe range anxiety. Unbeknownst to the ordinary observer, these displays may
Furthermore, compared to drivers facing an unambiguous appear highly precise, while in fact, they are merely
display of range, drivers presented with an ambiguous estimates, or predictions, based on assumptions that the user
range display reported improved driving experience, and can neither observe nor verify.
exhibited driving behavior better adapted to road and For example, remaining battery life is often displayed at a
remaining range conditions. level of accuracy (e.g. minutes for mobile devices, and
single digit miles for automobiles) that is at least one order
Author Keywords of magnitude higher than it should be. As a consequence,
Range anxiety; information display; progress bar; trust in many systems create user expectations that cannot be met.
automated systems; car; electric vehicle; ambiguity. This becomes particularly important when resources such
ACM Classification Keywords as power for transportation, navigation or life support are
H.1.2; H.5.2; J.4; J.7 rapidly depleting, or when there is utmost urgency to take
drastic corrective action. Therefore the implications of how
INTRODUCTION uncertainty is made salient to a user go far beyond the
Have you ever been startled by your laptop’s warning that examples at the beginning of the paper. A case in point is
“you only have a few minutes of battery power left” even the example of the Royal Majesty, a cruise ship that ran
though only moments earlier it had indicated closer to an ashore miles off course. When the GPS system aboard
hour of resources? Maybe you have also waited in front of failed, the navigation system switched to dead reckoning
mode and displayed an estimate to the crew without making
Permission to make digital or hard copies of all or part of this work for personal this switch salient. Deviating from its programmed course
or classroom use is granted without fee provided that copies are not made or
distributed for profit or commercial advantage and that copies bear this notice
due to winds and currents, the system indicated a precise—
and the full citation on the first page. Copyrights for components of this work although no longer accurate—numerical position to the
owned by others than ACM must be honored. Abstracting with credit is crew [19]. Similar situations can occur in many system
permitted. To copy otherwise, or republish, to post on servers or to redistribute
to lists, requires prior specific permission and/or a fee. Request permissions
control scenarios, including nuclear power plants, or large
from Permissions@acm.org. experiment or mission control centers..
CHI 2015, April 18 - 23, 2015, Seoul, Republic of Korea.
Copyright is held by the owner/author(s). Publication rights licensed to ACM. Signaling accuracy when in fact, there is none, can also
ACM 978-1-4503-3145-6/15/04…$15.00 have important consequences for drivers of electric vehicles
http://dx.doi.org/10.1145/2702123.2702479
(EVs), as charging stations are not nearly as pervasive (yet)

2201
UI Impact on Performance & Decisions CHI 2015, Crossings, Seoul, Korea

as gas stations. Drivers build expectations about available referring to EVs. Perceptions of range even influence EV
power resources and, through often unforeseeable changes purchasing decisions [1]. Second, drivers of EVs and
in conditions, will encounter situations in which these especially novice EV drivers are likely to rely more on the
expectations are disappointed [3, 16]. Being surprised by range display than drivers of combustion engine vehicles
what may come as a sudden lack of available range can [20]. Drivers of standard cars have a sense about the best
cause anxiety and stress, and can lead to potentially and worst case of fuel consumption and have developed a
hazardous situations. This concern stemming from limited feel for the available range based on the indicated fuel
range is often referred to as range anxiety, “an anxiety or level. As mileage is not a measure typically discussed with
fear that one may not reach a target before the battery is EVs it is much harder to translate a 30% battery charge into
empty, which can occur while driving or prior to driving as an available range. Third, EVs are a suitable test case
the user worries about later planned trips [9, p. 202].” because over-reliance on the information display can have
severe consequences for the consumer of that information.
We have thus focused on the range display in electric
Misjudging the battery life of a laptop might prevent one
vehicles, with the guiding question for our study being:
from sending an important email, but misjudging the
How do expectations about available energy resources
battery life of a car might leave a driver stranded on a busy
affect drivers’ attitudes towards the vehicle, the driving
highway or remote mountain road.
experience—with an emphasis on range anxiety—and
driving behavior?
BACKGROUND AND APPROACH
We focused on the format of the range display in an electric Trust and Range Displays
car, in particular, for three reasons: First, the range display How a device, process, or system displays information has
is a central instrument in electric vehicles. While rarely consequences for attitudes that users and operators develop
more than a gimmick in standard combustion engine towards that system. Situations in which expectations are
vehicles, the range display is a central instrument for set up based on a high level of presumed precision and
drivers of electric vehicles and it influences driving authority, which are subsequently not met, can affect trust
experience to a large degree [1, 20]. This focus is reflected not only towards the information displayed, but also
in the public discourse that has shifted from emphasizing towards the system as a whole [10].
efficiency and miles per gallon typical for internal
combustion engine vehicles to range in miles when The main challenge with current range displays is that they

Figure 1. The four study conditions. Each panel shows the dashboard as participants saw it when entering the vehicle. Remaining
range was displayed as either highly precise values (top row) or a diffuse color band that varied in width directly with remaining
range (bottom row).

2202
UI Impact on Performance & Decisions CHI 2015, Crossings, Seoul, Korea

represent an estimate. Remaining range is inferred based on regarding energy capacity and usage, mass and coefficient
factors that include battery capacity, vehicle dynamics, of drag, road topography, atmospheric conditions and
environmental conditions and driving style, and this driving behavior. The problem with these approaches to
estimate often leads to inaccurate predictions. An inaccurate increase prediction accuracy through improved algorithms
prediction can have several implications for reducing a is that, no matter how sophisticated, they cannot avoid
driver’s trust in the vehicle. For example, there might be a situations of unmet expectations: there will always be
situation where the vehicle displays a remaining range of 30 situations in which predictions will suddenly be unmet due
miles on a level road, and the driver wants to reach a to a change in driving context, such as the decision to take a
destination just 15 miles away on top of a hill. Even though mountain road instead of the highway.
the original estimate suggests sufficient reserve to reach the
Exceptions to the efforts of prior research to improve range
destination, the driver might not reach the top, as the steep
prediction accuracy are the work of Lundström and col-
incline leads to a fast decline in battery charge, resulting in
leagues. For example Lundström and Bogdan investigated
an effective range of just 10 miles. Sudden drops of range
“coping strategies” used by highly experienced EV drivers
are particularly frustrating for EV drivers [7], and whenever
to down-regulate stress when driving in limited resource
a prediction doesn’t meet the actual range, the car can be
conditions in order to inform the design of range
perceived as having misled, or even lied, to the driver. Due
information display to help less experienced drivers cope
to these instances it is understandable that EV drivers and
with similar situations [8]. Additionally, Lundström and
especially experienced drivers lose trust in the information
colleagues investigated alternative ways of displaying
presented by the vehicle [20] and often refer to the range
driving range, such as special visualizations of range on
meter as a “Guess-O-Meter” [7]. A large body of research,
maps [9], or visualizing factors that influence the available
built upon the Computers Are Social Actors paradigm [17],
range into the display [7]. While these approaches depart
supports the idea that people perceive and treat machines in
from the design strategy to increase accuracy of range
much the same way as they do other people. In that context,
algorithms, they do not question the utility of a specific
with cars that effectively, and repeatedly, “lie” to drivers, it
range number displayed to the driver.
seems reasonable that drivers might lose trust in their
vehicles.
Reflecting Uncertainty
Having the car make clear that the range display is just a To deal with situations of unmet or missed expectations, we
guess, drivers might feel a higher level of responsibility took an alternative strategy that has generally been
towards their ability to reach a destination. Even when overlooked by researchers concerned with energy source
running out of battery, drivers with an ambiguous range displays of mobile devices and cars. Rather than exploring
display might consequently not lose trust in their vehicle, as new algorithms that expand on the parameter space to
the car had not lied to them in the first place. increase prediction accuracy, we explored how a display
can not only make the nature of an estimate more salient,
Predicting Resources but also reflect the considerable amount of uncertainty
Other researchers have been aware of the problems arising associated with that estimate. Following the idea that
from inaccurate predictions of resources in cars and other ambiguity can be a valuable resource for designers, and
battery-dependent technologies. However, most of this prior particularly the guidance to use imprecise visualizations to
work has focused on designing algorithms that improve emphasize the uncertainty of information [22], we designed
prediction accuracy by increasing the number of parameters a range display that highlights prediction uncertainty. We
that the system takes into account. were particularly interested in the question of how the
ambiguous display of range information can mitigate
In their exploration of mobile phone battery charging
negative effects of missed expectations on attitudes, driving
behavior patterns, charge level indicators and user
experience, and driving behavior.
knowledge and reactions, Rahmati and colleagues [15]
found that inaccurate predictions of remaining battery life
THE STUDY
affected users’ sense of trust towards the phone, and We explored how the display of ambiguous range
suggest that improving the accuracy and resolution of information affects EV driving experience and behavior
predictions could make charging more convenient. In the through a 2 x 2 (state of charge (SOC) at start: low vs. high)
context of implantable or wearable medical devices, where x (range information ambiguity: low vs. high) factorial
lives depend on predictable power, designers have study design. Participants were asked to complete a specific
improved estimates of remaining battery life by including 19-mile long route to a predefined location driving a fully
ongoing operating parameters, such as pulse counters, along electric vehicle. Half of the drivers started out with what
with known device electrical characteristics [12]. In the appeared to be a 35% battery charge (about 30 miles) while
case of electric vehicles, Lundström et al. [9] suggest the other half started out with what appeared to be a 100%
including temperature into range prediction algorithms to battery charge (about 85 miles). Additionally, half of the
improve accuracy, while Zhang and colleagues [21] present drivers were presented with range information that
an algorithm that includes the vehicle’s immediate status

2203
UI Impact on Performance & Decisions CHI 2015, Crossings, Seoul, Korea

suggested precision (2 significant digits), whereas the other lighting conditions, and bright red for the range indicator to
half of the drivers were presented with ambiguous range make it particularly prominent in the overall design.
information (see Fig. 1). This setup was designed to provide
insight into how participants react to the range display in Participants
situations when expectations are not met in everyday and We recruited 73 participants for the study. All participants
anxiety provoking situations. had a valid US driver’s license, were students, were a
minimum 21 years of age, had no previous EV driving
The display experience, but sufficient general driving experience.
The car used for the study was a Honda Fit fully electric General driving experience was required so that any stress
pre-production prototype that was further modified by and cognitive load induced through the display changes
Honda engineers to allow for the integration of our display would not get confounded with stress and cognitive load
setup. In order to manipulate the apparent level of battery from standard driving scenarios. Participants were recruited
charge (low versus high) and the mode of range display via email announcements and were asked to sign up on an
(low and high range information ambiguity), we developed online scheduling system. Participants were randomly
a dashboard interface on the basis of an android tablet that assigned to one of the four study conditions and were
could be placed in front of the actual dashboard of our gender balanced as much as possible across conditions. We
study vehicle (see Fig. 2). also scheduled all study runs outside of rush hours, and all
runs were conducted under similar weather conditions to
minimize potential influence from traffic and weather on
our measures. Each participant received a $50 Amazon gift
card in exchange for his or her participation.

Procedure
Upon arrival at the lab, participants were greeted and asked
to take a seat in front of a laptop that displayed a survey.
Participants were told that their help was needed to explore
the experience of driving an electric vehicle with a newly
designed display. After asking for their informed consent,
participants signed in on a log sheet that determined the
experimental condition for the participant. Participants were
then asked to enter their participant id into an online survey
tool, and to follow the instructions given on the screen. The
survey tool introduced the participants to the following
Figure 2 Photo of the in-car display setup. specifics of the task:
The tablet was directly connected to the car’s CAN-bus via Your task will be to take a short drive with our Electric
a CAN-bus reader, which allowed one-way access to data Vehicle to Alice's Restaurant and park the car in the
about current battery charge, vehicle speed, and remaining parking lot (see picture below). The drive to Alice's
range. The tablet was fastened with Velcro in front of the restaurant should take about 30 minutes.
dashboard in such a way that it fully covered the original We have programmed this route into the car's navigation
instruments but also such that it was easily removable in system. Please follow the directions as stated by the
case the software crashed during the drive. Additionally, a navigation system. As soon as you arrive at Alice's
frame was constructed around the tablet to prevent light Restaurant and safely parked the car, please call the study
reflections that might inhibit the readability of the display. administrator to let him or her know that you have arrived.
Participants were specifically instructed to remove the He or she will give you his phone number before the drive.
tablet in case something seemed wrong with the display.
While parked at Alice's Restaurant you will be asked to fill
The display itself was designed with several parameters in out a short survey about your driving experience. As soon
mind—in particular: content, layout and color. Regarding as the survey is completed, please drive the car back to the
content, we reduced the information presented to drivers to starting location. The route will be already programmed
only what was needed to safely navigate on the road and to into the navigation system of the car. In case of any
experience the study manipulations: current speed, current problems with the car, please stop the car somewhere safe,
state of charge and remaining range. Regarding layout, we and give the study administrator a call.
arranged the indicators so that range was most prominent,
and they were all visible through the inner rim of the As part of the task description, participants were shown a
steering wheel. Regarding color, we chose light blue on a map of the roughly 19 mile long route they were asked to
black background for high readability under various drive, as well as a picture of the destination. Participants
were told to park the car the car safely in front of the

2204
UI Impact on Performance & Decisions CHI 2015, Crossings, Seoul, Korea

restaurant shown on the picture and then call the everyday driving: suburban streets, highway, mountain and
experimenter for further instructions. At the beginning of countryside. Third, and most importantly we designed the
the study participants were not made aware that the driving route to have an elevation profile that would remain flat for
task would be over once they reached Alice’s restaurant. the first 3/4th of the drive and then suddenly increase in
slope going up the mountain road (see Fig. 4). As the range
Before starting the drive, participants were asked to wear a
estimation of the vehicle is calculated by extrapolating from
skin conductivity sensor (affective Q sensor) measuring
current driving conditions, this profile was intended to set
electro dermal activity on their wrist, as well as a heart rate,
expectations of sufficient range during the initial phases of
body movement and respiration monitor on their upper
the drive and then disappoint those expectations when
torso (Zephyr Bioharness BT). Both sensors were calibrated
driving up the hill which lead to a rapid decline in range
by asking the participant to relax for 5 minutes while their
due to increased energy consumption.
vitals are being recorded. This paper will however not
report on data collected from these sensors.
As soon as finished with the online survey, participants
were introduced to the vehicle and instructed to drive
according to the directions from the GPS navigation
system. Everything the experimenter said and did was
carefully scripted and the experimenter was instructed to
respond to possible questions about remaining range with a
standard answer and not to make any additional
explanations: “If there is any problem, just give me a call.”
This evasive answer was to prevent the experimenter from
inadvertently making participants feel psychologically safe
when encountering doubts about available range. However
none of the participants mentioned experiencing stress
greater than what they had previously encountered when
nearing an empty gas tank. The research was IRB approved Figure 3. Map of the 19 miles test route
and all drivers were fully insured for the study. Additionally
the vehicle could be stopped safely along the road at any
time during the study and the experimenter made sure that
each participant had a working cell-phone and gave his or
her cell phone number to the participant before the drive.
Upon arrival at Alice’s Restaurant, participants called the
experimenter who instructed them to first fill out a
prescribed questionnaire package to reflect on their main
driving experience, and then call again. Afterwards
participants were debriefed over the phone at which point
the study was formally over. In the debriefing, participants
were told that the car had been fully charged with plenty of
resources to return to the starting location. Upon arrival at
the start location each participant was then asked to
complete an exit survey about participant demographics
after completion of which each participant received the $50
gift certificate.
Figure 4. Elevation and road/driving context of the 9 miles test
The Route route
The route was carefully designed to fulfill a set of specific
characteristics. First, the drive was about 30 minutes and Manipulating SOC and Range Information Ambiguity
long enough to create an immersive driving experience. We Initial Values
chose a rout that would ask people to drive up a long and We manipulated the apparent level of initial battery charge
winding road to give drivers a sense that they are on their by simulating a state of battery charge at the beginning of
own. This was also amplified by choosing a destination the drive (SOC at start) to read 100% in the “high”
away from the starting location rather than a circular route. condition and 35% in the “low” condition. Unbeknownst to
We hoped that any sense of range anxiety would be participants, and independent of what level of charge was
amplified by this setting. Second, the route featured displayed in the vehicle, the experimenters always ensured
different road conditions and terrains encountered in that the actual level of charge was high enough to allow

2205
UI Impact on Performance & Decisions CHI 2015, Crossings, Seoul, Korea

safe completion of the course for all participants. The initial decreased. The width of the interval was designed so that it
low charge state of 35% was carefully determined such that still signaled uncertainty even when approaching zero.
when driving in “eco” mode (the default economic setting
Both low and high ambiguity display conditions were
as participants entered the car), the display indicated 32
carefully designed such that the only difference between
miles of remaining range at the start of the drive, and
them was in the level of ambiguity around the range
(typically) indicated 2.17 miles of remaining range upon
prediction. Therefore both displays were equal in size and
reaching the destination. Given that the driving course was
layout. Even the red box showing the remaining range value
only 19 miles long, we expected participants to be
was roughly equal in size to the red uncertainty interval. In
comfortable starting a drive with this level of charge.
addition to changing the display we also modified the
For participants in the 100% condition, the displayed state spoken limited range warning from “The battery charge is
of charge and range decreased as they ordinarily would not sufficient to reach the destination” in the low ambiguity
over time, given current driving and road conditions. For condition to “The battery charge might not be sufficient to
participants in the 35% condition, the displayed state of reach the destination” in the high ambiguity condition.
charge and range decreased at a lower, scaled rate, to ensure
that they would never reach negative values. Measures
The main measures of interest were Driving Experience,
Warnings Trust towards Vehicle, and Driving Behavior. As controls,
To increase participants’ awareness of a low state of charge we measured age, gender and driving experience. However
we programmed the tablet to issue warnings at specific we did not include these variables in our final analyses as
levels or SOC. These low range warnings were displayed in they did not significantly change our results.
all conditions when the remaining range reached certain
thresholds; however, only the participants in the low charge Driving Experience
conditions reached an SOC low enough to trigger the Driving experience was measured by assessing i) range
warnings. The intensity of the warning messages increased anxiety, ii) driving pleasure, and iii) level of calmness. The
in three stages: First, a yellow battery warning symbol (in range anxiety measure was operationalized through the
between the speed and range indicators) lit up. Second, the following four items: “I sometimes felt frightened while
battery symbol started to blink with a frequency of 1 Hz. driving the car,” “I often felt anxious while driving the car,”
Third, a spoken range warning was issued twice informing “Driving the vehicle often made me nervous,” and “I got
the driver about insufficient range to reach the destination. totally scared during the drive.” The items deliberately did
The range warnings were carefully adjusted so that they not refer explicitly to “range anxiety” as we did not want to
appeared only after the participants had started to enter the tap into lay theories about range anxiety that people are
road with the steep incline. The late triggering of the range likely to have and therefore assessed directly the experience
warnings was intended to heighten range awareness but of anxiety. Additionally, we did not want to make
prevent participants from aborting the drive before driving participants aware that the study was about range. Due to
up hill on a road section that would make it difficult to turn the randomized assignment of the participants to study
around. conditions, and since nothing but SOC and display were
changed between conditions, differences in anxiety can be
Information Ambiguity attributed to differences in the experience of range anxiety.
We manipulated the level of information ambiguity of the Participants rated for each statement from “Not at all (1)” to
range display by modifying how range was displayed to the “Totally (10)” how much they agreed with each statement
driver. In the low ambiguity condition, the remaining range on a 10-point scale. The scale was reliable (α = .89).
was displayed with two digits after the decimal point, Driving pleasure was measured by taking the pleasure
signaling precision and a high level of confidence in an rating from the affect grid, a validated and widely used
accurate prediction. The range value was set within a red measure of pleasure and arousal [18]. Level of calmness
box along the range indicator, which moved leftward as was measured by asking people on a 10-point scale ranging
participants drove (since range values decreased to the left). from “not at all” to “extremely” how much they agreed with
the statement “I was calm while driving.”
In the high ambiguity condition, the remaining range was
displayed without showing a number, but instead by
Trust towards the vehicle
displaying a kind of uncertainty interval around the Trust towards the vehicle was operationalized by adapting
calculated range—represented visually as a colored and combining two items from a widely used survey tool
gradient, fading to the background on either side. The originally designed to measure trust towards automated
interval also moved leftward as participants drove, systems [5]. We asked participants to rate on a 10-point
mirroring the position of the range value box in the low scale ranging from “not at all” to “extremely” to what
ambiguity condition. The width of the uncertainty interval extend they agreed with the following two statements: “I
was determined as a percentage of the estimated range, so am confident in the vehicle,” and “I am familiar with the
that its horizontal width shrank as the remaining range

2206
UI Impact on Performance & Decisions CHI 2015, Crossings, Seoul, Korea

vehicle,” (α = .74). We deliberately framed the questions to manipulation on how ambiguous the range information was
refer to the vehicle rather than the display, as we were perceived, F (3,54) = 42.92, p < .0001. When the range
interested in how changes affect the overall perception of information was displayed unambiguously (M = 7.29)
the car. people on average perceived the displayed information
almost twice as ambiguous than when the information was
Driving Behavior displayed ambiguously (M = 4.18) (Figure 5).
Driving behavior was measured by assessing how efficient
participants drove. The measure of efficiency was We also found a significant main effect of our SOC
operationalized by calculating the difference between the manipulation on perceptions of sufficiency of resources, F
battery state of charge level (SOC) before and after (3,54) = 85.05, p < .0001). When the car was fully charged
completing the main driving task. at the beginning of the drive (M = 8.21), participants on
average were more than twice likely to report that the
RESULTS charge was sufficient than when the car was only partially
Analysis of Variance (ANOVA) was used for data analysis. charged (M = 3.80) (Figure 6).
Data from 57 out of 73 participants contributed to the final Therefore we found strong evidence that our manipulations
data analysis. 10 participants had to be excluded because had the intended effects. We were able to create a situation
they called the experimenter before reaching the final in which people believed that their resources were barely
destination and therefore had a driving experience that was sufficient for the course given and we were successful in
not comparable to the other participants. An additional 6 designing a display that communicates ambiguity when
participants had to be excluded due lack of adherence to the presenting an estimate.
study protocol and due to technical errors. Table 1 shows
Driving Experience
the final participant distribution across conditions.
Range anxiety
SOC
Low High We found a significant main effect of the starting SOC on
Range Information low 11 (6m/5f) 17 (8m/9f) range anxiety (Figure 7). Range anxiety was significantly
Ambiguity high 15 (11m/4f) 14 (7m/7f) higher when the SOC at start was low (M = 3.68, SE = .29)
in comparison to when the SOC at start was high (M = 2.36,
Table 1. Participants’ distribution across conditions and by SE = .29), F (3,54) = 8.6, p < .01.
gender (male/female)
Calmness
Manipulation Checks We also found a significant main effect of the range
Two measures served as manipulation checks. To assess information ambiguity on participants’ perceptions of how
whether we successfully manipulated range information calm they felt during the drive (Figure 9). Participants felt
ambiguity we used a 3-item scale. Participants were asked significantly more calm in the “Range Ambiguity: high”
to rate on a 10-point Likert-type scales ranging from “Not condition (M = 7.20; SE = .39) than in the “Range
at all (1)” to “Totally (10)” how much they agreed with Information Ambiguity: low” condition (M = 6.04, SE =
statements such as “The range display was very precise in .41), F (3,54) = 4.42, p < .05. Participants also felt
providing the remaining number of miles given the battery marginally significantly more calm in the SOC high
charge.” The scale was reliable (α = .73). To assess whether condition (M = 7.10; SE = .38) than in the SOC low
we successfully manipulated beliefs of sufficient resources condition (M = 6.14, SE = .42), F (3,54) = 2.88, p = .10.
we created a 3-item scale including questions such as “The Interestingly the display type seemed to be able to
car had plenty of battery charge for the route I was given.” compensate for the SOC differences: Even though not
Participants rated on a 10 item scale from “Not at all” to statistically significant it is interesting to see that people felt
“Totally” how much they agreed with each statement. The on average more calm when driving with a low SOC and
scale was reliable as well (α = .91). We found a significant high range information ambiguity display than when
main effect of our range information ambiguity driving with a high SOC and low ambiguity display.

Figure 5: Impact on display changes Figure 6: Impact of charge at the Figure 7: Range Anxiety was increased
on perceived accuracy of range beginning on perceived sufficiency of when participants started the drive
information. charge (SOC). with a low battery charge.

2207
UI Impact on Performance & Decisions CHI 2015, Crossings, Seoul, Korea

Driving Pleasure Limitations

We found a marginally significant effect of the SOC level Several limitations should be noted. First, the complex
on driving pleasure (Figure 10). Participants on average felt study setup made it difficult to obtain data from large
better when driving in the “SOC: high” conditions (M = numbers of participants. However, given that research
6.44; SE = .39) than when driving in the “SOC: low” around on interactive systems in the car is often conducted
conditions (M = 5.36; SE = .44) F (3,54) = 4.18, p < .05. in simulators to ease data collection, our research provides
The type of display did not have any significant effect on findings that have higher ecological validity and therefore
participants overall emotional state when driving. provide insights that we might not have been able to gather
if we had run the study in a simulator. Second, we did not
Trust towards Vehicle include conditions in which drivers did not encounter a
We found a significant main effect of the SOC of the car on large discrepancy between expected and actual range. For
trust towards the car (Figure 11). On average, participants example we could have asked additional participants to
trusted the car significantly more if they were driving with a drive a 19 mile route on a flat terrain. While additional
full battery (M = 6.62, SE = .36) than with an almost empty study conditions might have given additional insights we
one (M = 5.21; SE = .41), F (3,54) = 6.70, p < .05. There believe that the conditions in which participants drove with
was also a marginally significant main effect of the range a full charge can be taken as a sufficient baseline. In fact,
information ambiguity on trust, as participants on average every time people drive a car there is a, however small,
trusted the car more when it displayed range ambiguously discrepancy between expected and actual range. Since the
(M = 6.43; SE = .37) rather than unambiguously (M = 5.39; high-SOC conditions did not cause situations in which
SE = .40), F (3,54) = 3.64, p = .06. people had to be concerned about running out of fuel, those
conditions can be taken as a “control.” While more
Driving Behavior amplified, the high SOC conditions were not fundamentally
We found a significant interaction between range different from everyday driving situations as they did not
information ambiguity and SOC on driving efficiency F create an experience of severe range anxiety. Finally, our
(3,54) = 4.28, p < .05 (Figure 8). When participants drove research only included participants who had no prior EV
in the “SOC: high” conditions, they used on average more driving experience. Previous research has shown that range
energy when driving with the ambiguous range display (M anxiety is particularly a phenomenon experienced by novice
= 34.37; SE = 3.38) than when driving with the drivers [2]. It might therefore be possible that some effects
unambiguous range display (M = 32.16; SE = 2.79). would be weakened with more experienced drivers of
However this trend was reversed in the “SOC: low” electric vehicles. However, Strömberg and colleagues [20]
conditions: Here drivers drove almost 10% more efficiently argued that even experienced drivers experience concerns
when driving with the ambiguous range display (M = 31.65; over range during unplanned trips.
SE =6.66) than when driving with the precise range display
(M = 34.60; SE = 5.00). DISCUSSION
Current range information systems are designed to take
incomplete information and present it as trustworthy to the
driver. Based on findings from a study about use of GPS
navigation systems, Leshed and colleagues suggested that it
might be advantageous to present error-prone information
ambiguously [11]. We explored this suggestion and, in fact,
found that presenting information ambiguously can have
advantages. Revealing, rather than hiding, error-prone
information led to improvements in driving experience,
behavior and trust towards the car.
Figure 8: Effects of charge and range information ambiguity
on energy consumption.

Figure 9: Participants felt calmer with Figure 10: Participants had a more Figure 11: Effects of charge and range
the ambiguous range information. pleasurable driving experience when the car information ambiguity on trust towards
was fully charged. the vehicle.

2208
UI Impact on Performance & Decisions CHI 2015, Crossings, Seoul, Korea

Surprisingly, the beneficial effects of presenting Finally, our findings contribute to research on interactions
information ambiguously were even stronger in critical and between humans and automated systems in general. The
highly stressful situations. premise of our research was that automated systems, like
people who do not keep a promise, might be perceived as
Maybe our most exciting finding is that the ambiguous
untrustworthy if set expectations are subsequently not met.
range display supported a driving style that was more
Parasuraman and Miller [14] demonstrated that following
adaptive to the available resources: People with the
social rules of etiquette in presenting errors to a user not
imprecise display drove more agile when having enough
only improved trust towards the system but overall system
resources but drove more carefully when driving with
reliability. Our finding extends this work by demonstrating
limited resources. In contrast people with the precise
that systems that are more careful in how they make
display took less useful driving strategies: They drove
promises not only improve user experience but also overall
economically when having plenty of resources and more
system performance (in this case driving behavior).
aggressively when driving with limited resources. This
finding, that the adaptiveness of driving style improved An open question for future research remains whether high
when information was presented ambiguously, is in line and low information ambiguity display styles can be
with Langer and Piper’s research on mindfulness [6]. In combined to promote more sustainable driving behavior. It
several studies, the authors showed that participants, who might be possible that transitioning from an unambiguous
were presented with new information in a way that did not to an ambiguous display design at an appropriate SOC level
communicate certainty, were subsequently more able to use might lead drivers to use less overall resources. Presenting
that information more adaptively. Our ambiguous display in an imprecise display, particularly when the battery depletes,
combination with the ambiguous range warnings might may also serve the purpose of avoiding serious system
have promoted a more “mindful” and adaptive driving style consequences such as deep discharge of a battery, which is
in the participants. undesirable and shortens battery life. Future research could
therefore explore the transitioning effect itself and how
Our study contributes to research on range anxiety. To our
switching between different range displays might affect
knowledge this study is the first to employ an experimental
driving experience and behavior. Another avenue for future
research paradigm to elicit and study this phenomenon
research is to explore possible long-term learning effects of
under controlled conditions. We found that driving with
ambiguous range information displays. Our findings
limited resources does create increased levels of anxiety
stemmed from observations of novice EV drivers and it is
and significantly impairs the overall driving experience.
unclear how experienced drivers react to range information
People not only felt more range anxiety when driving with
that is displayed ambiguously and how drivers interpret
limited resources, they also experienced driving as less
ambiguously visualized information over the long term.
pleasurable and less calm. An encouraging finding was that
Interestingly, past research found that while experienced
the ambiguous range display had a tendency to make people
drivers develop an explicit understanding of the
feel calmer when driving. This finding is particularly
unreliability of the predicted range, experiences of
interesting since most drivers had told us in earlier
discrepancies between range prediction and actual range
prototypes evaluations of our display that they did not like
still lead to considerable frustration [7]. This suggests that
seeing the range information presented in an ambiguous
displaying range information ambiguously might help even
display. Our findings about range anxiety are also
more experienced drivers cope with frustrations stemming
interesting as prior work on range anxiety has relied
from unreliable predictions.
predominantly on survey research methods and not
explored whether this phenomenon can be experimentally Conclusion
stimulated [3, 16]. Therefore our test protocol provides a Our study demonstrates that it can be advantageous to
useful starting point for other researchers to study range highlight the uncertainty associated with a measure rather
anxiety and to evaluate novel designs to counteract it. than to hide it. While presenting a single number can be fast
Our study also contributes to research on progress bars [13]. to read and easy to apprehend, designers have to carefully
Progress bars indicate that an operation is proceeding consider the implications that disguised uncertainty may
successfully but also make predictions about completion have on user experience and behavior especially in critical
time. Similar to range predictions, these completion time situations. Although tested on drivers of electric vehicles
predictions are often unreliable due to non-linear only, the findings of the present study may not be limited to
progressions of the tracked processes [4] and are therefore EV applications alone, but may potentially be considered
prone to elicit user frustrations. While previous research has for other applications that require solemn decision-making
predominantly focused reducing perceived durations to in the face of unreliable information during rapidly
improve user experience, displaying progress information depleting battery power sources, including critical function
ambiguously might be an alternative strategy to improve portable medical, communication and electronic devices.
user experience.

2209
UI Impact on Performance & Decisions CHI 2015, Crossings, Seoul, Korea

ACKNOWLEDGMENTS 10. Lee, J. D. & See, K. Trust in automation: Designing for

The authors are grateful to our reviewers for their helpful appropriate reliance. In Human Factors: The Journal of
comments and suggestions and to Honda R&D Americas the Human Factors and Ergonomics Society 46:1
for their financial support during the conduct of this study. (2004), 50-80.
We would also like to acknowledge the help of several
11. Leshed, G., Velden, T., Rieger, O., Kot, B., & Sengers,
highly dedicated students who helped us carry out this
P. In-car gps navigation: engagement with and
research: Halsey Hoster, Leko Murphy, Scott Purpura, Jea
disengagement from the environment. In Proc, CHI
Min Koo, and Brian Mok.
2008, ACM Press (2008), 1675-1684.
REFERENCES 12. Mallela, V., Ilankumaran, V. & Srinivasa Rao, N.
1. Bunch, D., Bradley, M., Golob, T., Kitamura, R. & Trends in cardiac pacemaker batteries. In Indian pacing
Occhiuzzo, G. Demand for clean-fuel vehicles in and Electrophysiology Journal 4:4 (2004), 201-212.
California: A discrete-choice stated preference pilot 13. Myers, B. A. (1985). The importance of percent-done
project. In Transportation Research Part A: Policy and progress indicators for computer-human interfaces. In
Practice 27:3 (1993), 237-253. Proc. CHI '85, ACM Press (1985), 11-17.
2. Franke, T., Cocron, P., Bühler, F., Neumann, I., & 14. Parasuraman, R., & Miller, C. A.. Trust and etiquette in
Krems, J. F. Adapting to the range of an electric high-criticality automated systems. Communications of
vehicle–the relation of experience to subjectively the ACM 47, 4 (2004), 51-55
available mobility resources. In Proc. of the European
15. Rahmati, A., Qian, A. & Zhong, L. Understanding
conference on human centered design for intelligent
human-battery interaction on mobile phones. In Proc.
transport systems, (2012), 95-103.
MobileHCI 2007, ACM Press (2007), 265-272.
3. Franke, T., Neumann, I., Bühler, F., Cocron, P. &
16. Rauh, N., Franke, T., & Krems J. F., Understanding the
Krems, J., Experiencing range in an electric vehicle:
impact of electric vehicle driving experience on range
Understanding psychological barriers. In Applied Psy-
anxiety. Human Factors (2014), 1-19.
chology: An International Review 61 (2012), 368–391.
17. Reeves, B. & Nass, C. The media equation: How people
4. Harrison, C., Amento, B., Kuznetsov, S., & Bell, R.
treat computers, television, and new media like real
Rethinking the progress bar. In Proc. UIST’07, ACM
people and places. CSLI Publications and Cambridge
Press (2007), 115-118.
University Press, (1996).
5. Jian, J. Y., Bisantz, A. M. and Drury, C. G., Foundations
18. Russell, J. A., Weiss, A., & Mendelsohn, G. A. Affect
for an empirically determined scale of trust in automated
grid: a single-item scale of pleasure and arousal. Journal
systems. International Journal of Cognitive Ergonomics
of personality and social psychology 57,3 (1989), 493-502.
4 (2000), 53-71.
19. Sheridan, T. B., & Parasuraman, R., Human-Automation
6. Langer, E. J., & Piper, A. I. The prevention of
Interaction, Reviews of Human Factors and Ergonomics
mindlessness. Journal of Personality and Social
1 (2005), 89-129
Psychology 53, 2 (1987), 280-287.
20. Strömberg, H., Andersson, P., Almgren, S., Ericsson, J.,
7. Lundström, A. Differentiated Driving Range: Exploring
Karlsson, M., & Nåbo, A. Driver interfaces for electric
a Solution to the Problems with the Guess-O-Meter in
vehicles. In Proc. AutomotiveUI 2011, ACM Press
Electric Cars. In Proc. AutomotiveUI 2014, ACM Press
(2011), 177-184.
(2014), 1-8.
21. Zhang, Y., Wang, W., Kobayashi, Y. & Shirai, K.
8. Lundström, A. & Bogdan, C. (2012) COPE1 – Taking
Remaining driving range estimation of electric vehicle.
Control over EV Range. In Adjunct Proc. AutomotiveUI
In Proc. IEVC 2012. IEEE (2012), 1-7.
2012. [Online]. 2012 pp. 17–18
22. Gaver, W. W., Beaver, J. & Benford, S. (2003, April).
9. Lundström, A., Bogdan, C., Kis, F., Olsson, I. & Fahlén,
Ambiguity as a resource for design. In Proc. CHI 2003,
L. Enough power to move: dimensions for representing
ACM Press (2003), 233-240.
energy availability. In Proc. MobileHCI 2012, ACM
Press (2012), 201-210.

2210