Content uploaded by Jorge Goncalves
Author content
All content in this area was uploaded by Jorge Goncalves on Oct 05, 2017
Content may be subject to copyright.
Monetary Assessment of Battery Life on
Smartphones
Simo Hosio, Denzil Ferreira, Jorge Goncalves, Niels van Berkel, Chu Luo,
Muzamil Ahmed, Huber Flores, Vassilis Kostakos
Center for Ubiquitous Computing
University of Oulu, Finland
{simo.hosio, denzil.ferreira, jorge.goncalves, niels.van.berkel, chu.luo,
muzamil.ahmed, huber.flores, vassilis.kostakos}@ee.oulu.fi
ABSTRACT
Research claims that users value the battery life of their
smartphones, but no study to date has attempted to quantify
battery value and how this value changes according to
users’ current context and needs. Previous work has
quantified the monetary value that smartphone users place
on their data (e.g., location), but not on battery life. Here we
present a field study and methodology for systematically
measuring the monetary value of smartphone battery life,
using a reverse second-price sealed-bid auction protocol.
Our results show that the prices for the first and last 10%
battery segments differ substantially. Our findings also
quantify the tradeoffs that users consider in relation to
battery, and provide a monetary model that can be used to
measure the value of apps and enable fair ad-hoc sharing of
smartphone resources.
Author Keywords
Smartphones; battery value; auction; user study; monetary
model; resource sharing.
ACM Classification Keywords
H.5.m. Information interfaces and presentation (e.g., HCI):
Miscellaneous
INTRODUCTION
Research on smartphone battery life has typically focused
on improving the energy efficiency of hardware, software
and network protocols [31], or on understanding user
strategies for battery management [10,11,33]. While the
energy efficiency of smartphones is a priority for hardware
and software providers [3], the increasing screen sizes and
sensor capabilities have practically stagnated the perceived
battery life available for end users [30]. For example, a
large-scale longitudinal study exploring the charging habits
of more than 4000 smartphone users found that they charge
their devices frequently throughout the day [10], and
showed that users perceive battery draining as a tradeoff
against obtaining value from an application. Thus, a user
may happily play games to kill time on full battery, but may
stringently conserve battery when it is almost depleted,
saving it for “valuable” use such as emergency calls or
wayfinding.
Given these concerns, an important way to characterise
smartphone use is to quantify the value that smartphone
users place on their devices’ battery. Doing so can be a first
step towards systematically characterising individual
applications based on the value that they provide to users
(as measured through the battery-tradeoff), as well enabling
fair ad-hoc resource sharing between devices [7].
Here, the research question we answer is: how much value
do smartphone users place on their battery life? Previous
studies have systematically quantified the monetary value
of sensitive data, such as location, communication logs, or
apps use (e.g. [6,34]), but surprisingly we are not able to
find studies that measure the perceived value of smartphone
battery life. We present our findings from a small-scale
pilot study and a field trial where 22 participants auctioned
their device’s remaining battery life in exchange for
monetary rewards. We also include results from semi-
structured interviews and a concluding workshop.
We begin by demonstrating that the monetary value of
battery is not constant, but inversely related to current
battery level. Battery life is valued about 3 times more
when it is near depletion than it is when fully charged.
Second, we show that users may associate intrinsic value
with battery life. For instance, they may be willing to
donate battery to friends in exchange for social capital.
Finally, we utilise a well-known methodology in a new
context, and describe how context and the renewable nature
of a mobile resource such as battery life pose obstacles to
similar methodologies that use second-price auctions.
RELATED WORK
Value-driven frameworks
Previous work has attempted to quantify the overall value
that users obtain when using smartphones. For instance, one
Permission to make digital or hard copies of all or part of this work for
personal 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 and the full citation on the first page. Copyrights for
components of this work owned by others than the author(s) must be
honored. Abstracting with credit is 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 from
Permissions@acm.org.
CHI'16, May 7-12, 2016, San Jose, CA, USA
Copyright is held by the owner/author(s). Publication rights licensed to
ACM.
ACM 978-1-4503-3362-7/16/05…$15.00
DOI: http://dx.doi.org/10.1145/2858036.2858285
study identified 15 value elements that users consider,
including convenience and pleasure, which both provide
satisfaction and influence choice of product [29]. A similar
study conducted on mobile internet usage specifically,
identified four types of value: functional (defined as
technical and practical benefits e.g., Internet, email),
emotional (explained as emotion related benefits e.g.,
watching movie), social (to fulfil interaction purposes e.g.,
chatting), and monetary (benefits in terms of money) [38].
Such value assessment frameworks depend on users’ age,
personality, or demographic characteristics [29], as well as
cultural factors [38]. In general, we find that monetary
value is an important aspect of how users value their
smartphones, and therefore could be considered in the
context of battery life. A battery-value framework that
encompasses an understanding of its value to users is
needed.
Smartphone Resource Sharing and Use
Battery life is crucial to opportunistic sharing of
smartphone resources, which in itself has the potential to
reduce overall energy consumption, improve application
responsiveness, and lead to new possibilities for mobile
services and applications [36]. For instance, by sharing
wireless connectivity across devices, it is possible to create
highly scalable sensor networks [22,23]. These networks
can be small in scale (personal sensing), include a group of
people with a common background (group sensing), or
consist of a larger user base (community sensing) [23].
Other opportunistic resource sharing studies involve sharing
of GPS location [36] or processing power [27].
Consequently, an important aspect of resource sharing is
power consumption, and understanding how owners of the
devices value their battery life. Previous studies have
quantified the value of mobile phone resources (albeit not
of battery life) and have identified differences in how users
value different types of personal data [26,34]. These
valuations may nudge day-to-day smartphone use, since
users constantly weigh various choices against each other:
for example, giving up privacy in return for added value of
an application. Furthermore, data collected through
smartphone use is of high interest to both commercial and
research entities, giving rise to studies on the monetisation
of personal information (PI) [34]. While the proclaimed
monetary value of PI differs across users, some general
characteristics can be found. For example, the identity of a
user (containing personal information) has been rated to
have double the value compared to revealing usage history
of social websites [6]. Hence, we can expect that while
users may have distinct strategies at evaluating their battery
life they may also exhibit similar overall trends.
Human-Battery Interaction
Smartphone battery life has been shown to be a major
concern for users. A survey from as early as 2007 revealed
that 80% of mobile phone users take measures to increase
the lifetime of their mobile devices [32]. Research on
improving the battery life of mobile devices has since
focused on improving energy efficiency of hardware and
software, for example by reducing the amount of data being
transmitted, increasing the capacity of the internal battery,
or restricting the resources allocated for idle (background)
applications [5]. Another approach being actively
investigated is code offloading: migrating mobile code to be
executed remotely in the cloud or on dedicated servers,
leading to energy savings on the mobile devices [15]. The
mentioned examples demonstrate how devices can be
designed and optimised for energy savings.
Related to our work, studies have shown how users
perceive and attempt to manage smartphone battery.
Rahmati et al. [32] investigated the interpretation of battery
life information. Their study indicated that users generally
have limited knowledge regarding the actual battery
characteristics of their phone, which suggests that detailed
wattage information may not be as useful to them as battery
percentile or hours of battery life. Following this, the
researchers state that current battery interfaces are too
complex (both cognitively and technologically) for users to
effectively interpret and configure battery-related settings.
A study investigating charging behaviour over a large group
of users has identified distinguishable patterns across them
[10], showing distinct preferences and keeping-alive
strategies. To facilitate battery life management, the Task-
Centered Battery Interface (TCBI) [35] and the Interactive
Battery Interface (IBI) [11] draw users’ attention to a set of
phone activities, or specific battery draining applications
and their impact in their device’s battery life, respectively,
allowing users to make informed decisions on what to do in
order to keep their device alive. Carat [1] turns battery
management into collaborative effort, where the “crowd”
provides application’s battery impact estimations,
generating post-hoc reports with suggestions to improve the
device’s battery life.
However, despite earlier studies providing a wide array of
findings on user strategies and expectations on battery life,
as well as automated battery management tools [40], they
do not clearly quantify the value users associate to battery
life. Users nowadays do know what to do to extend their
battery life [11]. Having a better understanding on how
battery is valued can help improve automated battery
management, resource sharing, and quantify smartphone
use.
EXPERIMENTAL DESIGN
We designed an experiment where participants could
auction their smartphone battery on an hourly basis, and
winners of each auction would collect monetary rewards in
exchange for rapidly depleting their device’s battery. Data
transfer between participants’ smartphones and our server
was performed in real-time using network sockets to ensure
that the auction winners were rewarded immediately.
Actual battery depletion was managed by our smartphone
software.
Auction Procedure
Each participant in our study was prompted by their
smartphone using a notification to bid on an hourly basis,
every day from 10:00 to 22:00 for the duration of the study.
This meant that each day we had 13 auctions. At each
auction participants bid their desired amount of money for
giving up exactly 10% units from their currently remaining
battery life. We decided to control this variable and keep it
constant at 10% units to increase the power of our statistical
analysis and make our results comparable across
participants. Our smartphone software detected whether a
participant charged their phone during these hours, and
would subsequently exclude them from bidding for the rest
of that particular day.
All auctions followed the same procedure: a reverse
second-price closed-bid auction (i.e., a reverse Vickrey
auction). This means that the bidder with the lowest bid
wins, but receives the amount indicated in the second
lowest bid. Bids of other participants are not revealed to
bidders, i.e., it is a sealed-bid auction. The mechanism has
been shown to be truth-telling, as the optimal strategy for
the bidders is simply to be honest in their bids [25]. This
auction model is also conceptually clean [9] and thus easy
to understand and explain to participants [34]. Finally, the
model has been used in auctioning personally identifiable
information [34] and Web browsing behaviour [6].
The daily starting and ending hours of auctioning were
chosen to ensure that most participants would be awake and
alert to place bids simultaneously. It would be easy for
participants to rig an auction where only 2 bidders are
awake to place bids, e.g. in the very early hours such as
04:00 or 05:00. For each auction notification, participants
were given a window of 10 minutes to place their bids.
Participants could also choose to dismiss the notification
and not place a bid at that particular auction. Once the
bidding window closed, the notification on participants’
smartphones who had not placed a bid was withdrawn.
The winner of each auction was determined 10 minutes
after the bidding closed. The 10-minute threshold ensured
that data was synchronized between participants and our
servers. Shortly after a winner had been determined by the
server, the winning device was notified and began draining
10% units of the remaining battery life. If a device was
unreachable (e.g., offline) the notification appeared
whenever the device was switched back on again. The
money for the winning bid was released only after the
smartphone’s battery was depleted ceaselessly by 10%
units, and the depletion verified by our software.
Smartphone Logger
The logger was responsible for collecting sensor data from
smartphones, as well as the actual bids. It was implemented
as a plugin to the open-source mobile sensing framework
AWARE [12]. AWARE enables collecting sensor data from
Android-powered smartphones, and runtime
synchronisation of the collected data to a server database.
The following data was collected from each participant:
Bids: user-indicated bids (in EUR) for draining 10
percent units of the currently remaining
smartphone battery life.
Battery level: battery level (percentage), power
related events (phone shutting down, rebooting),
and user-driven contexts (initiating a charge and
unplugging the device).
Location: coarse network-based location (i.e., no
GPS), collected every 5 minutes.
Application Usage: application launches (name of
application and timestamp), starting and stopping
of background services, notifications and crashes.
Screen Status: the phone’s screen status, such as
turning the screen off and on, or locking the
screen.
Battery bids from participants were collected using the
Experience Sampling Method [24] provided by AWARE.
Using an interval contingent trigger, participants received
one alert every hour during the auction days. Figure 1 (left)
depicts one of these alerts asking how much money the
participant wants in exchange for exactly 10% units of their
battery life. The popup was not triggered if the user had less
than 10% units of battery left, as it would not be possible to
“sell” as much battery at the time.
Figure 1. The user interface of the auction. From left: asking
for a bid; notifying the participant about a win; notifying the
participant about a successfully completed drain.
Battery drainer
A key characteristic of our experiment is that participants
did not bid hypothetically: we actually drained their
smartphone’s battery. To ensure the draining took place, we
built a background service that ran on participants’
smartphones. The server would notify our software of
potential auction wins and inform the winner (Figure 1,
middle). The software kept track of the device’s battery
level until 10% units were ceaselessly drained, and then
notified the user (Figure 1, right) and the server of the
successful battery draining.
To enable participants to potentially bid in every hourly
auction, we had to ensure that draining of 10% battery units
could be achieved within one hour. We conducted a series
of maximum battery draining tests to assess different
methods of battery depletion. For our tests, we used a
reference handset model: Motorola XT1032 (Moto G) with
a Qualcomm MSM8226 Snapdragon 400 processor, 1 GHz
RAM and Non-removable Li-Ion 2070 mAh battery. We
tested multiple draining approaches (Figure 2) that rely on
continuously activating commonly available hardware and
performing computationally intensive tasks, as follows:
Camera: activate the camera of the phone, without
storing images;
Microphone: activate audio listening on the
microphone, without storing audio;
Sensors (environment): activate all available
environmental sensors, such as accelerometer,
temperature sensor, gravity sensor, gyroscope,
light sensor, linear accelerometer, magnetometer,
pressure sensor, proximity sensor, relative
humidity sensor, and rotation vector sensor.
Sensors’ availability may differ with handset
models;
GPS: activate GPS location requests;
Flash: activate the flashlight of the phone. Due to
the camera API, this also activates the camera;
Computational processing: compute
exponentiations with large integers;
All of the above: all aforementioned battery
draining approaches.
Figure 2. For each battery draining method we show how the
battery level (y-axis) depletes over time (x-axis, logarithmic).
The tests were performed with the phone in an idle state,
i.e., not running applications, and the display turned off.
Our time-to-drain (TTD) results are therefore, upper bound
since naturalistic device usage and any running application
will accelerate the depletion of the battery. Our results
found that the camera approach was the slowest, requiring
3.5 days to deplete the whole battery (or 8.4 hours for
10%), while all methods combined required 30 minutes for
10% units of battery. For our study we decided to use the
flash method, which took about 35 minutes to deplete 10%
battery units, and did not affect significantly normal device
usage – using all methods simultaneously had a noticeable
effect on the device’s performance. Our participants would
have ample time to drain their battery between two
consecutive auctions, without drawbacks in device
performance.
STUDY
Pilot
We conducted a brief 3-day pilot with five colleagues. The
participants were rewarded with a movie ticket plus any
money they would win from the auction. To end the pilot,
we conducted semi-structured interviews to discuss their
bidding experience.
First, the lack of a persistent application interface was
confusing to 2 participants, who felt unsure whether the
auction actually was happening, and thus did not feel
comfortable placing bids. As a result, in the main study we
explained this more clearly in our instructions, and assured
that our software was running in the background. We made
a conscious decision against a constantly accessible and/or
visible interface for bidding, as we wanted to minimise the
disruption to participant’s daily routine and usage patterns.
Second, the bidding notifications were not disruptive
enough, causing them to remain unnoticed and expire on
their own. This led to not having enough bids in many of
the auctions. To increase the popup’s noticeability, we
added an auditory cue to the notification (overriding the
phone’s current default notification, which could be vibrate-
only or silent). After these changes we conducted our main
study, which we will discuss next.
Participants and Rewards
Our main study had 22 participants (5 females, 17 males,
average age 24.3, SD=3.0) recruited from University of
Oulu in Finland, using email lists and posters placed at the
campus. The requirements for participating were i) own an
Internet-connected Android smartphone to use in the study
– we wanted participants to use their own phones, ii) bid
daily at least four times, and iii) participate in a workshop
including a semi-structured interview at the end of the
study. The mean hourly salary in Finland is above 18 EUR.
Therefore, and to comply with the country’s work
guidelines, upon completing the study each of our
participants was compensated with 50 EUR plus the money
won in the auction. We estimated the 1-on-1 briefing,
participating in the auction, and the post-study workshop
together amounting to at least 3 hours per participant (3 *
18 EUR = 54 EUR).
Participation
The study began with a 2-day enrolment phase, followed by
8 days of auctions. The participants were incrementally
enrolled into the study during this 2-day enrolment phase,
to allow us to individually explain the study details. During
the enrolment phase, the auction system was not active,
although participants could place non-winning bids to get
acquainted with the system. The data collected during this
period was excluded from analysis. All participants
participated simultaneously during the 8 days of auctions.
Given 13 auctions on any full day (from 10:00 to 22:00),
the study offered a total of 102 bid opportunities (on the
first day, bidding began exceptionally at noon 12).
During the 8 days of auctions, we employed a motivational
strategy to elicit sustained participation: we sent daily
motivational messages to all bidders, using our software’s
popup functionality. The messages leveraged two
previously studied psychological motivators: perceived
self-efficacy [2] (e.g., “Your participation has been
awesome so far! Please keep bidding whenever you can.”)
and causal importance [41] (e.g., “Because of your help, we
are able to conduct a much better study! Keep bidding!”).
Both of these motivation types have been found effective in
eliciting sustained participation in a similar mobile data
collection study [17].
Finally, we invited the participants to an open-ended
discussion about the study and issues around the value they
assign to battery life. We organised two workshops to
accommodate everyone’s availability. Each participant took
part in only one of the workshops. In the workshops, one
researcher led the discussion and showed statistics from the
experiment, focusing on issues such as auction and bidding
strategies, themselves from data patterns, battery valuation
contexts, and the mental models around smartphone battery
in general. Two additional researchers scribed the
discussions and collected further insights on the issues
directly from the participants’ observations. A short data
collection form was also distributed, containing questions
on demographic data, self-perceived truthfulness of the
bids, and a free textual feedback item.
RESULTS
Data
In the end, we had bidding data from 20/22 smartphones.
Two participants’ data was discarded due to data quality
issues (software-phone incompatibilities led to sporadic
data collection). We expected to collect a maximum of
2040 bids if the 20 active participants responded to all bid
notifications. Ultimately, we collected 1211 bids. In
addition to these bids, participants cancelled the bid
notification 120 times, i.e., they actively decided not to bid
during that bidding round. Thus the total amount of user
interactions to bid notifications was 1331. In addition, 342
bid requests expired on their own (i.e., no user interaction
was registered). Finally, 367 bids are missing because
participants’ phones were either disconnected or turned off.
During the 8 days of the study we recorded 795,374 state
changes in the battery levels of our participants around the
clock (i.e., 24-hours per day), and 480 charging events (i.e.,
participants charging their phones). We collected 14,852
location events, 34,231 screen state changes, and 221,808
application-related events. Finally, we summarised the key
insights from the workshops. We defer our workshop
findings to this paper’s discussion section.
Analysis
We analyse participants’ battery management patterns,
bidding behaviour, and derive a model to observe human
behaviour through their application use in different battery
contexts. Our initial analysis of location and screen status
data did not yield interesting insights in the scope of this
paper.
Battery management
Figure 3 depicts the Probability Density Function (density
plot) of participants’ battery levels during the study (right:
auction hours only, left: 24-hour basis). On average, the
aggregated battery level of smartphone users during any
hour of the day is seldom less than 65%, also reported in
[10]. Here we noticed that participants very frequently
allowed their battery levels to deplete much lower than this,
similar to what has been reported in [11]. This is not a
surprise, but rather indicates the auction being successful in
its purpose. We specifically instructed participants not to
charge their phones during bidding hours, as they would not
be allowed to bid otherwise.
Figure 3. Left: battery level fluctuation during the entire study
on a 24-hour basis. Right: battery levels during auction hours
only (from 10:00 to 22:00).
To illustrate the diversity in participants’ battery
management behaviour, we show the density plot for
participants P1, P2, and P3 in Figure 4 (during auction
hours). We notice that P1 seldom had low battery levels,
indicating very frequent device charges. P1 did not win the
auction even once, since participants that charged their
device during auction hours were not eligible to win. In
contrast, P2 and P3 spent considerable time on low battery
levels, and for example P3 won 8 auctions.
Figure 4. Different battery management behaviours by three
different participants during the auction hours.
In Figure 5 we show for P1, P2, P3 the mean battery level
per hour of day. We notice that P2 and P3 charged during
the night and gradually discharged their battery during work
hours [8h-16h]. On the other hand, P1 discharged their
phone during the night, began charging during work hours,
and in the afternoon began discharging again.
Figure 5. The aggregated battery level of all participants stays
high throughout the day, but individual participants’ battery
levels vary a lot.
We also calculated the aggregated battery level across all
participants (“mean” in Figure 5), which gradually declines
during working hours. The peak hours when the battery
level of the entire population is highest (81%) are between
05:00 and 07:00 while the lowest (61%) hours are at night:
22:00 - 24:00. These findings are in line with previous work
[10], with few exceptions.
Bids
Based on our workshop findings (discussed later), we
removed 9 bids above 50 EUR as evident outliers. The
mean bid across all participants was 2.22 EUR (SD=4.3),
and the median bid 0.70 EUR. The high standard deviation
indicates participants altered their battery valuation, which
in our experiment is desirable as it denotes price elasticity.
In Figure 6 we show the density plot of all bids valued less
than 20 EUR (upper limit 20 for a cleaner visualisation,
there were not many bids over 20).
Figure 6. A density plot of the placed bids. Y-axis denotes
probabilities, while x-axis represents bid values in EUR.
We find most bids were worth less than 1 EUR, with spikes
at round values ranging between 1 to 5 EUR. During the
study, 18/20 participants won at least one auction. On
average, participants won approximately 6 times (5.67,
SD=5.36). And, as we expected, draining 10% units of
battery took 23 minutes on average. Two participants
constantly bid very low and thus won exceptionally many
times – 18 and 19 wins, respectively. Finally, the winning
bids summed up to 17.14 EUR, with a mean winning bid of
0.17 EUR (min=0.01 EUR, max=2.30 EUR).
For every bid we received from each participant, we had a
record of the corresponding battery level at the bidding
moment. Analysing the correlation between bids and
battery levels, we found a weak reverse correlation
(Pearson product-moment, r=.-16, p<.05). In other words,
as the battery level decreases, the value increases. This
finding suggests that participants valued battery higher as
their devices’ battery depleted.
Next, we binned battery levels into 9 bins, each
representing a 10% unit range: 10-20, 20-30, 30-40,…, and
90-100. For each battery level bin we can calculate the
mean bid value (Figure 7). The trend is linear whereby bid
values increase as battery level drops, and especially we
observe a sharp increase in the final bin (20-10% battery).
A post-hoc Tukey HSD test showed that the lowest battery
level bin (20-10) differed significantly (p<.05) from bins
50-60 (diff = 2.25), 70-80 (diff = 3.05), 80-90 (diff = 3.00)
and 90-100 (diff = 2.65).
Figure 7. Mean, median, and the percentage of bids placed per
battery level categories. As battery level decreases, bids
increase.
The same graph also shows how often we received bids
from each battery level bin (in percent). We observe that
while participants mostly bid when battery levels were
between 40% and 90% we still received a fair amount of
bids even when battery levels were lower.
To illustrate the differences between participants’ bidding
strategies, we show the density plots for the bidding by P4,
P5, and P6 in Figure 8. We observe that P4 tended to bid at
integer values (2, 3, etc.), P5 tended to bid one order of
magnitude higher (10, 20), while P6 bid one order of
magnitude lower (0.1, 0.5). Finally, we examine the mean
bid across all participants during the auction in Figure 9.
The figure aggregates all bids from all auction days. The
bidding value in general increases as the day progresses.
Figure 8. Different bidding behaviours by three different
participants during the study.
Figure 9. The mean bid per auction hour (continuous line) and
the bid trend line (dashed line).
Using the monetary model: evaluating apps
We demonstrate the feasibility of using the monetary
valuation of battery life as a lens to quantify user behaviour.
We quantified the value that users place on individual
applications by considering “when” participants run them,
in terms of how much battery is left. Certain applications
are more battery-intensive than others. Our findings and
data on battery valuation demonstrate that users clearly take
application-battery use into account, especially when
battery is running low. This allows us to quantify the value
that users place on specific applications with our monetary
estimation for battery life.
We collected each application launch, and the amount of
battery left at that moment on the device. We then
generated a density plot for each individual app across all
participants. The curve indicates the application’s launch
frequency, or probability (y-axis) on a given battery level
(x-axis, continuous from 0 to 100). This is a highly similar
approach to what Jones et al. [21] use for analysing
smartphone application patterns. Only in our case, the
context variable is battery life. Here, this provides detailed
insight into how frequently the participants used different
applications at varying battery levels. In Table 1 we
summarise the density curves for the most popular
applications in our dataset.
Observing the curves, we notice how some apps are
launched more often on high battery levels (e.g., YouTube),
while other apps are launched regardless of the current
battery left (e.g., Chrome), or on lower power levels (e.g.,
Instagram). Not surprising, regardless of the application,
their use drops close to zero when battery levels are very
low. This is indicative of the high value users associate with
the last remaining battery on their phones.
Application
Frequency
EMV
Launch frequency per
battery levels 0-100
Viber
992
2.54
Chrome
904
2.39
Instagram
271
2.36
WhatsApp
2649
2.36
Facebook
1552
2.33
Spotify
254
2.31
Twitter
212
2.11
Gmail
279
2.09
YouTube
106
2.03
Table 1. For the most-often used applications in our study we
calculate the number of times it was launched in our study
(frequency), and the Expected Monetary Value (EMV) that
the user population implicitly associates with that app. The
axes of the density plots are probability (y-axis: [0,1]), and
battery level (x-axis: [0,100]).
Next, for each application we consider its probability of
being launched at the 90-100%% battery level bin, 80-90%,
and so on. For each such bin we have an associated
monetary value extracted from Figure 7, which effectively
places higher value on low-power bins.
Because we did not collect bids when battery level was less
than 10%, we used the bid value from the second-lowest
bin (10-20%) as the bid value of the lowest bin (0-10%).
For each app we multiply the probability of each battery
level bin with the monetary value of each bin. Summing
those 10 products we obtain a measurement of “relative
importance”. This metric is in EUR, and is typically called
the Expected Monetary Value (EMV), shown in Table 1.
This metric is used to evaluate the potential payoffs for a
set of possible outcomes, and thus can be used to compare
the relative value of different actions [28].
Workshops
We held two workshops after the 8 auction days were
concluded. Prior to the workshops we generated anonymous
graphs and statistics regarding the study, which we used in
slides to drive the discussion. The workshops revealed that
the auctions were very successful in making participants
think about battery value in new ways in their everyday
context. Analysis of the (anonymous) questionnaires we
collected during the workshops reveals that the self-
assessed honesty of the bids for battery was high: 4.09/5.00
(SD = 0.67). Finally, although we use most of the workshop
results in supporting the discussion, we note three key
findings here:
1. Because chargers are ubiquitous in most
environments, battery life is not really considered
as a “real problem.” Only when battery level
becomes very low, its perceived value increases
rapidly.
2. The value judgement regarding battery is highly
context-dependent.
3. While the auctions were determined to be truth-
telling, the renewable nature of smartphone battery
imposes challenges for conducting similar studies
in the future.
We inquired about a handful of extremely high bids placed
during the auction, such as 10,000 EUR. Our participants
revealed that such bids were submitted to “play with the
system” or as an attempt to get lucky, even if the
participants seemed to be aware they would most likely not
win with such bids. Based on the discussions, we set the
upper cut-off limit in data analysis to 50 EUR, meaning that
all higher bids were omitted from the analysis in this article.
On the other extreme, we discussed the reasons behind the
extremely low bids (0.001 EUR and similar). The bids were
placed either to test if the auction actually works, or to
maximise the chances of winning a round, where
participants reported not caring about battery life at all.
However, typically after placing a very low bid and
winning a round, participants started bidding honestly, as
losing 10% units of battery for, say, 0.0001 EUR was
perceived highly unpleasant. Participants also noted that
“there was always someone bidding even lower”, so they
just started bidding honestly.
We did not find out any of the participants knowing each
other prior to the study, although we naturally cannot
completely rule out this possibility. Further, the participants
did not seem to collaborate in their bids during the study.
DISCUSSION
A recent BBC article claims that permanently-powered
smartphones are “a necessity in a world where more of us
suffer from nomophobia, also known as smartphone
separation anxiety” [20]. Researchers have also discussed
how users experience even heavy anxiety when deprived of
their smartphone [8]. So, it is no wonder that battery life –
or rather the lack of it – has repeatedly been framed as a
major challenge for smartphones: the culprit for
smartphones dying on their loyal users. Even so, research
has overlooked the assessment of battery life from a value-
driven perspective, and it has been assumed that battery is
simply a valuable resource. Reflecting on our experiment,
we discuss battery life, its perceived value in different
contexts, and how the economic model we developed is
useful in analysing user behaviour and smartphone
applications. First, however, it is important to discuss the
auction itself.
Reverse Second-price Auction for Measuring Battery
Valuation
Previous studies have auctioned personal information [34]
or web browsing habits [6] using reverse second-price
auctions. This mechanism produces honest and truthful
results, coinciding with the theoretical assessment for this
mechanism: it “makes sense to bid your true value even if
other bidders are overbidding, underbidding, colluding, or
behaving in other unpredictable ways” [9]. The duration of
our study (8 days) is admittedly shorter than in some of the
most related previous studies (12 days in [4] and 6 weeks in
[34]). However, in our auction the data collection frequency
is higher (13 times per day vs. 1~4 times per day in [4,34]),
overall yielding a higher number of entries per participant
(average of 60.6 bids). Thus, we feel the shorter duration of
our study is sufficiently balanced by the richer data
collection.
While the collected data does show rationale and expected
differences in battery valuation per different battery levels,
we identified two key challenges with an auction: bid
honesty, and bid strategy. For instance, although the best
strategy in this type of auction is bidding honestly [9,25], in
the workshops many respondents confessed bidding very
low during the first rounds of auction. This was either to
verify everything working correctly, or attempting to
simply win without caring about bid honesty: “I started by
bidding really low, just to check that everything is working.
And after that I started to bid more realistically…” or
“Even before the study I was convinced there was no
auction or at least real money involved”. The latter was
voiced by a participant bidding extremely low for a long
time in the auction just to win many rounds, i.e., very
similar behaviour to P6 in Figure 8.
Second, the bidding context had an impact on some of the
participants: “Time of day matters much more. Usually at
night, closer at night, battery is cheaper because I know it
is charging time” and “Time of the day has a big impact,
much more than the battery level. Close to night, we know
that we can charge it very soon and know that will be home,
safe.” Surprisingly, when aggregated across all participants,
and depicted in Figure 9, bids did not significantly decline
towards the end of the day, but in fact slightly increased.
This suggests that personal bidding strategies were being
employed by participants. Other real-world contexts could,
in theory, affect bidding strategies and lead to adaptive
behaviour right after winning a bid.
These comments reveal an inherent and fascinating
challenge to studying battery value using an auction
process: the renewable nature of battery life. Users charge
their devices whenever convenient for them, however
following a preferred charging schedule routine [11].
Despite participants indicating that they were truthful in
their bidding (4.09 on 5-point Likert scale), in many cases
battery was perceived as an endless resource due to ample
charging possibilities: e.g., participants could bid low (to
“game“ the auction) when they knew the next opportunity
to charge is near, e.g., at the end of the day. In a sense they
felt they would gain “something for nothing” if they won
the round in such a context. This is a crucial difference with
previous studies examining the value of data, such as
location [18], personal information [34], or web browsing
data [6]. By employing an auction to improve our
understanding of battery value, we also learned more about
how its value affects how the participants’ spend their
device time, i.e., application usage.
Smartphone Battery Valuation
As expected, the battery value rose as battery level went
down (r=-.16). However, only during the very lowest
battery levels (below 20%) did this value substantially
increased. Although we did not collect bids in the last
battery level bin (below 10%), the workshops discussions
revealed that the battery value becomes very high on those
occasions. For example, one participant commented: “I
stopped bidding when the battery value became infinite!”
The density plots depicting launch trends per battery
conditions in Table 1 show that users drastically reduce
application usage as the battery level nears zero. In other
words, the value of the last drops of battery is perceived
much higher than the value a single application can deliver
at the time.
Battery value is also dependent on location, and mobility
context. When next to charging facilities, typically when
someone is at home or at the office, battery loses its value:
“When I am next to a charger, I don't really care". The
same was noted by many of our participants, indicating that
the renewable nature of battery greatly deteriorates its
monetary value when charging opportunities are near. On
the other hand, mobility has the opposite effect on battery
value, despite the current available level: “When traveling, I
did not know when I can charge, and when I need the
phone. So I bid really high” or “In festivals or traveling,
then it's a massive problem”, in reference to the problem of
battery potentially running out in the future. Indeed
festivals were reported as a special case in the previously
mentioned BBC article [20], and portable chargers for
festivals is now a business case [37]. When at airports, it is
now common to find travellers on the lookout for elusive
power sockets [14]. The racks of power outlets for visitors
are seen as a mechanism to offer highly sought value to
customers in a situation where charging the device is not
possible in the foreseeable near future. Based on our study,
we find that smartphone battery becomes a real concern
only when the battery level is low, since that is when it
seems to be valued the most.
Resource-sharing Applications
Our monetary assessment of battery value provides insight
for fairer resource sharing amongst users. Several use cases
for resource sharing or donating have been proposed in
literature. For instance, Pering et al. [31] envision how a set
of mobile devices, with no need for additional hardware,
can form a local device conglomerate that shares radio
interfaces in order to save battery life. They highlight
significant energy savings introduced by such a scheme,
and there can also be financial benefits for mobile users. An
example is roaming abroad. Roaming is typically
expensive, and a sharing scheme enables nearby devices to
connect online via a local gateway device with fixed rate
bandwidth -- either for free, for a small monetary fee, or
even employing some other type of compensation scheme.
The introduced scenario naturally takes a toll on the
gateway device’s battery life. In our vision the gateway can
passively run in the background and offer sporadic
connectivity for nearby sensors equipped with low-range,
low energy consumption devices. The battery depletion
experienced by the mobile user is compensable by a micro-
payment scheme. The question is, then, how much should
the users be paid for their battery, if at all, by whom, and
under which circumstances?
Economic theory suggests that not all money is equal, or “a
dollar is not a dollar”: people value and earmark money
from different sources in different ways [39]. Similarly, in a
crowdsourcing context it has been recently shown that some
prefer receiving tangible goods rather than money for their
work [16,19]. Using the resource sharing scenario to frame
the discussion, we asked the workshop participants how
they would feel about sharing or donating battery to friends,
strangers, or institutions. Most participants indicated they
would be willing to part from their battery for altruistic
purposes, to share it with friends for free. Social charging
infrastructures [13] are created symbiotically among
friends, where phone chargers are placed, shared, and
expected to be available during social events.
Moreover, some participants felt uncomfortable to ask for
money for battery: “I would rather say no for a friend then
ask for money!”. One participant also wished for
“something else” than money in exchange for battery,
similar to [19]: “Not money from friends...give me battery, I
give you lunch....So something tangible instead of money”.
This presents a welcome opportunity: while a micro-
payment mechanism can be considered, there may also be
potential for exchange of small goods.
The Economics of HCI and User Behaviour
Because the relationship between users and their
smartphones is complex and evolving, quantifying the value
that users place on these devices is challenging. For
example, a recent study found smartphone users suffer from
severe psychological and physiological effects, such as
elevated heart rate, when prevented from interacting with
their devices [8]. That study suggests phones are an
“extension to self,” meaning that when the phone is taken
away, the user loses a very part of oneself. Park et al. [29]
report that some of the aspects of smartphones’ value relate
to their convenience (e.g., checking the weather,
navigating) and pleasure (e.g., listening to music, watching
movies).
As we show in our analysis, user interactions can be valued
by considering the battery level at the given moments. The
model we empirically derive places greater value on actions
occurring when on low battery, and in this manner it
enables us to estimate the expected value of other functions:
maps, music, movies, and communication functions can be
systematically valued in this manner.
Effectively, we can consider the battery-price plot (Figure
7) as a demand curve: price goes up as quantity goes down,
due to battery scarcity. However, what is interesting in the
case of smartphones is that users face intermittent scarcity,
since while they charge their phone they effectively have an
unlimited energy (i.e., value) supply, but to access that they
need to give up their mobility [10]: one typically does not
charge their device while they are moving. Due to these two
constraints (intermittent scarcity and energy-mobility
tradeoff) it may be possible to estimate how much value
users place on their mobility, or to be more precise their
potential for mobility. For instance, users who charge their
phone when their battery is at 90% would be considered to
value their mobility less than users who charge their battery
only once it reaches very low levels. While this assessment
is not profound, our work provides the tools to quantify
such behaviour in a systematic manner, using our metric
that reflects Expected Monetary Value [28].
More broadly, our work applies to HCI research in general.
The early days of HCI focused on the benefits of usability
by arguing that improved usability saved time, which in
turn saved on salary costs. The archetypical example would
be call-centres, where improved user interfaces would
reduce operators’ time, and thus reduce costs. The metrics
popularised in that era were largely linked to task
completion time and error rate, and ultimately “time was
money” on such desktop systems. However, on
smartphones “battery is money”: without battery all
functionality becomes unavailable. Thus, studying user
behaviour from an energy aspect (rather than time & error
performance) may present a fruitful avenue to explore for
further research on smartphones. Because every user
instance of interaction depletes battery, our approach can
systematically quantify user behaviour by considering its
energy impact.
Limitations
An economic model such as ours is always a simplified
description of reality, designed to yield testable hypotheses
about behaviour. An important feature of an economic
model is that it is necessarily subjective in design because
there are no objective measures of economic outcomes.
In our analysis we cannot reliably analyse certain factors
like hour of day, as the experimental design is not suitable
for this. Also, the fact that we asked individuals to avoid
charging during bidding hours resulted in atypical
behaviour. However, without such constraints it would have
been very hard to gauge battery value low battery levels,
since users naturally tend to avoid those [10].
Another limitation is that our software did not take into
account the winning bidder’s current energy expenditure.
For example, if a participant was watching a movie when
winning, the amount drained was likely not 10% units of
the entire battery, but slightly less. The software simply
drained until the level was 10% units lower than at the time
of starting the drain. Again, we argue that the high amount
of bidding rounds compensates this.
Finally, we acknowledge that the results likely depend on
cultural and societal backgrounds, demographic
characteristics, and the personality of participants. Despite
the limitations, the framework for measuring user value of
smartphone resources is applicable to other populations.
CONCLUSION
One of the most prominent contextual elements of
smartphones, battery life, has not been quantified from the
perspective of perceived monetary value. In this paper, we
presented the first auction-based study aiming to assess the
value users assign to their remaining battery life in the
context of daily life. We also discovered the renewable
nature of battery to impose challenges for the de-facto
auction protocol (reverse second-price auction). Overall, we
observed that users place different values on battery
depending on the current level of battery, and social and
mobility contexts.
Our study provides a look into how monetary battery value
can be quantified. It offers a replicable method to examine
applications, features, and user behaviour, based on their
use patterns across different battery levels. Future work
may expand this assessment for other mobile resources,
such as bandwidth or storage space.
ACKNOWLEDGEMENTS
This work is partially funded by the Academy of Finland
(Grants 276786-AWARE, 285062-iCYCLE, 286386-
CPDSS, 285459-iSCIENCE), and the European
Commission (Grants PCIG11-GA-2012-322138 and
645706-GRAGE).
REFERENCES
1. Kumaripaba Athukorala, Eemil Lagerspetz, Maria H.
von Kügelgen, Antti Jylhä, Adam J. Oliner, Sasu
Tarkoma and Giulio Jacucci. 2014. How Carat Affects
User Behavior: Implications for Mobile Battery
Awareness Applications. In CHI'14.
http://dx.doi.org/http://dx.doi.org/10.1145/2556288.255
7271
2. Albert Bandura. 1992. Exercise of personal agency
through the self-efficacy mechanism. APA PsycNET.
3. Nilanjan Banerjee, Ahmad Rahmati, Mark D. Corner,
Sami Rollins and Lin Zhong. 2007. Users and Batteries:
Interactions and Adaptive Energy Management in
Mobile Systems. In UbiComp 2007: Ubiquitous
Computing, Springer Berlin Heidelberg, 217-234.
http://dx.doi.org/10.1007/978-3-540-74853-3_13
4. Omer Barak, Gabriella Cohen, Alla Gazit and Eran
Toch. 2013. The Price is Right?: Economic Value of
Location Sharing. In Proceedings of the 2013 ACM
Conference on Pervasive and Ubiquitous Computing
Adjunct Publication, ACM, 891-900.
http://dx.doi.org/10.1145/2494091.2497343
5. Raffaele Bolla, Raees Khan, Xavier Parra and Manuela
Repetto. 2014. Improving Smartphones Battery Life by
Reducing Energy Waste of Background Applications. In
International Conference on Next Generation Mobile
Apps, Services and Technologies, IEEE, 123-130.
http://dx.doi.org/10.1109/NGMAST.2014.10
6. Juan P. Carrascal, Christopher Riederer, Vijay Erramilli,
Mauro Cherubini and Rodrigo de Oliveira. 2013. Your
Browsing Behavior for a Big Mac: Economics of
Personal Information Online. In International
Conference on World Wide Web, International World
Wide Web Conferences Steering Committee, 189-200.
7. CES 2013: Share Battery Power Between Mobile
Devices - IEEE Spectrum. Retrieved 13/08/2015 from
http://spectrum.ieee.org/tech-talk/consumer-
electronics/portable-devices/mobile-devices-share-
everything
8. Russell B. Clayton, Glenn Leshner and Anthony
Almond. 2015. The Extended iSelf: The Impact of
iPhone Separation on Cognition, Emotion, and
Physiology. J Comput-Mediat Comm 20, 2: 119-135.
http://dx.doi.org/10.1111/jcc4.12109
9. Easley David and Kleinberg Jon. 2010. Networks,
Crowds, and Markets: Reasoning About a Highly
Connected World. Cambridge University Press.
10. Denzil Ferreira, Anind K. Dey and Vassilis Kostakos.
2011. Understanding human-smartphone concerns: a
study of battery life. In International Conference on
Pervasive Computing, Springer-Verlag, 19-33.
http://dx.doi.org/10.1007/978-3-642-21726-5_2
11. Denzil Ferreira, Eija Ferreira, Jorge Goncalves, Vassilis
Kostakos and Anind K. Dey. 2013. Revisiting Human-
Battery Interaction with an Interactive Battery Interface.
In International Joint Conference on Pervasive and
Ubiquitous Computing , ACM, 563-572.
http://dx.doi.org/10.1145/2493432.2493465
12. Denzil Ferreira, Vassilis Kostakos and Anind K. Dey.
2015. AWARE: mobile context instrumentation
framework. Frontiers in ICT 2, 6: 1-9.
http://dx.doi.org/10.3389/fict.2015.00006
13. Pedro Ferreira, Moira McGregor and Airi Lampinen.
2015. Caring for Batteries: Maintaining Infrastructures
and Mobile Social Contexts. In Proceedings of the 17th
International Conference on Human-Computer
Interaction with Mobile Devices and Services, ACM,
383-392. http://dx.doi.org/10.1145/2785830.2785864
14. Finding those elusive power sockets at airports.
Retrieved 16/09/2015 from
https://www.kayak.co.uk/news/finding-elusive-power-
sockets-airports/
15. Huber Flores, Pan Hui, Sasu Tarkoma, Yong Li, Satish
Srirama and Rajkumar Buyya. 2015. Mobile code
offloading: from concept to practice and beyond.
Communications Magazine, IEEE 53, 3: 80-88.
http://dx.doi.org/10.1109/MCOM.2015.7060486
16. Jorge Goncalves. 2015. Situated Crowdsourcing:
Feasibility, Performance and Behaviours. Ph.D
Dissertation.
17. Jorge Goncalves, Vassilis Kostakos, Evangelos
Karapanos, Mary Barreto, Tiago Camacho, Anthony
Tomasic and John Zimmerman. 2014. Citizen
Motivation on the Go: The Role of Psychological
Empowerment. Interacting with Computers 26, 3: 196-
207. http://dx.doi.org/10.1093/iwc/iwt035
18. Jens Grossklags and Alessandro Acquisti. 2007. When
25 Cents is Too Much: An Experiment on Willingness-
To-Sell and Willingness-To-Protect Personal
Information. In WEIS.
19. Simo Hosio, Jorge Goncalves, Vili Lehdonvirta, Denzil
Ferreira and Vassilis Kostakos. 2014. Situated
Crowdsourcing Using a Market Model. In User
Interface Software and Technology, ACM, 55-64.
http://dx.doi.org/10.1145/2642918.2647362
20. How far will we go to charge our phones? - BBC News.
Retrieved 18/08/2015 from
http://www.bbc.com/news/magazine-33295211
21. Simon L. Jones, Denzil Ferreira, Simo Hosio, Jorge
Goncalves and Vassilis Kostakos. 2015. Revisitation
Analysis of Smartphone App Use. In International Joint
Conference on Pervasive and Ubiquitous Computing,
ACM, 1197-1208.
http://dx.doi.org/10.1145/2750858.2807542
22. Aman Kansal, Michel Goraczko and Feng Zhao. 2007.
Building a Sensor Network of Mobile Phones. In
Proceedings of the 6th International Conference on
Information Processing in Sensor Networks, ACM, 547-
548. http://dx.doi.org/10.1145/1236360.1236433
23. Nicholas D. Lane, Emiliano Miluzzo, Hong Lu, Daniel
Peebles, Tanzeem Choudhury and Andrew T. Campbell.
2010. A Survey of Mobile Phone Sensing.
Communications Magazine, IEEE 48, 9: 140-150.
http://dx.doi.org/10.1109/MCOM.2010.5560598
24. Reed Larson and Mihaly Csikszentmihalyi. 1983. The
Experience Sampling Method. In Flow and the
Foundations of Positive Psychology (eds.). Wiley
Jossey-Bass, San Francisco, 15, 41-56.
25. R. P. McAfee and John McMillan. 1987. Auctions and
bidding. Journal of economic literature: 699-738.
26. Sandra J. Milberg, Sandra J. Burke, H. J. Smith and
Ernest A. Kallman. 1995. Values, Personal Information
Privacy, and Regulatory Approaches. Commun. ACM
38, 12: 65-74. http://dx.doi.org/10.1145/219663.219683
27. Derek G. Murray, Eiko Yoneki, Jon Crowcroft and
Steven Hand. 2010. The Case for Crowd Computing. In
Proceedings of the Second ACM SIGCOMM Workshop
on Networking, Systems, and Applications on Mobile
Handhelds, ACM, 39-44.
http://dx.doi.org/10.1145/1851322.1851334
28. R. R. Officer and A. N. Halter. 1968. Utility Analysis in
a Practical Setting. American Journal of Agricultural
Economics 50, 2: 257-277.
http://dx.doi.org/10.2307/1237541
29. Jaehyun Park and Sung H. Han. 2013. Defining user
value: A case study of a smartphone. International
Journal of Industrial Ergonomics 43, 4: 274-282.
http://dx.doi.org/10.1016/j.ergon.2013.04.005
30. Kostas Pentikousis. 2010. In search of energy-efficient
mobile networking. Communications Magazine, IEEE
48, 1: 95-103.
http://dx.doi.org/10.1109/MCOM.2010.5394036
31. Trevor Pering, Yuvraj Agarwal, Rajesh Gupta and Roy
Want. 2006. CoolSpots: Reducing the Power
Consumption of Wireless Mobile Devices with Multiple
Radio Interfaces. In Proceedings of the 4th
International Conference on Mobile Systems,
Applications and Services, ACM, 220-232.
http://dx.doi.org/10.1145/1134680.1134704
32. Ahmad Rahmati, Angela Qian and Lin Zhong. 2007.
Understanding human-battery interaction on mobile
phones. In MobileHCI 2007, 265-272.
http://dx.doi.org/10.1145/1377999.1378017
33. Ahmad Rahmati and Lin Zhong. 2009. Human–battery
interaction on mobile phones. Pervasive and Mobile
Computing 5, 5: 465-477.
http://dx.doi.org/10.1016/j.pmcj.2008.08.003
34. Jacopo Staiano, Nuria Oliver, Bruno Lepri, Rodrigo de
Oliveira, Michele Caraviello and Nicu Sebe. 2014.
Money Walks: A Human-Centric Study on the
Economics of Personal Mobile Data. In UbiComp,
ACM. http://dx.doi.org/10.1145/2632048.2632074
35. Khai N. Truong, Julie A. Kientz, Timothy Sohn, Alyssa
Rosenzweig, Amanda Fonville and Tim Smith. 2010.
The Design and Evaluation of a Task-centered Battery
Interface. In Proceedings of the 12th ACM International
Conference on Ubiquitous Computing, ACM, 341-350.
http://dx.doi.org/10.1145/1864349.1864400
36. Narseo Vallina-Rodriguez, Christos Efstratiou, Geoffrey
Xie and Jon Crowcroft. 2011. Enabling Opportunistic
Resources Sharing on Mobile Operating Systems:
Benefits and Challenges. In Proceedings of the 3rd
ACM Workshop on Wireless of the Students, by the
Students, for the Students, ACM, 29-32.
http://dx.doi.org/10.1145/2030686.2030696
37. Volt. Retrieved 16/09/2015 from http://getvolt.dk/
38. Ranti Yulia Wardani and Sony Warsono. 2012. A cross-
cultural study on the value structure of mobile internet
usage: comparison between Indonesia and Japan.
Journal of Indonesian Economy and Business : JIEB.
27, 3: 355-372,420-421.
39. V. A. R. Zelizer. 1994. The Social Meaning of Money.
BasicBooks.
40. Lide Zhang, Robert P. Dick, Z. M. Mao, Zhaoguang
Wang and Ann Arbor. 2010. Accurate Online Power
Estimation and Automatic Battery Behavior Based
Power Model Generation for Smartphones. In
CODES/ISSS '10, 105-114.
41. Marc A. Zimmerman and Julian Rappaport. 1988.
Citizen participation, perceived control, and
psychological empowerment. American journal of
community psychology 16, 5: 725-750.
