Jeff A. Johnson
Sun Microsystems
2550 Garcia Ave., MS UMTV12-33
Mountain View, CA 94043-1100
415-336-1625
jeffrey.johnson@eng.sun.com
An experiment was conducted to determine which of several
candidate user interfaces for panning is most usable and intuitive:
panning by pushing the background, panning by pushing the view/window,
and panning by touching the side of the display screen. Twelve subjects
participated in the experiment, which consisted of three parts: 1)
subjects were asked to suggest panning user interfaces that seemed
natural to them, 2) subjects each used three different panning user
interfaces to perform a structured panning task, with experimenters
recording their performance, and 3) subjects were asked which of the
three panning methods they preferred. One panning method, panning by
pushing the background, emerged as superior in performance and user
preference, and slightly better in intuitiveness than panning by
touching the side of the screen. Panning by pushing the view/window
fared poorly relative to the others on all measures.
Panning user interfaces for touch-displays have not been studied much,
if at all. Some work has been done on scrolling techniques for more
conventional input devices. Informal (and unpublished) investigations
at Xerox in the late Seventies focused on whether scrollbars in the
Star user interface should treat the document as moving under the
window, or the window as moving over the document. Star's designers
chose the former [14], but the marketplace for mouse-controlled
graphical user interfaces eventually settled on the latter. Bury et al.
[3] compared scrolling user interfaces for keyboard-controlled
character-display terminals and found that subjects performed best when
keys indicated the direction of movement of the viewing area (e.g., the
"up" key moved it upward in the document and hence the document content
downward). Beard et al. [1] and Glassner [5] described panning and
scrolling user interfaces for specific task-domains. Beck and Elkerton
[2] compared user interfaces for traversing long lists. MacLean et al.
used panning user interfaces as an example in their analysis of [8]. However, all of this work has focused on user interfaces controlled through keyboards and indirect pointing devices such as mice, not touch-displays.
Though empirical studies of touch-display interfaces have been
performed [10, 11, 12], these have not included studies of panning and
scrolling. Several ways of using touch-displays have been proposed [13]
but panning through a flat information space is not among them.
Recently, a product that incorporates a user interface for controlling
panning has appeared in the marketplace [15], but nothing describing
how that design was validated has been published.
The goal of the current study was to determine the best -- meaning most
easily used -- user interface for panning and scrolling in a simulated
physical world displayed on a touch-controlled display.
Initial experience at FirstPerson (a Sun subsidiary) with a prototype
information-space controlled by a touch-display raised in a new form
the old issue of whether the scene should be treated as moving under
the view-window or the window as moving over the scene. The prototype
had been designed such that swiping one's finger across the display
caused the scene to shift. The issue was: What should the relationship
be between the direction of the swipe and how the scene shifts? Is the
user's finger swipe best regarded as pushing the scene background or
pushing the "camera" that is viewing the scene? With a Push Background
user interface, users swipe in the direction they want the scene to go.
This seems completely natural until one considers previous findings on
scrollbars and scrolling keys, as well as the fact that to bring
information onto the screen that is off-screen left, users would swipe
towards the right. With a Push Camera user interface, users swipe in
the direction of the off-screen information.
Figure 1.Two contrasting panning methods.
The correct interface depends on how users think about it: do they view
themselves as shifting the background or panning the camera? Therefore,
the right way to resolve the issue is via user testing. The designers
of the initial prototype chose the Push Camera approach, based on an
informal paper-and-pencil test they conducted using a few co-workers as
subjects. However, experience with the prototype suggested that a
significant number of users considered its panning user interface to
work "backwards" from what they expected.
In an attempt to resolve this issue, a more formal study was devised,
using subjects from outside the company who had no experience with the
existing prototype and so were not already familiar with its user
interface for panning. The idea of the study was that subjects would
first be asked -- before they had tried or seen any panning user
interface -- to describe one that would seem natural to them. Subjects
would then perform a panning accuracy test with different panning user
interfaces. Finally, they would be asked to say which panning user
interface they preferred of those they had tried.
As the study was being designed, it was decided that panning
user-interfaces other than Push Background and Push Camera should be
considered. One plausible alternative to panning by swiping a finger is
panning by touching the sides of the touch-display. In fact, this
user-interface for panning was already used in the prototype in certain
situations. When a user is dragging an object to a new location that is
off-screen, it is impractical to have to drop it, pan the scene, pick
the object up, and resume dragging. Therefore, in the prototype, when a
user drags an object to a side of the display and holds it there, the
"camera" pans towards that side. It is reasonable to consider having
panning work that way all the time rather than only when the user is
dragging something. This interface is best regarded as Touch Edge
Camera because the user feels as if hitting an edge of the display
pushes the "camera" or "viewing window" in that direction. This
interface was added to the set to be compared.
Another possible panning user interface is the opposite of Touch Edge
Camera: users press on the side of the display they want the background
to move towards. However, because this interface -- referred to herein
as Touch Edge Background -- is in the author's experience very
counter-intuitive and also seems incompatible with the goal of having
panning work well with dragging, it was not included in the
comparison.
Regarding the first part of the study, when subjects would be asked to
suggest a panning method, it is clear that the appearance of the
touch-display can influence what people suggest. This is what Gibson
[4] and, later, Norman [9] refer to as an "affordance": when an aspect
of an artifact's design suggests how it is to be used. We thought that
adding a brightly colored border around the displayed image might
suggest "touch here" to users, and might therefore suggest Touch Edge
panning (camera or background). We therefore designed the study so that
in the initial (elicitation) phase of the study, half of the subjects
would see such a border, and half would see the "normal" display (with
the displayed scene taking up the entire screen).
Figure 2.Touch-display apparatus.
Subjects participated in the study individually. A video camera was
focused on the display for the entire session, recording what happened
on the display as well as the subject's hand when near or on it. Each
session was conducted by an experimenter and an assistant (to operate
the camera and a timer).
Subjects were told at the start of the session that we (the
experimenters) "are testing various design ideas for our product, to
make the product easy to use." The experimenter explained that some
aspects of the prototype might be hard to use, but if so, that
indicated a bad design rather than anything wrong with the subject.
Subjects were seated in front of a prototype touch-sensitive display,
and their attention was directed to it. The session consisted of three
phases: Panning Elicitation, Panning Accuracy, and Panning Preference.
Subjects were shown that the living room scene was wider than the
display (by panning the scene remotely via keyboard commands to the
right and then back to the initial position). In other words, they were
shown the panning function without being shown a panning user
interface. They were then asked to show and tell how they would expect
to operate the touch-display to "bring that other part of the scene
back into view." Words such as "pan" that might suggest a particular
mental model or user interface were intentionally not used in the
instructions. Subjects' responses were categorized (see Analysis and
Results).
The experimenter then demonstrated to the subject how to operate the
first of the three panning user interfaces. Subjects were not allowed
to practice panning before starting the timed trials. Subjects were
instructed to move the line given on each trial to the indicated letter
position, and once the line was positioned to his or her satisfaction,
to say "OK" so that the assistant would know when to stop the timer.
The next trial began after a brief pause, with the scene positioned
where it had been left by the previous trial. Twenty three panning
trials followed (i.e., all lines moved to all reachable targets) in
which the experimenter stated the line number and target letter, the
assistant timed the trial, and the experimenter recorded the time on a
data sheet. All trials were videotaped to allow rechecking of the times
and collection of other data.
Figure 3.Panning accuracy test display.
After the first set of panning accuracy trials, the experimenter
changed the panning method, demonstrated the new panning method, and
commenced a second set of trials with a different panning user
interface. These were followed by a set of trials with a third panning
user interface.
Subject responses were grouped into five categories, corresponding to
the four panning methods under consideration plus an Other category.
The panning categories were:
Figure 4.Panning elicitation results.
The foregoing analysis treated all 12 subjects as one group. However,
the scene shown to half of the subjects had a bright blue border, which
might suggest a Touch Edge panning user interface. Also, orthogonally
to this grouping, half the subjects were of each gender. Neither the
presence/absence of a blue border nor gender appeared related to
subjects' suggested panning user interfaces.
For each dependent measure, a Friedman ranks test for matched scores
[6] was used to test for an effect of panning user interface type. To
perform this test, each subject's three scores on a given measure were
ranked, then the rank scores were submitted to a formula that yields a
chi-squared statistic indicating whether any interface had more low or
high ranked scores than would be expected by chance.
Figure 5.Panning accuracy: time data.
Table 1.Panning accuracy: time data (avg. over trials).
For all three measures, a significant effect of interface type was
observed (time: X2(2) = 13.17, p < .01; moves: X2 = 12.17, p < .01;
errors: X2 = 15.79, p < .01; see Figures 5-7). With the Push Background
panning user interface, subjects took less time, required fewer moves,
and made fewer direction errors than with the other two panning user
interfaces. The effect was particularly strong for the direction-errors
measure: when using the Push Background, subjects made almost no
direction errors, in sharp contrast to the other two panning user
interfaces (see Figure 7). A post-hoc t-test of difference scores
between the Push Camera and Touch Edge Camera interfaces showed no
significant difference between those two interfaces for any of the
dependent measures.
Figure 6.Panning accuracy: moves data.
Table 2. Panning accuracy: moves data (avg. over trials).
No gender differences appeared on any measure, either for overall
performance or effect of panning user interface type.
Figure 7.Panning accuracy: error data.
Table 3.Panning accuracy: error data (sum over trials).
One possibility worth checking is that the three panning user
interfaces might differ in how performance improves with practice. For
example, performance with one user interface might start out worse than
with others, but improve faster. Learning effects were examined by
comparing, for each subject, performance in the first half of the
trials with that in the second half. Since each user-interface-block of
trials consisted of an odd number (23) of trials, the "first half" was
defined as trials 1-11, and the "second half" was defined as trials
13-23. Trial 12 was ignored.
Overall, there was clear evidence of learning over the trial blocks.
Superimposing the trial blocks for the three panning user interfaces,
performance in the second half of the trials in a block almost always
exceeded that in the first half. Simple sign tests were significant for
all three performance measures: time (p < .01), number of moves (p <
.05), and direction errors (p < .01).
Since the distance from the starting position to the target position
differed from trial to trial, a performance difference between the
first and last half-blocks could result from a difference in the
distance panned in the two half-blocks. In this case, that explanation
can be ruled out, because the distance subjects had to pan in the two
half-blocks was almost equal (averaging 102.4 pixels/trial for trials
1-11 vs. 103.2 pixels/trial for trials 13-23) and was greater in the
second half-block anyway.
To determine whether the learning rate depended on the panning user
interface, subject's difference scores (first half-block minus second
half-block) were submitted to a Friedman ranks test. For all three
performance measures (time, moves, and direction errors), the test
showed no significant difference in learning between the three
interfaces. Another possibility worth checking is whether panning time
and number of moves depends on the distance panned, and whether the
dependency is affected by the panning user interface. It might be, for
example, that panning time is directly proportional to distance for one
panning user interface but not for another. For each subject,
regressions were computed for panning distance vs. time and distance
vs. moves for each of the three panning user interfaces.
Overall, both time and number of moves were positively related to the
panning distance. This was determined by a simple sign test on the
regression scores: many more of them were positive than would be
expected by chance (p < .01). On the other hand, Friedman ranks tests
applied to the regression scores showed no effect of panning interface
type on the strength of the relationship between distance and either
time or number of moves.
The subjects who preferred Touch Edge Camera panning were subjects 4
and 8. Since neither of these subjects (in fact, no subjects at all;
see Tables 1-3) performed better using Touch Edge Camera than with Push
Background, performance cannot be an explanation for their divergent
preference.
Of the ten subjects who were asked to state a preference, five were
male and five were female. The distribution of preference scores was
exactly the same for males as for females: four out of five preferred
Push Background, one out of five preferred Touch Edge Camera. Thus, no
gender difference in preference was observed.
Abstract
Keywords:
Touch display, touchscreen, panning, scrolling, navigation.
Introduction
Often, computer-based displays provide views of a scene or
information that is too expansive (wide or tall) to be shown in its
entirety. A common solution is to provide a panning or scrolling
function, allowing users to control which portion of the subject is
visible in the display. If such a function is provided, a user
interface -- a way for users to invoke and control panning -- must also
be provided.
METHOD
Subjects
The subjects were six males and six females, ranging in age from 18 to
65 years old. The distribution of subject ages was similar across
gender. Subjects were recruited by company employees, and were paid $15
for the one-hour test session. The subjects' occupations include
student, homemaker, retiree, clerical, salesperson, and business
planner. None are computer engineers, but some do use computers.
Design
The experiment consisted of three tests: Panning Elicitation, Panning
Accuracy, and Panning Preference. Since gender differences have
occasionally been observed in studies involving hand-eye coordination
[7], all three tests were designed to allow gender effects to be
distinguished from individual differences.
Panning Elicitation Test
The six subjects of each gender were randomly assigned to one of two
display types -- plain or bordered -- in such a way that three subjects
of each gender had each display type. The dependent measure was the
type of panning user interface the subject suggested (see Procedure and
Materials), i.e., a categorical variable. However, as is described in
the Analysis and Results section, the response categories were not
predetermined, but rather emerged as the study was conducted.
Panning Accuracy Test
Because high intersubject variability was expected in the panning
accuracy task, a within-subject design was used for that part of the
experiment. Each subject was tested with all three panning user
interfaces (Push Background, Push Camera, and Touch Edge Camera), with
the order of presentation of the interfaces counterbalanced across
subjects both within gender and overall. For each panning user
interface, there were twenty-three trials (see Procedure). The
following dependent measures were taken on each trial: time, number of
moves, and whether the subject started by shifting the scene in the
wrong direction (direction errors). Thus, the design of the panning
accuracy test was a three (panning user interface) by two (gender)
design, with panning user interface varied within subject.
Panning Preference Test
The panning preference test was a very simple design: Ask all 12
subjects (six of each gender) which panning user interface they
preferred of the three they had tried, to determine whether any
interface is systematically favored or disfavored.
Procedure and Materials
The display was a small (15 cm diagonal) color liquid-crystal screen
mounted in a flat case. The case added about 2 cm of border to each
edge of the display. A transparent touch-pad was affixed to the front
of the display. The display housing was mounted on a pedestal, which
held the display about four inches above the surface of a desk (see
Figure 2). The display and touch-pad were connected to a Sun
Sparcstation 2 workstation.
Panning Elicitation
A cartoon scene of a living room was visible on the display. For half
of the subjects, the living room display occupied the full area of the
display; for the other half, the display had a quarter-inch, bright
blue border that decreased the area available for the living room
scene.
Panning Accuracy
In this phase of the study, the experimenter placed a sticker on the
display's bottom bezel that marked three horizontal positions: "A",
"B", and "C". Then the experimenter started a program that changed the
displayed image from the living room scene to a set of evenly spaced
vertical lines, each labeled by a number (0 - 10 from left to right) at
its top end (see Figure 3). The experimenter directed the subject's
attention to the numbered lines and said that in this part of the test,
he would ask the subject to shift the scene so that certain numbered
lines were over specified letters. The experimenter explained that the
subject would be timed while shifting the scene to the target letter.
The experimenter continued: "We'll do that several times, then I'll
change the way the scene is shifted and ask you to shift it some more,
then we'll change it again and I'll ask you to shift it some more." The
subject was reminded that we were testing our designs, not the
subject.
Panning Preference
After completing all three sets of panning accuracy trials, subjects
were asked "Which of the three shifting methods do you like best?"
Their answers were recorded, both on the data sheet and on videotape.
ANALYSIS AND RESULTS
Panning Elicitation
When subjects were asked to indicate how they would operate the
touch-display to bring into view the part of the living room scene that
was off-screen-right, they gave a variety of responses. Note that at
this point, subjects had not yet seen any actual panning method for the
prototype display (though some had used scrolling mechanisms on
computers). All they had seen was that the living room scene was wider
than could be seen in the display at once, and that the off-screen
portion was the right side of the scene. Thus, their responses may be
considered to be based upon their prior experience and whatever the
prototype display suggested to them.
Six (i.e., half) of the twelve subjects indicated by their actions and
words that Push Background was the panning method they first thought of
when faced with the prototype touch-display. Three subjects suggested
Touch Edge Camera. The remaining three subjects suggested methods that
were classified as Other. A statistical test of the skewedness of the
observed distribution requires a null hypothesis specifying what
distribution would be expected by chance. Because a null hypothesis of
equal category probabilities seems naive and no other chance
distribution seems any more plausible, a statistical test wasn't
feasible. Nonetheless, it is notable (if not "significant" in the
formal sense) that half of the subjects suggested Push Background, and
that no subject suggested Push Camera or Touch Edge Background (see
Figure 4).
Panning Accuracy
Each subject's time scores were averaged over each trial block,
yielding, for that subject, an average time for each of the three
panning user interfaces. Similarly, each subject's moves scores were
averaged for each of the panning user interfaces. Direction errors were
summed for each panning user interface. Tables 1, 2, and 3 show each
subject's aggregated data for the time, moves, and error measures,
respectively. Note that though treatment means are included in the
tables for informal comparison purposes, the table columns are not
independent: all three scores in each table-row are from the same
subject.
Panning Preference
Two of the twelve subjects, one male and one female, were accidentally
not asked which of the panning user interfaces they preferred of those
they had tried. Of the ten remaining subjects, eight preferred Push
Background panning, two preferred Touch Edge Camera panning, and none
preferred Push Camera panning. A multinomial calculation showed that
this distribution is significantly skewed (p < .01).
