Please enable JavaScript.
Coggle requires JavaScript to display documents.
Effects of Display Size and Navigation Type on a Classification Task…
Effects of Display Size and Navigation Type
on a Classification Task
ABSTRACT
We report on a controlled experiment
that uses this task to compare physical navigation in
front of a wall-size display with virtual navigation using panand-zoom
on the desktop. Our main finding is a robust interaction
effect between display type and task difficulty: while
the desktop can be faster than the wall for simple tasks, the
wall gains a sizable advantage as the task becomes more dif-
ficult. A follow-up study shows that other desktop techniques
(overview+detail, lens) do not perform better than pan-andzoom
and are therefore slower than the wall for difficult tasks
INTRODUCTION
a. . This increased density, in turn, affords
physical navigation. Users simply approach the screen to see
detail and step back for an overview, similar to the pan-andzoom
navigation available on a desktop display. This raises
the question as to the relative trade-offs between physical navigation
with a wall-size display versus virtual navigation on a
desktop
b. However, our observations of actual
users during prototyping and real-world tasks shows that they
want to reorganize data displayed on the wall: users move
items around and group them in a way that is meaningful to
the task at hand
c. Our challenge is how to design an abstract task that operationalizes
the critical aspects of data manipulation in order
to conduct controlled experiments that compare task performance
for both wall-size and desktop displays. To increase
internal validity, the task should reduce the cognitive load
associated with the decision-making process and focus on
actual data manipulation. To increase external validity, it
should feature the same typical interactions found in realworld
tasks. Finally, experimenters should be able to vary,
in a controlled way, the difficulty of the task.
d. We conclude with a discussion of the main
result, which found a robust interaction effect between display
type and task difficulty: Although the desktop is often
faster for simple tasks, the wall-size display performs signifi-
cantly better with increased task difficulty
MOTIVATION
a. Application Examples
Scheduling a large conference on a wall-size display: Teams
and individuals move close for detail or stand back for an overview.
s. The most common
scheduling task consisted of identifying a misplaced or
in-conflict presentation, finding a better slot, and moving it
there
. Because the schedule was so heavily constrained, one
move often triggered another and schedulers, often working
in groups, had to juggle sequences of updates.
y, we are currently designing an application that lets
users organize sequences of video clips on the wall.
b.Implications
These applications share three elements: A complex
decision-making task that relies on the users’ expertise as
well as their ability to quickly access the full content of the
wall; a structured display in which information is logically
organized in a grid; and a need to manipulate data by moving
items from one cell to another
The above examples illustrate how benefits of wall-size displays
extend beyond visualization of extremely large images
to include tasks requiring human judgment and the active manipulation
of large datasets
. Here, manipulating content is an
integral part of the task, whether to better understand the data
set, to form an opinion, or to enact a decision.
The size of the dataset is the primary reason why users want to
move off the desktop onto a wall-size display3
. Users can see
all of it at once, yet access details simply by walking toward
the wall. Users take better advantage of their spatial memory
since it is coupled with their physical movement in space.
In contrast, the virtual navigation imposed by a desktop interface
can be disorienting, and the overhead of constantly
navigating the data set, e.g. with pan-and-zoom, can distract
users and increase their cognitive load. Even so, physical locomotion
is more time-consuming and tiring than virtual navigation,
and manipulating data with well-known devices and
widgets may be more efficient than using mid-air techniques
on a wall-size display [18]
RELATED WORK
a. Large Displays in Desktop Settings
For instance, Czerwinski
et al. [10] observe higher productivity and satisfaction with
larger display surfaces when performing complex daily tasks.
Bi and Balakrishnan [7] compare a large projected wall display
with single and dual desktop monitors. Their results suggest
that large displays facilitate tasks with multiple windows
and rich information because they offer a more immersive experience,
with enhanced peripheral awareness.
In sense-making tasks, high resolution displays not only display
more information but also provide a virtual space where meaning is encoded in the spatial relationship between data,
documents, the display, and the user [1]. Similarly, increasing
display size and resolution both improve user performance in
rich-information environments
Tan et al. [26] demonstrate that large projection displays support
spatial orientation tasks better than desktop monitors,
and argue that these more immersive environments encourage
egocentric rotations, leading to improved performance
Although these studies consistently show the benefits of
larger displays, most were conducted in traditional desktop
settings where users sit before a monitor, with limited or no
physical locomotion
b. Physical Navigation with Wall Displays
As display size and pixel density increase, standing and moving
in front of large displays becomes necessary. Ball et al. [4]
show that larger displays promote physical navigation and
improve user performance for search, navigation and pattern-
finding tasks. However, their tasks do not involve data manipulation,
and they do not include a desktop condition for comparison.
Ball and North [3] investigate the key advantages of
large displays and find that physical navigation is more important
than increasing the field of view. Yost et al. [27] show
that user performance improves with larger displays despite
the need for physical navigation.
Large displays also affects perception. For example, Endert
et al. [12] demonstrate the impact of visual encodings on
physical navigation on large displays and show that physical
navigation improves user performance
. In summary, these studies show the benefits of physical
navigation in some situations for certain tasks, none of which
feature data manipulation.
c.Multiscale Interfaces and Display Size
Multiscale interfaces [9] were designed to visualize large
quantities of data on displays that are too small
. Surprisingly, the large
display is not always faster, and is sometimes slower than the
medium display. The authors suggest that some techniques
require increased target searching time on the large display.
These results suggest only a small or no benefit of large displays
when using multiscale navigation techniques.However, these studies were conducted in desktop settings where users
sit in front of the display, with tasks that involve only visualization
or target acquisition.
d. Summary and Approach
Our goal is to build upon previous work to improve our understanding
of the trade-offs between wall-size displays and
physical navigation on the one hand, and desktop monitors
and virtual navigation on the other, for data manipulation
tasks. We must identify which input techniques are most appropriate
for each setting and construct an abstract data manipulation
task that captures the essential elements of the realworld
tasks we observed.
ABSTRACT CLASSIFICATION TASK
Scheduling tasks are
even more complex because they add more constraints, such
as avoiding conflicts across parallel sessions.
To classify the items, we need to know when two items are
in the same class. In practice, such decisions are domainspecific,
e.g., two conference papers on the same topic or two
brains with similar features, and often require expert judgment
or incur a heavy cognitive load.
Our experiment uses the simplest
solution i.e. we operationalize information density by
adjusting font size.
We control the complexity of our classification task via several
parameters: number of items, number of classes, number
of containers, and representation of the item, including the label
font size. These factors define a rich yet easy-to-control
design space for experimental tasks based on the abstract task
a. Experimental Task:
Label size affects legibility at a distance and thus influences
the level of physical or virtual navigation required to be able
to read a label and make a decision
This not only reduces the time
needed to solve each task, but it also has ecological validity,
since in the real-world tasks we observed, participants built
upon an initial classification made by others or a computer
had generated an initial pass for which some errors remain
EXPERIMENT 1: WALL VS. DESKTOP
Our goal is to investigate the trade-offs between physical and
virtual navigation and how they affect task performance.
. Based on our review of the
literature and our experience using the wall-size display, we
formulate three hypotheses:
● Wall performs better than Desktop for smaller labels;
● Wall performs better than Desktop for harder tasks;
● Desktop performs better than Wall for larger labels and
simpler tasks.
The use of a different input device for the Wall and Desktop
conditions is meant to maximize external validity
Procedure
The experiment is a [2×3×2] within-participants design with
three factors:
● DISPLAY: display type, Wall or Desktop;
● LABELSIZE: label size, Large, Medium or Small;
● DIFFICULTY: number of classes, Easy (2 labels) or Hard (4
labels).
Participants
are told to complete the tasks as quickly as possible but
to avoid dropping items into the wrong containers, to discourage
a trial-and-error strategy.
The experiment is split into two sessions, one per DISPLAY.
Half the participants start with Wall, the other half with Desktop.
The order of the DIFFICULTY and LABELSIZE conditions
are counterbalanced across participants using Latin Squares.
To minimize the potential order effect between DISPLAY conditions,
we use the same sequence of trials and symmetric
layouts for each participant between the Wall and Desktop
conditions. The experiment lasts about one hour
Results
(i) Movements of the Virtual Viewpoint and of the Participants:
With large labels, no navigation
is needed to perform the task, and indeed almost no viewpoint
movement occurs in the Desktop condition. However,
most participants did move in front of the wall (about 482cm
per trial on average), with no evidence that this depends upon
DIFFICULTY. For the other conditions, the amount of movement
increases significantly both with smaller LABELSIZE
and higher DIFFICULTY. These differences correlate with the
differences in task completion time. In particular, viewpoint
movements increase sharply for Small-Hard (in scene space).
For Small and Medium sizes, in scene space, the length of
the virtual navigation is longer than that of physical navigation
(Fig. 5). This is not surprising, and can be attributed to
the users’ ability to move their head and eyes
However, the distance in screen space (Fig. 5) for the desktop
is shorter or close to that of the wall. This indicates that virtual
navigation competes with physical navigation in terms of
distance covered in motor space, and therefore the difference
in performance between Wall and Desktop for difficult tasks
must have another explanation.
(ii) Physical vs Virtual Reach:
To complement our analysis of participants’ movements, we
now look at their ability to interact with distant targets. Indeed,
the larger size of the wall-size display enables users to
reach targets at a distance without moving while on the desktop
they must pan and/or zoom the scene.
, relative to the center of the view.
With Medium and Small labels, the points are more closely
clustered for the desktop than for the wall, indicating that the
participants’ reach is larger on the wall. Indeed, while on the
desktop users must bring the target into view with pan-andzoom,
they can act at a larger distance on the wall, reducing
the need for navigation. On the other hand, with Large labels
no navigation is needed for the Desktop nor the Wall. However
the Wall requires more head movements, which might
explain why the desktop is faster
(iii) Angular Size of the Labels:
In our design, LABELSIZE operationalizes information density:
smaller text size forces participants to get closer to the
display through either physical or virtual navigation in order
to make an informed decision
Most participants found
the small labels with four letters (hard task) tiring, in both
conditions.
Fig. 11 summarizes participants preferences between the
desktop and the wall. Except for Large labels, almost all participants
preferred the wall
This may be due to the novelty effect of using a wall-size display
as well as other factors yet to be identified, including
spatial memory.
In the
hard conditions, they tried to remember the positions of the
misplaced items and of the containers to reduce navigation
In summary, these results show a robust interaction effect between
display type and overall task difficulty (label size and
number of classes), with the wall up to 35% faster in the hardest
condition. This difference can be attributed to the ability
to use more efficient strategies on the wall, as evidenced by
the larger reach of users. Other factors are likely at work,
though, such as a better use of spatial memory.
EXPERIMENT 2: THREE DESKTOP TECHNIQUES
Experiment 1 showed a strong performance advantage of
physical navigation on a wall-size display when compared
with pan-and-zoom navigation on a desktop interface for
difficult classification tasks. Could these results be different
with other types of virtual navigation? To test this hypothesis,
we compared three desktop techniques in a second
experiment: the baseline pan-and-zoom technique, an
overview+detail technique and a focus+context technique [9]
Overview+detail adds a miniature view of the scene (the
overview) displayed in a corner of the main view (the detail
view)
The
literature suggests that adding an overview to a pan-and-zoom
interface increases user satisfaction [14, 19] and that an interactive
overview can be very efficient for search tasks [21]
We tested an interactive overview but found that it slowed
users down. The switching cost between views was too high
when they performed pick-and-drop actions. This shows how
a data manipulation task can affect the usability of a technique
that has been tested only for search or visualization tasks. Instead,
we chose to test a PZ+OV technique, which adds a
passive overview in the lower-right corner of the screen, with
a rectangle showing the current position of the detail view.
Lenses [25] are another way to combine focus and context in
a single view. We implemented a fisheye lens that is permanently
attached to the cursor and has the same radius as the
disks. The entire scene is scaled down to fit the display and
the lens has a magnifying factor of 6, making the small labels
readable. To avoid occlusion during pick-and-drop, the disk
being picked is attached to the bottom of the lens.
a. Method:
The experiment is a within-subjects design with one factor
(TECHNIQUE): PanZoom, PZ+OV, Fisheye.
b. Results
The analysis of variance reveals no significant effect of TECHNIQUE
on task completion time
In summary, this experiment confirmed that the wall-size display
out performs the desktop for difficult data classification
tasks. Although new techniques could be devised to improve
the desktop condition, e.g., using multiple or adaptive
lenses [23], we believe that they are unlikely to help the desktop
beat the wall for complex data manipulation tasks.
7 CONCLUSION AND FUTURE WORK
This paper introduces a classification task that abstracts out
a wide category of tasks that involve data manipulation and
operationalizes two key factors: information density and task
difficulty. This abstract task was informed by our observations
of users of an ultra-high-resolution wall-size display,
raising the question of the advantages of this type of display
over a traditional desktop display.
We ran a controlled experiment comparing physical navigation
in front of a wall-size display vs. virtual navigation on
a desktop display for a data classification task. Our results
show a robust interaction effect, such that the desktop is more
efficient for easy tasks, but the wall is significantly more ef-
ficient (up to 35%) for difficult tasks. We tested three other
desktop techniques with the difficult task in a follow-up experiment,
but none could compete with the wall-size display
This is but a first step in understanding the interaction environment
provided by wall-size displays