Docstoc

Studies in One-Handed Mobile Design

Document Sample
Studies in One-Handed Mobile Design Powered By Docstoc
					                     Studies in One-Handed Mobile Design:
                            Habit, Desire and Agility
    Amy K. Karlson, Benjamin B. Bederson                                      Jose L. Contreras-Vidal
        Human-Computer Interaction Lab                                 Cognitive-Motor Behavior Laboratory
          Computer Science Department                                          Kinesiology Department
   Univ. of Maryland, College Park, MD 20742                         Univ. of Maryland, College Park, MD 20742
          {akk, bederson}@cs.umd.edu                                             pepeum@umd.edu
ABSTRACT                                                           to remain.
In this paper we explore single handed mobile computing.
Our approach is to understand how devices are currently            Much effort could be saved if there was an understanding of
used (habit), how users would prefer to interact with              these usability issues, but, to the best of our knowledge, no
devices (desire) and the relationship between device size          such empirical studies have been conducted. In this paper,
and thumb freedom (agility). We use a different                    we present the first of such studies. We are concerned with
methodology to study each factor: a field study, survey, and       three main questions: (1) when do users of handheld
empirical evaluation, respectively. We found that users are        devices use and prefer one hand vs. two hands? (2) which
already operating mobile phones with one hand. While the           surface regions on a handheld device are more easily
majority of users would like to perform personal                   accessed with the thumb? and (3) what types of gestures are
information management tasks with a single hand, current           easiest to make with a thumb on a handheld device?
mobile device designs do not well support this mode of             Answers to these questions have broad implications for the
interaction. Lastly, thumb mobility tends to decrease as           world of mobile devices, especially as devices are required
device size increases, but users generally can perform             to offer more and more functionality. There is currently a
localized movement well, even for large devices.                   growing demand for hybrid devices; consumers
                                                                   increasingly want a single device that combines both
Author Keywords                                                    telephone and information management features [10]. With
Thumb movement, mobile devices, one-handed design.                 expanding uses, it is likely that device ergonomics, which
                                                                   depend heavily on use scenarios, will become increasingly
ACM Classification Keywords                                        difficult to design well. Results from empirical studies such
H5.m. Information interfaces and presentation (e.g., HCI):         as ours can help provide the insights that will lead designers
Miscellaneous.                                                     to enjoyable, ergonomically-sound devices from the outset.
INTRODUCTION                                                       We answer these questions in three ways. We performed (1)
The handheld market is growing at a tremendous rate; the           a survey to record perceptions of and preferences for
technology is advancing rapidly and experts project that the       personal device use as well as (2) a field study to capture
annual mobile phone sales will top 1 billion by 2009 [10].         device types and interaction methods used in situ. We found
With this rapid growth have come vast differences in               that users often interact with mobile devices with only one
design. To meet customer demand for comfort and                    hand, and would do so more often if information
uniqueness, hardware designers continually introduce               management tasks better supported single handed use. To
smaller, sleeker profiles to the market. Unfortunately,            understand the biomechanical issues related to one handed
changes to the devices’ interfaces are not keeping up with         device interaction, we performed (3) an empirical study to
their form factor. At best, we must rely on evolution to           capture thumb movement over devices of varying form.
remove poor designs from the market and allow good ones            The results from this study suggest favorable regions and
                                                                   directions for movement, and offer evidence that device
                                                                   size impacts performance. Optimized target placement
                                                                   based on these findings has the potential benefit of
                                                                   increasing the speed of device interaction, at the same time
                                                                   protecting against repetitive stress injury.

                                                                   RELATED WORK
                                                                   As the “desktop in the pocket”, more mobile devices access
                                                                   similar data sets to desktop computers (i.e., web pages,
                                                                   contacts, appointments, email, etc.) often via familiar


                                                               1
interfaces. But the disparity in screen size between desktop        gestures, while AppLens interpreted thumb gestures for
and handheld computing has inspired many to explore                 indirectly controlling an input cursor [13]. To accommodate
novel software designs for data presentation, navigation,           a variety of hand geometry, both systems allowed gestures
and interaction on small screens, most notably for web              to be issued anywhere on the screen, but no attempt was
content. Page decomposition algorithms intended to reduce           made to optimize positions of on-screen targets. In a
or eliminate scrolling [8] have been combined with text             qualitative evaluation, participants responded favorably to
summarization for 1D (PowerBrowser [7]), 2D (WEST                   the one-handed designs but showed strong inclinations
[4]), and layout-preserving (Thumbnail Summaries [15])              toward direct tapping interaction rather than either gesture
presentation, while other approaches support user control           language, suggesting gestures may be best used to
over display content to maximize screen utility (Collapse-          complement direct thumb interaction techniques. This is
to-Zoom [2] and WebThumb [24]). However, the                        likely to hold true for gesture solutions based on device
fundamental issue such works address is data presentation           position and orientation [21, 23], although quantitative
for comprehension on small screens, which is not the focus          comparisons to thumb entry mechanisms have not been
of this paper.                                                      made.
Because text entry remains the input bottleneck for mobile          Scientists in the medical community have studied the
devices, many are working on improving this modality,               biomechanics of the thumb extensively for the purposes of
some even for one-handed use. Standard text input                   both reconstruction and rehabilitation. The static structure
hardware takes three general forms: miniaturized                    of the thumb has been understood for the last decade [1],
QWERTY keypads, touchscreens, and 12-key numeric                    but only now are scientists beginning to reliably quantify
keypads. Peripheral keypads for one-handed text entry are           the functional capabilities of the thumb. In the medical
also available (FrogPad, half QWERTY), but require                  domain, strength is the primary parameter used to assess
support from a hand, desk or lap, which violates the spirit         mechanical ability, and the influence of movement direction
of one-handed use as we define it. Software interfaces for          upon thumb strength has been established both at the
touchscreen devices primarily assume interaction with a             anatomical [20] and functional [5, 16] levels. However,
pen, featuring input targets that are much smaller than a           these strength-direction, or force vector measurements have
finger tip, and which are placed for maximal real estate            only been recorded for standard anatomical planes of
utilization. At best these design choices make thumb                movement. This excludes the plane parallel to the palm -
interaction error-prone (e.g., inaccurate targeting), and at        the one in which we are interested.
worst impossible (e.g., targets are out of reach). Therefore
                                                                    As a complement to force capabilities, others have looked
we see only the 12-key design supporting one-handed use
                                                                    at the extent of thumb movement. Kuo [14] has developed a
well, and algorithmic alternatives to improve upon the
                                                                    model for the maximal 3D workspace of the thumb, and in
original multitap text entry approach abound. Generally,
                                                                    our own field Hirotaka [11] has quantified an average for
speed improvements are sought by reducing the average
                                                                    thumb rotation angle. But the experimental conditions for
keytaps per word, either through predictive models such as
                                                                    these studies do not account for constraints that would be
T9, novel key-to-letter encodings [12], or chording [17], but
                                                                    imposed by holding objects of varying size, as would be the
not yet via ergonomic optimization. Predictive models [19,
                                                                    case when using a handheld device.
22], emulators [6] and peripherals [17] have been used to
compare approaches, but no formal studies have
                                                                    SURVEY
specifically targeted one-handed interaction on an actual
                                                                    One component of understanding mobile device
device.
                                                                    requirements is to capture current usage patterns. We
Hirotaka [11] redesigned the cell phone for improved                therefore polled users about how they currently use and
ergonomics and interaction, but the exercise was based on           would like to interact with devices. We designed a brief
center of mass and static range of thumb motion, rather than        survey to capture user perceptions of and motivations for
the model based on dynamics that we propose. Qualitative            their device usage patterns as well as thoughts on how
user feedback on his design was positive, but was not               devices of today fail to meet their needs.
evaluated formally. We are also aware of a single
commercial phone, the Identity (www.smartskins.com), that           Survey Method
has adapted the classic 12-key phone for improved one-              Our survey consisted of 18 questions presented on a single
handed input: a kidney bean shaped body with keys placed            web page which was accessed via an encrypted connection
above the screen. As intuitively appealing as these design          (SSL) from a department-owned server. An introductory
choices may be, the company offers no quantitative                  message informed potential participants of the goals of the
evidence for them.                                                  survey and assured them that their participation and
                                                                    responses would be kept anonymous. Notification that
Generalized one-handed interaction techniques have also             results would be posted for public access after the survey
been explored. Previously we designed two variations for a          was closed provided the only incentive for participation.
one-handed application manager for PDAs; LaunchTile was             Participants were solicited from a voluntary subscription
designed using thumb-sized targets and direct manipulation


                                                                2
mailing list for individuals interested in the activities of our
laboratory. In addition the solicitation was propagated to
one recipient’s personal mailing list, a medical informatics
mailing     list,    and     a    link     to    the     survey
(http://www.cs.umd.edu/hcil/mobile/survey/) was posted on
the two undergraduate course web pages.

Survey Measures
For each user, we collected age, sex and occupation
demographics. Users recorded all styles of PDAs and/or
phones they currently owned as well as whether they
typically travel with more than one device at a time. If so,
users were asked to complete the remainder of the survey
with only one device in mind - the one used for the majority
of information management tasks. We collected general
information about the primary device used, including usage
frequency, purchase criteria, input hardware, as well as
actual and preferred methods for text entry. We then asked
a variety of questions to understand when and why people
use one vs. two hands to operate a device. We asked users
to record the number of hands used (one and/or two) for                 Figure 1. (a) Hands currently used and (b) preferred for
eighteen typical mobile tasks, and then to specify the                  18 mobile tasks as a % of observed population. Tasks are
number of hands (one or two) they would prefer to use for               grouped by reading (light gray background), combined
each task. Three pairs of tasks were designed to distinguish            reading/writing (medium gray background), and writing
                                                                        (dark gray background).
between usage patterns for different tasks within the same
application, which we call read (email reading, calendar
                                                                       CURRENT USAGE: Of the 18 activities users typically
lookup, address lookup) vs. write (compose mail, enter an
address, enter an calendar appointment). Users then                    perform with devices, 9 were performed more often with
recorded the number of hands used for the majority of                  one hand, 6 more often with two hands, and 3 were about
device interaction and under what circumstances they chose             equal.
one option over the other. Finally, users were asked how               Figure 1a shows the division between these three groups.
many hands they would prefer to use for the majority of                Upon closer inspection, all of the “writing” activities were
interactions (including no preference), and allowed to enter           performed more often with two hands, and all “reading”
additional comments before submitting the survey.                      activities with one hand.

Survey Results                                                         Overall, 45% of participants stated they used one hand for
Two hundred twenty-nine participants (135 male)                        nearly all their device interactions, as opposed to only 19%
responded to the survey solicitation. Participant ages ranged          who responded similarly for two hands. When participants
from 18 to 70+ years with a median age of 38.5 years.                  use one hand, the majority (60%) perceive that they do so
                                                                       whenever the interface allows for it, rather than based on
DEVICES: The three most common devices owned were                      use scenario. It follows then that when participants use two
flip phones (52%), small candy bar phones (23%) and Palm               hands, the majority (64%) perceive they do so only when
devices without a keyboard (20%). We use the term “candy               the interface prevents it rather than based on use scenario.
bar phone” which is the industry term for a phone shaped
like a traditional candy bar – a fixed rectangle, typically            PREFERENCE: When asked the number of hands users
about 3 times longer than wide. Next, Palm devices with a              preferred to use while performing the same 18 tasks, one
keypad tied with Pocket PCs without a keypad (14%).                    hand was preferred overwhelmingly to two hands for all
Participants most often sited small size (67%) as a criterion          tasks (Figure 1b). The activities with the closest margin
for purchase, followed by low cost (51%), synchronization              between the number of participants who preferred one vs.
with desktop (42%) and color display (35%). For the                    two hands were playing games (13%) and composing email
devices participants used for personal information                     (16%). With one exception (gaming), the activities for
management, 54% had numeric keypads, 42% had touch                     which more than 20% of users stated a preference for two
screens, and 25% had QWERTY keypads.                                   hands were “writing” tasks, or those that required text
                                                                       entry: contact address entry, calendar entry, email writing,
TEXT ENTRY: Of the 88% of participants that perform text               text messaging and text entry. Based on these data, it is
entry, 37% use multi-tap text entry, 20% use touch screen              consistent that 66% of participants stated they would prefer
gestures (e.g., Graffiti), 20% use QWERTY keypads, and                 to use one hand for the majority of device interaction,
11% used T9 predictive text entry.                                     versus 9% who would prefer two hands for all interaction.


                                                                   3
Twenty-three percent did not have a preference and 6 users           interactions that included both the dialing and talking
did not respond.                                                     phases of use. All observations were performed
                                                                     anonymously without any interaction between the observer
While there were many interesting trends in user comments,
                                                                     and observed.
none strictly pertain to the focus of this paper. We refer the
reader to the permanent survey URL for those details.
                                                                     Confounding factors
                                                                     The choice of observation location may have biased our
Analysis of Survey
                                                                     results from those found in the general population because
Based only on current use patterns, there is no obvious
                                                                     due to the nature of travel, users may have been more likely
winner between one-handed and two handed use. Excluding
                                                                     to be: 1) carrying additional items; 2) standing or walking;
phone calls, the number of activities for which a majority of
                                                                     and 3) using a phone vs. PDA. Different environments,
responders use one hand (7) vs. two hands (6) is nearly
                                                                     information domains, populations, and scenarios will yield
balanced, and for many of these activities, the margin
                                                                     unique usage patterns (e.g., more seated users riding buses,
between the two groups is not strikingly large. But user
                                                                     more PDAs used by doctors, more text messaging by teens,
perceptions of why this is the case indicates that the
                                                                     etc.) Our goal is not to catalogue each possible
interface is the culprit, rather than preference; they use one
                                                                     combination, but to learn what we can from a single in-
hand if at all possible and only two hands when the
                                                                     transit scenario for which a variety of populations converge.
interface makes a task impossible to do single handedly.
Other than gaming, tasks involving text entry are the only
                                                                     Field Study Measures
ones for which users may be willing to use two hands,                For each user observed, we recorded sex, approximate age,
possibly because the mental attention required is great              and device type used: candy bar phone, flip phone,
enough that they cannot perform other activities                     Blackberry, or PDA. For phone use, we recorded the
simultaneously – not so for browsing activities. It seems            hand(s) used to dial (left, right or both) and the hand(s) used
clear that the onus is on interface designers to help bridge         to speak (left, right or both). We also noted whether users
the gap between how users would like to interact with                were carrying additional items, and their current activity
devices, and how they currently must do so.                          (selected from the mutually exclusive categories: walking,
                                                                     standing, or sitting).
FIELD STUDY
Reporting upon one’s own actions is problematic for a
                                                                     Field Study Results
number of reasons: time distorts memory, people can be               Only two users were observed operating devices other than
influenced by suggestion, and personal bias affects                  mobile phones, one used a PDA and the other a Blackberry.
perception. Recall of routine actions is particularly                Both were seated and using two hands. The remainder of
susceptible to error because those actions do not require            the discussion focuses on the 48 phone users, 62.5% of
deliberate attention: how many of us know for certain                which used a flip phone, 37.5% a candy bar phone. Overall,
which sock or shoe we typically put on first? As                     74% used one hand to dial. By activity, 65% of one handed
investigators, we must therefore assume some error in                users had a hand occupied, 54% were walking, 35% were
reports on behavior gathered in our survey. Although user            standing, and 11% were sitting. Figure 2 presents the
perceptions and preferences might justly be given primary            distribution of users within the population who used one vs.
weight in influencing future designs, habit is an indicator of       two hands for phone dialing, segmented by concurrent
preference and therefore of interaction patterns likely to be        activity (walking, standing, or sitting). The distribution of
transferred to new devices. To capture any disparity                 users engaging in the three activities reflects the airport
between reflective reporting and actual behavior, we                 scenario where many more people were walking or standing
conducted an in situ study of user interaction with mobile           than sitting. It is plain from Figure 2 that the relative
devices. The study targeted an airport environment for the           proportion of one handed to two handed dialers varied by
high potential of finding mobile device users and the
relatively easy access to observe people.

Field Study Method
We observed 50 travelers (27 male) at Baltimore
Washington International Airport’s main ticketing terminal
during a six hour period during peak holiday travel.
Because observation was limited to areas accessible to non-
ticketed passengers, seating options were limited. We
expected to observe the use of both PDAs and cell phones
since travelers are likely to be planning transportation or
rendezvous, catching up on work, and pursuing
entertainment. Because the majority of users talk on a cell            Figure 2. Number of hands used for phone dialing by
phone with one hand, we recorded only the cell phone                   activity, as a percentage of the observed population.


                                                                 4
activity; the vast majority of walkers dialed with one hand,         are the intended result of many Fitts’ Law studies, which
about two-thirds of standers dialed with one hand, and but           derive them experimentally via linear regression. We
more seated dialers used two hands. However, we also                 instead use Fitts’ Law as an input; no matter the constants,
noted whether one hand was occupied during the activity,             if distance and target size are equal, then differences in
and found walkers were more likely to have one hand                  speed between tasks suggest the influence of external
occupied (60%), followed by standers (50%), and finally              factors. We hypothesize that one such factor is the influence
sitters (25%), which may be the real explanation for why             of device size on thumb movement and that the speed
walkers were more likely than standers to dial with one              differentials between conceptually equivalent trials will
hand, followed by sitters. Regardless of activity, when both         allow us to identify surface regions that are most
hands were available for use, the percentage of one vs. two          appropriate for interactivity.
handed dialers were equal (26%).
                                                                     Equipment
Analysis of Field Study
                                                                     Device models
Figure 2 suggests that the activity performed while dialing          For real devices, design elements such as buttons and
may influence the number of hands used. However, since               screens communicate to the user the “valid” input areas of
the percentage of users with one hand occupied correlates            the device. But instead of design details influencing user
with the distribution of one-handed use across activities,           interaction, we wish for user capability to dictate the
hand availability may be the more accurate indicator of the          potential positions of interface elements. Thus to remove
number of hands used to dial. While scenario clearly                 the bias inherent in existing devices, we modeled four
impacts usage patterns, the fact that users were as likely to        common handheld devices: (1) a Siemens S56 candy bar
use one hand as two hands when both hands were available             phone measuring 4.0 x 1.7 x 0.6 in (10.2 x 4.3 x 1.5 cm);
suggest preference, habit and personal comfort also play a           (2) a Samsung SCH-i600 flip phone measuring 3.5 x 2.1 x
role. Regardless of scenario, we can safely conclude that            0.9 in (9 x 5.4 x 2.3 cm); (3) an iMate smartphone
one-handed phone use is quite common, and thus an                    measuring 4 x 2.0 x 0.9 in (10.2 x 5.1 x 2.3 cm) and (4) an
essential consideration in mobile phone design.                      HP iPAQ h4155 Pocket PC measuring 4.5 x 2.8 x 0.5 in
                                                                     (11.4 x 7.1 x 1.3 cm). We refer to these as simply SMALL,
THUMB MOVEMENT STUDY
                                                                     FLIP, LARGE, and PDA. We removed all superficial design
The third component of this work is an empirical study that
                                                                     features, leaving only form. The 3D models were printed
examines the ergonomics of thumb mobility in the context
                                                                     using Z Corp.’s ZPrinter 310 (http://www.zcorp.com/) rapid
of mobile device use. We believe that effective designs for
                                                                     prototyping system. Device models were hollow to hide
one-handed use must take into account the areas on the
                                                                     materials required for touch sensing, but weight was
surface of the device that are accessible to the thumb when
                                                                     reintroduced to provide a realistic feel. Once printed and
held in one hand, and as importantly, during repeated
                                                                     cured, the models were sanded and lightly shellacked to
interaction. The thumb has a wide range of motion and is
                                                                     achieve a comfortably smooth finish.
extremely versatile, but is most adapted for grasping tasks
in opposition to the fingers [5]. Thus one-handed mobile
                                                                     Target Design
interaction introduces novel requirements for the thumb –            A maximal orthogonal grid of circular targets 0.6 inches in
repetitive pressing tasks issued on a plane parallel to the          diameter was fitted to the surface of each device. Circles
palm. Although we can make (possibly good) guesses about             were used for targets so that the effective target size would
the resulting range of movement, empirical evidence is a             not be influenced by the angle of approach [18]. The target
better guide. Unfortunately, no strictly relevant studies have       size was estimated to be large enough for an average-sized
yet been conducted. In response, we have designed and                thumb to hit without overlapping other targets, but small
conducted a study to help us understand how device form              enough to provide adequate surface coverage for each
(small candy bar phone, large candy bar phone, flip phone,           device. The size of the grid of touch targets for the devices
and PDA) influences thumb mobility. Device interactivity             were: small candy bar phone (5x2), large candy bar phone
is characterized by a series of actions. Success in                  (7x3), flip phone (4x3), and PDA (6x4). Targets were
maintaining control of the device depends on the sequence,           numbered for identification.
location, and speed with which the user interacts with
surface. Thus instead of using strength as our measure for           Measurement
success, we use speed of access as our measurement                   A measurement strategy for Fitts’ Law tapping tasks might
parameter, and have based our study design on Fitts’ Law.            involve a surface-based sensor to detect finger contact.
                                                                     However, since our manufactured plaster devices possessed
Fitts Law Overview
                                                                     no input capabilities, we instead employed Northern Digital
Fitts’ Law is a model of human movement performance that             Inc.’s OPTOTRAK 3020 motion analysis system designed
states movement time (MT) is proportional to target                  for fine-grained (0.2 mm) tracking of motor movement.
distance (D) and inversely proportional to target size (W),          OPTOTRAK uses 3 cameras to determine the precise 3D
to yield the equation: MT = a + b log2 (2A/W) [9]. Because           coordinates of infrared emitting diodes (IREDs). Four
constants a and b differ by both appendage and task, they


                                                                 5
 IREDs affixed to the surface of each device defined a local          Department of Computer Science, with the only restriction
 coordinate system for the device. Positions of two markers           that participants be right-handed. Participants (15 male)
 (for redundancy) affixed to the tip of each participant’s            ranged in age from 18 to 35 years with a median age of 25
 thumb were then translated with respect to the coordinate            years. Participants received $20 for their time.
 system of the device to establish relative movement
 trajectories, from which taps were then derived. Diode               Design
 positions were sampled at 100Hz.                                     For each of the four devices (small, flip, large, and PDA)
                                                                      users performed all combinations of distance unit (1 and 2)
 While the OPTOTRAK senses movement with high
                                                                      x direction (N↔S, E↔W, NW↔SE, NE↔SW in compass
 resolution, it does not explicitly indicate when the thumb
                                                                      notation) as supported by the geometry of the device. For
 actually touches the surface of the device. So in an attempt
                                                                      example, the small phone did not accommodate trials of
 to more robustly detect tap time, we investigated surface
                                                                      unit 2 in the E↔W, NW↔SE, or NE↔SW directions. For
 touch sensing technologies. We sought an option that was
                                                                      the large phone and PDA, trials of distance 4 units were
 inexpensive and flexible enough to work across our variety
                                                                      included when possible. Finally each device included a
 of device sizes. Initial experimentation with wiring Phidgets
 (www.phidgets.com) touch sensors to copper tape on the               single trial each of NW↔SE and NE↔SW to opposite
 surface of a device was promising, although taps were often          corners of the target grid. The resulting number of trials for
 lost if done in rapid succession. We found the loss could be         each device were as follows: small candy bar phone (32),
 mitigated by ensuring successive taps were either not too            flip phone (48), large candy bar phone (108), and PDA
 fast or were received by different sensors. This lead to a           (128). There were more trials for the larger devices because
 design in which no adjacent target was wired to the same             the grid of touch targets was larger on the larger devices.
 touch sensor. For each model, a hole was drilled in the
                                                                      Tasks
 center of each target position. A copper tab approximately
                                                                      Users performed reciprocal tapping tasks in blocks as
 1cm square was placed over each hole Figure 3c), and
                                                                      follows. For the small and flip phones, trials were divided
 soldered from the back side such that each was connected to
                                                                      equally into two blocks. For the large phone and PDA, trials
 one of four series of wires (Figure 4). A printout of circular
                                                                      were divided equally into 4 blocks. To allow users to focus
 targets was then placed over the grid of tabs. To avoid cross
                                                                      attention on the device, trials were announced by audio
 talk, each series was connected via sheathed wire to a single
                                                                      recording. Users were presented with the name of two
 touch sensor placed approximately 2 feet from the user. All
                                                                      targets by number. For example, a voice recording would
 wires were enclosed in the device, except for an umbilical
                                                                      say “1 and 3”. After 1s, a voice-recorded “start” was
 chord composed of 4 IRED wires and up to 4 sensor wires.
                                                                      played. Users tapped as quickly as possible between the two
 Software
                                                                      targets, and 5s later, a “stop” was played. After a 1.5s delay
 Data collection and experiment control software was run on           the next trial began. Trials continued in succession to the
 a Gateway 2000 Pentium II machine with 256 MB of RAM                 end of the block, at which point the user was allowed to rest
 running Windows 98.                                                  as desired, with no user resting more than 2 minutes.
                                                                      Devices and blocks were counterbalanced across subjects
 Participants                                                         and within-block trials were randomized.
 Twenty participants were recruited via fliers posted in our
                                                                      Procedure
                                                                      Each participant session began with a brief description of
                                                                      the tasks to be performed and the equipment involved. Two




Figure 3. Device models at varying stages of development; (a)
large, (b) small, (c) PDA and (d) flip.                               Figure 4. Example sensor setup.



                                                                  6
IRED markers were then attached to the right thumb with                Phidgets hardware is not designed for use under
two-sided sticky tape. One diode was placed on the leftmost            experimentally stringent conditions.
edge of the thumb nail, and a second on the left side of the
                                                                       We instead analyzed the 3D thumb tip trajectory data to
thumb. The orthogonal placement was intended to
                                                                       derive surface taps as follows. If we call the distance of the
maximize visibility of at least one of the diodes to the
                                                                       thumb from the surface of the device its Z distance, then
cameras at all times. The two marker wires were tethered
                                                                       intuitively, a tap occurs when both the value and change in
loosely at the participant’s right wrist with medical tape to
                                                                       Z distance is minimal. Intuitively, this would require
avoid disturbance.
                                                                       finding the local minima of a sinusoidal-like wave that
The participant was seated in an armless chair with the                represents the change in Z over time. From a signal
OPTOTRAK cameras positioned over the left shoulder. At                 processing perspective, this amounts to analyzing the signal
this point users were given more detailed instruction about            to detect maxima and minima in the presence of noise and
the nature of tasks and the block design. The participant              then finding points where the first derivative of the distance
was also instructed that if at any point at least three of the         in the Z direction crosses zero from negative to positive,
IREDs on the surface device or at least one of the diodes on           indicating an inflection point in the Z trajectory, or tap.
the thumb were not within line of sight of the camera, an
                                                                       Before this analysis, data was first preprocessed to extract
out-of-sight error sound would be heard, at which point he
                                                                       the middle 3 seconds of each trial and select the thumb
or she should continue the trial as naturally as possible
                                                                       diode that had the most complete data (the fewest number
while attempting to make adjustments to improve diode
                                                                       of missing frames, or if equal, the one with the smallest
visibility. The participant was then given the first device
                                                                       maximum window of missing frames). Linear interpolation
and began a practice session of 24 trials, selected to
                                                                       was performed on missing frames as long as the gap was
represent all condition types and a variety of surface
                                                                       less than 100 ms. Missing frames included those lost due to
locations. During the practice trials, the administrator
                                                                       out-of-sight errors, as well as occasional frame drops by
intentionally occluded the diodes to give the participant
                                                                       collection hardware. Raw trial data was then analyzed by
familiarity with the out-of-sight error sound and proper
                                                                       the PICKEXTR Matlab function to identify extrema in a
remedies. After completion of the practice trials and
                                                                       signal. This function is provided with the RelPhase.Box
indication that the participant was ready, the study proper
                                                                       Matlab toolbox for relative phase analysis of oscillatory
was begun. After all trials for a device were completed,
                                                                       systems, by Tjeerd Dijkstra.
users were allowed to rest for as long as it took to set up the
next device, typically 3 to 5 minutes. After the last device,          Extrema output from PICKEXTR were used in conjunction
the participant completed a questionnaire, recording                   with a velocity vector derived from the Z distance to
demographics and subjective ratings. Total session time                determine taps as follows. An initial approximation of the
was approximately 2 hours.                                             velocity vector was determined by taking the first derivative
                                                                       of the Z distance which was then passed through a
Measures                                                               Butterworth filter to remove high frequency noise. The two
Dependent variables collected during the study included                data streams were then analyzed to find zero-crossings in
task time and for each device, subjective rating of difficulty         the velocity vector that occurred between signal extrema,
from 1 (easy) to 7 (difficult), and device regions that were           indicating an inflection point in the Z distance (i.e., a tap.)
easiest/hardest to access overall. Task times were                     Trials for which each tap did not have an intermediate peak,
determined by analyzing the middle 3 seconds of each 5                 or for which the standard deviation was suspicious were
second trial. This sub-sequence was chosen to avoid                    kicked back for analysis by hand.
artifacts resulting from initiation lag, or prediction of trial
completion (a phenomenon routinely observed by the study               Approximately 20% of the data was coded by hand for a
administrator.) Next, the location in time of all taps were            sanity check using ginput, a visual MATLAB tool for
identified within the 3 second interval and a single average           inputting 2D data points. In these cases, the first and last
tap time computed from the difference in time between the              taps that were identified by the automated process were
onset of the first tap to the onset of the last tap, divided by        retained so as not to bias the total time. Since average tap
one fewer than the total number of taps detected.                      time is calculated as the number, not placement, of
                                                                       intervening taps, this method minimized bias that could
Analysis                                                               result from human annotation. A total of 10 trials were
We expected to use touch sensor data for tap analysis.                 excluded from the statistical analysis because they were not
Unfortunately, too few taps were recorded by the sensing               classifiable by either human or computer.
hardware, making that data unreliable. Possible causes for
this failure were that actual trials were performed more               Results
quickly than pilot trials, that actual subjects did not press as       Times
hard or as long, or there could have been an intermittent              Due to the differing geometries of each device, unequal
problem with the wiring. Loss was not systematic enough to             numbers of trials were performed for each device (SMALL,
be traced to a specific cause, but it is also possible that the        FLIP, LARGE, PDA) x direction (N↔S, E↔W, NW↔SE,


                                                                   7
                                                                       0.45

                                                                        0.4

                                                                       0.35

                                                                        0.3
                                                                                                                                                   NS
                                                                       0.25                                                                        EW
                                                                        0.2                                                                        NESW
                                                                                                                                                   NWSE
                                                                       0.15

                                                                        0.1

                                                                       0.05

                                                                         0
                                                                              Small      Flip     Large     PDA        Flip     Large     PDA
                                                                              1, 1.4   1 & 1.4   1 & 1.4   1 & 1.4   2 & 2.8   2 & 2.8   2 & 2.8


                                                                       Figure 6. Tap time in seconds by device, direction and
                                                                       distance.
 Figure 5. Best (green) and worst (red) trials by unit distance
 and device. Green trials are significantly faster and red            thumb’s natural axis of rotation and thus requires little
 trials are significantly slower than 25% of the trials of the        flexion. Movement in the NW↔SE direction on the other
 same distance.                                                       hand, requires a great deal of flexion, which likely serves as
                                                                      a mechanical encumbrance.
NE↔SW) x actual distance (1.0, 1.4, 2.0, 2.8 units), with
some combinations having no trials. Thus for each device, a
                                                                      Directions
univariate analysis of variance (UNIANOVA) was carried
                                                                      The previous analysis suggests that distance may be a factor
out on tap time for the single factor trial, grouped by actual
                                                                      in whether device region influences movement
distance. Main effects of trial were observed at distance 2.8
                                                                      performance. However, because distances for N↔S and
units for PDA F(15, 305)=2.08, p=.011, LARGE F(9,
199)=2.84, p=.004, and FLIP F(5,114)=2.58, p=.03 for                  E↔W trials differ from those in the NW↔SE and
distance of 2.8 units. Trials of this distance were performed         NE↔SW trials, we cannot not directly compare all four
                                                                      directions. Instead we averaged tap times by device,
in only the NW↔SE and NE↔S directions. While no main
effects were observed for the remaining device x distance             distance (1, 1.4, 2, and 2.8 units) and direction. We found
combinations, specific pairs of trials differed significantly         that N↔S trials were on average 0.7% to 5.6% faster than
from one another. To observe trends these trials are                  E↔W trials of the same distance for all devices. We also
represented in Figure 5. Red trials are those for which users         found that NE↔SW trials were on average between 3.44%
were significantly slower than at least 25% of the trials of          to 19.46% faster than NW↔SE trials of the same distance
the same distance, while green trials were significantly              for all devices. Figure 6 shows the average tap times by
faster than at least 25%.                                             device, direction and distance, where N↔S and E↔W
                                                                      trials have distances 1 and 2, while NE↔SW and NW↔SE
The fact that no differences were found between trials for
                                                                      tirals have distances 1.4 and 2.8.
the small phone indicate that its small size allows users full
access to, and therefore command of, its surface area.                Regions
However, the interactions among trials for the larger                 Even when distance is not a factor, we would like to
devices seem to suggest that critical dimension in                    understand whether there are specific regions, not just
determining thumb agility is device width, as the SMALL               trajectories that are especially suitable for interaction. To
device is the narrowest but has a height which falls between          answer this question, we induced movement regions by
that of the FLIP and LARGE. While FLIP and LARGE                      averaging the trial times for trials of distance 1 or 1.4 for
differ dramatically in height the number of trials that differ        every 2x2 subgrid of the target space of each device for a
significantly from one another are comparable for distances
                                                                      total of six trials per region (2 N↔S, 2 E↔W, 1 NW↔SE,
≤ 2.0 units. While the PDA also has a height which falls
                                                                      and 1NE↔SW). A repeated measures analysis of variance
between those of the FLIP and LARGE, it has considerably
                                                                      (RM-ANOVA) on region revealed a main effect for the
more differentiable trials at distance ≤2.0 units and less,           LARGE F(11,209) = 2.3, p=.011 and PDA F(14, 266) =
which may be explained by its larger width. These findings            2.57, p=.002. For both devices we segmented the regions
suggest that width play an important role in the device               into three groups as follows: (1) regions for which at least
control, with wider devices negatively impacting thumb                one other region was significantly faster, (2) regions for
movement. It also appears that in general movement is                 which at least one other region was significantly slower,
easiest in the areas toward the center of the device. Finally         and (3) the remaining regions. However, averaging region
movement is favorable in the NE↔SW but unfavorable in                 times within each group showed that for both devices, only
the NW↔SE direction, at least for longer distances. This              modest speedup (2.3% - 7.2%) might be gained by
last observation may be explained by the fact that                    interacting within the fastest regions versus the slowest
movement in the NE↔SW direction is consistent with the


                                                                  8
                                                                     relationship exists between the easy regions (dark gray) in
                                                                     column 1 and fastest regions (light gray) in column 3.
                                                                     In addition to region marking, we asked users to rate the
                                                                     overall difficulty of managing each device with one hand
                                                                     on a scale from 1 = comfortable to 7 = uncomfortable.
                                                                     Average ratings from most to least comfortable were as
                                                                     follows: small (1.65), flip (2.65), large (3.96) and PDA
                                                                     (5.05). It is not surprising that this ordering correlates to the
                                                                     sizes of the devices, since as device size grows, more effort
                                                                     is required to maintain control of the device, which inhibits
                                                                     thumb movement.

                                                                     DISCUSSION
                                                                     While results from each stage of analysis were fairly
                                                                     modest, they reinforce one another to yield compelling
                                                                     messages. Upon reflection, many of these findings seem
                                                                     obvious, yet until now there has been no quantitative
                                                                     evidence to justify design decisions based on such intuition.
                                                                     The first message is that device size impacts thumb
                                                                     mobility; the smaller the device, the easier it is to use. It is
                                                                     not just that larger devices have regions that are farther
                                                                     away and thus harder to reach, but very close regions also
                                                                     become harder to access. For far targets, large devices get
 Figure 7. Movement maps by device. Depth of color in                in the way of full opposition capability. For near targets,
 columns 1 and 2 indicate user agreement.                            presumably key muscles in the hand are dedicated to
                                                                     maintaining device control which detracts from thumb
regions. These results are shown visually in the right-hand
                                                                     agility.
column in Figure 7.
                                                                     The second message is that locality of movement aids
The narrow margin between groupings seems to indicate
                                                                     performance. Users can adapt to a variety of regions,
that users can interact reasonably efficiently in nearly any
region so long as the movement distances are small. In fact,         presumably by repositioning the device, and perform
                                                                     reasonably for tasks over short distances. When movement
we intentionally did not control for hand position, since in a
                                                                     distances increase, location and direction factor into
real scenario users will adjust the device for optimal
movement. The regions were indistinguishable with respect            performance, with movement in the NW↔SE direction
to performance for smaller devices (SMALL and FLIP),                 particularly vulnerably to degradation over distance.
which again suggest that users have greater freedom of               Finally, it is important to remember that all study
movement over smaller devices than larger devices.                   participants were right-handed and used their right hand in
                                                                     the experiment.      Presumably, the results would be
Subjective Preferences                                               symmetric for left-handed users.
Although results from the region analysis were not very
discriminating, they are more compelling when considered             CONCLUSION
with subjective opinions about the difficulty of device use          In an effort to understand the interaction needs of mobile
by target. After completing all trials, users were presented         device users, we looked at a broad range of device use. Our
with diagrams of each device similar to those in Figure 7            field study showed that for at least one class of user
and asked to identify the targets they found most easy and           (travelers), mobile phones are often used with one hand and
most hard to interact with. Aggregating results across users         seem to correlate with activity, such as walking or holding
yielded a preference “heat” map for the most accessible              items in the other hand. Our survey revealed that a
targets of each device (column 1), with darker regions               remarkable percentage of users want to use one hand for
indicating more agreement among participants. Similarly,             interacting with mobile devices, but that current interfaces
column 2 shows the agreement among users for the most                are not designed to support dedicated single handed use.
difficult targets. We see that for each device the two               Finally, an empirical evaluation of thumb interaction on
representations are roughly inverses of one another. Recall          devices of varying size confirmed many of our intuitions
the center of each 2x2 subgrid of targets defines what we            about single-handed device use: thumb mobility degrades
called a region in the previous analysis. We see from                with increases in device size hence reducing the accessible
column 3 that there is indeed correspondence between the             real estate; mid-device regions are easier to access than near
targets that were seen as difficult (dark gray) in column 2          and far regions; finally NW↔SE movement for right-
and red and pink regions in column 3. A similar


                                                                 9
handed users degrades with movement distance and device            12. Hwang, S., Lee, G., QWERTY-like 3x4 keypad layouts
size.                                                                  for mobile phone. Posters CHI 2005. ACM Press
                                                                       (2005), 1479-1482.
ACKNOWLEDGMENTS                                                    13. Karlson, A.K., Bederson, B.B., SanGiovanni, J.
We appreciate the advice of François Guimbretière                      AppLens and LaunchTile: Two designs for one-handed
throughout the project, especially in designing the                    thumb use on small devices. Proc. CHI 2005. ACM
prototypes for the study. We also appreciate Dave Levin’s              Press (2005), 201-210.
helpful comments on drafts of this paper, and Pekka Parhi’s
help in analyzing the survey comments.                             14. Kuo, L-C., Cooney, W.P., Kaufman, K.R., Su, F-C., and
                                                                       An, K-N. A kinematic method to calculate the
This work was supported in part by Microsoft Research.                 workspace of the trapeziometacarpal joint. Proc. Inst.
                                                                       Mech. Eng. H 218 2 (2004) 143-149.
REFERENCES
1. Barmakian, J.T. Anatomy of the joints of the thumb.             15. Lam, H., and Baudisch, P. Summary thumbnails:
   Hand Clin. 8 4 (1992), 683-691.                                     readable overviews for small screen web browsers, Proc
                                                                       CHI 2005. ACM Press (2005), 681-690.
2. Baudisch, P., Xie, X., Wang, C., and Ma, W-Y.
   Collapse-to-zoom: viewing web pages on small screen             16. Li, Z-M, Harkness, D.A. Circumferential force
   devices by interactively removing irrelevant content.               production of the thumb. Medical Eng. & Physics 26
   Proc. UIST 2004. ACM Press (2004), 91-94.                           (2004) 663-670.
3. Bederson, B.B., Clamage, A., Czerwinski, M.P., &                17. Lyons, K., Starner, T., Plaisted, D., Fusia, J., Lyons, A.,
   Robertson, G.G. DateLens: A Fisheye Calendar                        Drew, A., and Looney, E.W. Twiddler Typing: one
   Interface for PDAs, Transactions on Computer-Human                  handed chording text entry for mobile phones. Proc.
   Interaction 11 1 (2004), 90-119.                                    CHI 2004. ACM Press (2004), 671-678.
4. Bjork, S., Holmquist, L.E., Redstrom, J., Bretan, I.,           18. MacKenzie, I.S. and Buxton, W. Extending Fitts’ Law
   Danielsson, R., Karlgren, J. and Franzen, K. WEST: a                to two-dimensional tasks. Proc. CHI 1992. ACM Press
   web browser for small terminals. Proc. UIST 1999.                   (1992), 219-226.
   ACM Press (1999), 187-196.                                      19. MacKenzie, I.S., Kober, H., Smith, D., Jones, T., and
5. Bourbonnais, D., Forget, R., Carrier, L. and Lepage, Y.            Skepner, E. LetterWise: prefix-based disambiguation for
   Multidirectional analysis of maximal voluntary                     mobile text input. Proc. CHI 2001, ACM Press (2001),
   contractions of the thumb. J. Hand Ther. 6 4 (1993),               111-120.
   313-318.                                                        20. Perlman, J.L., Roach, S.S., and Valero-Cuevas, J. The
6. Butts, L. and Cockburn, A. An evaluation of mobile                  fundamental thumb-tip force vectors produced by
   phone text input methods. Proc. CHI 2002. ACM Press                 muscles of the thumb. Journal of Orthopaedic Research
   (2002), 55-59.                                                      22 (2004), 306-312.
7. Buyukkokten, O., Garcia-Molina, H., Paepcke, A. and             21. Sazawal, V., Want, R. and Borriello, G. The unigestures
   Winograd, T. Power browser, efficient web browsing                  approach. Proc. Mobile HCI. Springer-Verlag (2002),
   for PDAs. Proc.CHI 2000. ACM Press (2000), 430-437.                 256-270.
8. Chen, Y., Ma, W.Y., and Zhang, H.J. Detecting                   22. Silfverberg, M., MacKenzie, I.S., and Korhonen, P.
   webpage structure for adaptive viewing on small form                Predicting Text Entry Speed on Mobile Phones. Proc.
   factor devices. Proc. WWW 2003. ACM Press (2003),                   CHI 2000, ACM Press (2000), 9-16.
   225-233.                                                        23. Wigdor, D. and Balakrishnan, R. TiltText: using tilt for
9. Fitts, P.M. The information capacity of the human motor             text input to mobile phones. Proc UIST 2003, ACM
   system in controlling the amplitude of movement. J. of              Press (2003), 81-90.
   Experiment. Psychology 47 (1954), 381-391.                      24. Wobbrock, J.O., Forlizzi, J., Hudson, S.E., and Myers,
10. Gartner. Forecast: Mobile Terminals, Worldwide, 2000-              B.A. WebThumb: Interaction techniques for small-
    2009. http://www.gartner.com.                                      screen browsers. Proc. UIST 2002. ACM Press (2002),
                                                                       205-208.
11. Hirotaka, N. Reassessing current cell phone designs:
    Using Thumb Input Effectively. Posters CHI 2003.
    ACM Press (2003), 938-939.




                                                              10

				
DOCUMENT INFO
Shared By:
Categories:
Stats:
views:54
posted:6/17/2010
language:English
pages:10
Description: Top Technology For 2010 Year
Sherin William Sherin William http://
About