*Current address: Dept. of Computer Science, Chalmers
University of Technology, S-412 96 Göteborg, Sweden
Research has suggested that rapid, serial, visual
presentation of text (RSVP) may be an effective way to scan and
search through lists of text strings in search of words, names,
etc. The Alphaslider widget employs RSVP as a method for rapidly
scanning and searching lists or menus in a graphical user interface
environment. The Alphaslider only uses an area less than 7 cm
x 2.5 cm. The tiny size of the Alphaslider allows it to be placed
on a credit card, on a control panel for a VCR, or as a widget
in a direct manipulation based database interface. An experiment
was conducted with four Alphaslider designs which showed that
novice Alphaslider users could locate one item in a list of 10,000
film titles in 24 seconds on average, an expert user in about
widget, selection technology, menus, dynamic queries
Selecting items from lists is a common task in today's
society. New and exciting applications for selection technology
are credit card sized phone directories; personal digital assistants
such as the Apple Newton with complete telephone, address, and
business registers; handheld computers for maintenance workers
with selections of prepared reports, objects, maps, and drawings;
selection mechanisms for Laser Disc players where frame numbers
between 1 and 54,000 need to be selected rapidly and electronic
calendars where hours, days, months and years must be selected
rapidly and accurately.
Obviously there is a need for methods for selecting
items quickly and accurately, without a keyboard and in a small
space. Traditional computers with large screens have used methods
such as scrolling lists, menus, and keyboard entry to select items.
For new emerging handheld technologies space is limited which
makes scrolling lists and menus hard to implement effectively.
Much of the research done on selection mechanisms
has focused on menus [5,16]. To make menu selections effective
various techniques have been explored, such as menus with different
ratios of breath and width, and menus where items are sorted by
how frequently they are selected [16,20]. The RIDE interface explored
in  allows users to incrementally construct strings from legal
alternatives presented on the screen and thereby elminate user
Scrolling lists [2,17] share many of the attributes
of menus and are often used for selecting items from lists [Figure
1]. Research has shown that items in scrolling lists should be
presented in a vertical format , items should be sorted ,
and that 7 lines of information is more than adequate for retrieving
alphanumeric information . Introducing an index to the scrolling
list can shorten the search time .
The Alphaslider [Figure 2] was first proposed by . It is used to rapidly scan through and select from lists of alphanumeric data. The essential components of an Alphaslider are a slide area, a slider thumb, a text output and an index to the elements that the slider operates over. Whereas a traditional slide area lets users page through the content of a scrolling list, the Alphaslider slide area lets users move directly to a certain part of the slider range by clicking in it. The index below the slide area guides that operation. The index, as shown in  and earlier proposed in , is proportionally spaced to the number of items that start with each character. The most infrequent starting characters of the items in the searched list does not show up in the index. The value of the Alphaslider is reflected in a single line text item, which should update immediately upon user movement of the slider thumb.
Figure 2: Alphaslider for selecting movie titles
The Alphaslider can be used in direct manipulation
database querying systems, such as Dynamic Queries . Applications
of Dynamic Queries have so far been limited to domains where the
attributes of the database are numerical, such as real estate
databases  and the chemical table of elements , but the
Alphaslider makes it possible to query alphanumeric attributes,
such as names, titles, and objects.
Although selecting words or names with an Alphaslider
might in some cases be slower than typing on a keyboard, the use
of an Alphaslider has several advantages compared to a keyboard.
Using a keyboard, inexperienced users must search the keyboard
for the appropriate key and the keyboard does not prevent misspellings.
Users may type a value for a field that is inappropriate such
as a number when a person's name is required . An Alphaslider
by definition contains all valid input choices and can continuously
have its query range updated, which effectively eliminates queries
that will result in an invalid or empty query result.
Some major design constraints for the Alphaslider
are the small size, one line of text output, and the mapping of
a large number of items to a small number of pixels, i.e. each
movement of the slider thumb corresponds to a large number of
items. Alphasliders, just as many other controls, should be operatable
without looking at them continuously. This is important if the
Alphaslider is used in a direct manipulation interface such as
a public information system or a control panel for a medical image
retrieval system, where users want to concentrate on the output
rather than the input - because they are visually separated.
The issue with the richest set of design possibilities
is how the slider thumb should be controlled. The Alphaslider
described in  was implemented with up to 320 entries, which
mapped one item to each pixel. In some applications this is sufficient,
but as has been argued above, in many emerging technologies there
is a need for a much larger range. This causes a problem when
there are more items than pixels. Traditional scroll bars solve
this by allowing users to click on the arrow buttons to change
the view without scrolling the slider thumb which is a good solution
in some cases.
Another solution is to separate the user's movement
of the mouse (trackball, finger on a touchscreen) from the display
of the slider thumb - so that when the mouse is moved the position
in the list is changed, but not necessarily the position of the
slider thumb. This technique makes it possible to map tens of
thousands of items, if not hundreds of thousands, to a slider.
The items are easily selectable and with proper feedback the task
can be accomplished rapidly. This class of techniques has the
advantage of users being able to operate the control without looking
The small size of the Alphaslider calls for a compact
text display, i.e. one line of text output. Displaying text in
a one-line display can be done in either of the following ways:
(i) by rapidly displaying text at a fixed location, referred to
as RSVP, rapid, serial, visual presentation. RSVP has been used
by psychologists to study reading behavior and it has been shown
that people can read text presented in RSVP format at approximately
the same speed as they can read text presented in page format
[11,15], (ii) by scrolling text horizontally from the right to
the left, referred to as Times Square. Reading comprehension using
Times Square with a smooth scrolling can be at least as high as
for RSVP, and with a higher user preference , and (iii) by
scrolling text vertically - a technique that is rarely used for
one-line displays .
For situations where the viewing window is narrow
and presentation rate is high, we conjecture that RSVP may be
a more suitable display method and also an efficient way to search
through lists . Accordingly, the Alphasliders described in
this paper all use RSVP as the display method.
An experiment was conducted to compare different
designs making it possible to map 10,000 items to a small number
of pixels in an Alphaslider.
The interfaces used in the experiment were built
using the Galaxy user interface development environment with the
Motif look and feel. A Sun Microsystems SparcStation with a 17-inch
color monitor and optical three button mouse was used. The resolution
of the screen was 1180 x 876 pixels. A 14 point Times Roman Medium
font was used to display the text. The experimental setup used
10.5 cm x 3.5 cm of the screen, while the Alphaslider used 7.5
cm x 2.5 cm within the larger area.
Four different designs of the Alphaslider were included
in the experiment [Figures 3 to 6]. Their look and feels were
similar in several aspects. The text output was one line RSVP
in all cases and was displayed over the slider. Under each slide
area, an index provided cues about the distribution of the elements
alphabetically. A timing mechanism for the experiment included
two buttons for each interface. The target title was display directly
above the Alphaslider value to minimize vertical eye movement.
The non-scrollbar Alphasliders would move the slider thumb directly
to where the mouse was clicked in the slide area. All the interfaces
were based on the Motif look and feel .
Position interface. The
first interface [Figure 3] allowed subjects to select the granularity
of their mouse movements by initiating dragging in different parts
of the slider thumb. The top part of the thumb corresponded to
the coarse granularity of 100 elements per mouse movement, the
middle part to the medium granularity of 20 elements per mouse
movement, and the lower part to the fine granularity of one element
per mouse movement. While dragging, the active part was turned
Figure 3: Position interface. Users select granularity
by clicking in different parts of the slider thumb.
Scrollbar interface. The
second interface [Figure 4] was based on the standard Motif scroll
bar [17, page 4-5]. To select and move by the coarse granularity,
subjects would drag the slider thumb. To move by the medium granularity,
subjects clicked or held down the mouse button on the slide area,
on either side of the thumb, and finally to move by the fine granularity
subjects would click on the arrow buttons at the ends of the slide
area. With this interface subjects were not able to move directly
to a particular part.
Figure 4: Scrollbar interface. Users select granularity
in a fashion similar to traditional scrollbars.
The third interface [Figure 5] let subjects
select granularity by moving the mouse at different speeds. If
subjects moved the mouse more than a certain trigger level of
pixels in one mouse event, the granularity would be changed to
the medium granularity, and if the speed reached a second trigger
level, the granularity would be changed to coarse.
Figure 5: Acceleration interface. Granularity is
proportional to the velocity of the mouse movements.
The fourth and last interface [Figure 6] allowed
subjects to change the granularity of their movements by moving
the mouse vertically - moving up or down switched to coarse and
fine granularity respectively. Upon release of the mouse button
the granularity switched back to medium. A simple stabilization
algorithm allowed users to move the mouse vertically without effecting
the setting of the Alphaslider.
Figure 6: Micrometer interface. Users select granularity
by moving the mouse vertically.
A very basic model for comparison of the time to
locate an item with different Alphasliders, Tlocate,
estimates it to be the time spent dragging and moving the slider
thumb to the correct position. Dragging can be estimated with
Fitt's Law [9,14], but dragging done with the Alphaslider differs
substantially from tasks described in those papers. A simple estimate
of Tlocate for comparison
purposes is the time users spend moving the thumb to approximately
the right spot, Trough-aim,
plus the time spent adjusting the thumb to find the correct item,
Based on these assumptions, the following hypotheses
were stated for expert mouse users:
Subjects were required to have previous mouse experience,
and having worked with mouse most probably implies that it was
done in a graphical user interface environment. Consequently,
subjects were expected to have used scrollbars before, which could
lead to better performance for the Scrollbar interface.
For subjective evaluations it was expected that the
Scrollbar interface would be preferred due to its similarity to
many commercially available scroll bars - especially the Windows
3.0 scroll bar which many subjects were assumed to have used previously.
The independent variable was the type of interface:
(i) Position interface, (ii) Acceleration interface,
(iii) Micrometer interface, and (iv) Scrollbar interface.
The dependent variables were:
(i) time to locate an item in the list
(ii) subjective satisfaction.
For each interface 25 tasks were generated by presenting
random items from a list of 10,000 film titles averaging 19 characters
in length. The tasks were generated at run-time when subjects
pushed the start button. For each interface subjects were presented
with 5 practice tasks. The slider thumb was returned to the middle
of the slider before each task.
Twenty-four subjects participated and were paid $10
each. Experience in using a computer mouse was required. Subjects
were recruited from the University of Maryland campus and were
mainly non-computer science undergraduate students in the range
of 18 to 35 years old. Nine females and fifteen males participated.
A counterbalanced within-subjects experimental design
was used. A pilot study with four subjects was conducted. Each
session lasted 1.5 hours. Subjects read a general instruction
sheet, were presented with interface-specific instructions for
each interface and were then given five practice tasks to complete.
While reading instructions and completing practice tasks subjects
were free to ask questions. During the timed tasks for each interface,
subjects were not allowed to ask questions and were asked to work
as quickly as possible. The experimenter sat next to the subject
and observed the interaction. When finished, subjects filled out
a shortened QUIS-form . After using all interfaces, subjects
filled out a forced-choice preference rating for each possible
pairing of interfaces.
Analysis of the timed tasks was done using an ANOVA
with repeated measures for interface type. Observing the mean
time for each subject to complete 25 tasks for each interface
shows a significant main effect, F(3,69) = 17.2, (p<0.001)
Tukey's post-hoc HSD analysis was used to determine which interface was significantly faster. The Position and Scrollbar interfaces were found to be significantly faster than the Micrometer and Acceleration interfaces (p<0.001). Subjects used approximately 24 seconds to complete all tasks for the Position interface and 25 seconds for the Scrollbar interface. For the Micrometer and Acceleration interfaces subjects used approximately 32 seconds. An expert Alphaslider user - the first author - used approximately 13, 16, 14 and 19 seconds respectively for the Position, Micrometer, Acceleration and Scrollbar interfaces.
Figure 7: Graph showing mean time to complete all
tasks for each interface. Standard deviation indicators
on top of bars.
The pairwise forced-choice preference ratings were
converted to ranks and the Freidman test was used to determine
the extent to which subjects ranked the interfaces in the same
order. The results indicate that subjects consistently rated the
Scrollbar interface highest, the Position interface second highest,
and the Micrometer and Acceleration interfaces worst (c2
= 30.6, p < 0.001). The mean preference rankings were 1.3 (0.7),
2.45 (1.0), 3.1 (0.7), and 3.1 (0.9) for the Scrollbar, Position,
Micrometer, and Acceleration interfaces respectively (standard
deviations in parentheses).
Subjects completed their tasks on the average one
second faster for the Position interface compared to the Scrollbar
interface, although the difference was not statistically significant.
The success of both interfaces was probably due to the fact that
they both were found to be stable and predictable by the subjects.
Observing subjects revealed different behavior for the two interfaces.
The Position interface was appreciated by some subjects for the
possibility to fine-tune without releasing the mouse button, while
the scrollbar interface was appreciated by others for the arrow
buttons which made it possible to fine-tune the setting by repeated
mouse clicks instead of dragging.
The hypothesis predicted the Acceleration and Micrometer
interfaces to perform better than the Position and Scrollbar interfaces,
but this was not the case. An explanation for the Micrometer interface's
bad performance may be found in that subjects found it somewhat
complicated. The time for changing granularity, i.e. the time
for moving the mouse vertically (Micrometer), was expected to
be less than the time to release the mouse button, locate a new
target, and hold down the mouse button again (Position & Scrollbar).
But the functionality of moving the mouse vertically probably
interfered with subjects' notion of mouse movements and slowed
them down. A similar explanation can be found for the Acceleration
interface. By unintentionally triggering the acceleration mechanism,
subjects overshot their targets and were discouraged by fast mouse
The expert Alphaslider user performed nearly twice
as fast as the experimental subjects. Observing the expert user's
mean times revealed a different ordering of the interfaces' performance.
The order followed the predictions of the hypotheses, except for
the Position interface which was the fastest for the expert user
Comparisons to other selection mechanisms
Landauer & Nachbar let subjects select words
and numbers from menus with 4,096 items, using the whole screen
. When subjects selected words of length 4-14 characters,
average selection times varied from 12.5 to 23.4 seconds for different
menu structures. Doughty & Kelso had subjects select numbers
from 1 to 4,096 and selection times varied from 9 to 17 seconds
for different menu structures . Alphaslider subjects had to
select from film titles, probably a more difficult task, from
a list which was 2.5 times as big, only using a fraction of the
screen size, and their selection times varied from 24 to 32 seconds
- a performance that compares favorably.
Several subjects were frustrated by doing
fine tuning work with the mouse while holding down the mouse button.
Holding down the mouse button while moving the mouse is a fairly
complicated motor action, and subjects were found to repeatedly
release the mouse button by mistake, which has been observed in
other studies too . Releasing the mouse button while dragging
caused the cursor to leave the slider thumb and forced subjects
to initiate dragging again. Subjects' ability to do the necessary
fine tuning was also affected by holding the mouse button down.
Subjects were observed pressing the button too hard and thereby
generating friction between the mouse and the mouse pad. For the
Scrollbar interface this behavior was not observed, as subjects
clicked the arrow buttons to fine tune the value of the Alphaslider.
It is reasonable to conjecture that a good design of an Alphaslider
should include arrow buttons for fine adjustments.
Feedback about subjects change of granularity was
provided for the Position, Micrometer, and Acceleration interfaces
through a speed indicator in the slider thumb. Although the thumb
was very close to the displayed film title, it is obvious from
the results of the subjective ratings of the interfaces that this
feedback is not enough. Feedback is an important design issue
for the Alphaslider and will be discussed further below.
For the Position, Micrometer and Acceleration interfaces
subjects were observed to mainly use the middle and fine granularity
and for the Scrollbar interface mainly the thumb and the arrow
buttons. The functionality of moving directly to a certain part
of the slider was used extensively by subjects.
Position interface. The
Position interface allowed subjects to select one of three parts
of the thumb to set granularity, which was greatly appreciated.
Subjects stated "With this interface I can exactly determine
by what speed I'm going to move". It also caused some problems
because the selection areas were small. As subjects were found
to nearly always use the middle and fine granularities, this could
be addressed by just allowing subjects to select from two granularities
on the thumb - with accordingly large areas to select.
Scrollbar interface. Subjects
found it easy to do fine tuning with the Scrollbar interface,
they just had to click the arrow button and the elements would
flash by rapidly. Some subjects experienced problems having to
move the mouse between the end points of the slide area to change
directions - this was particularly the case for expert mouse users
who were more comfortable with the position interface where they
could change directions by just moving the mouse.
The acceleration interface was expected to
do well in performance, but the reverse occurred; it both performed
badly and was rated low. Subjects overshot the goal by mistake,
by moving too fast and thereby triggering the acceleration. Feedback
was provided in the slider thumb but, as subjects concentrated
on the text value of the slider, this feedback was overlooked
in many cases. Most subjects found changing granularity with the
Acceleration interface too abstract. Some subjects did adjust
very well to the acceleration interface and found it easy because
they did not have to do anything else than move the mouse horizontally
to set the value.
Whereas it was expected that the Micrometer
interface would perform well, some subjects found it surprisingly
difficult to operate. An experienced user operating the Alphaslider
can concentrate on the output without looking at the Alphaslider
itself. Subjects were confused by the different semantics of moving
the mouse vertically and horizontally.
When releasing the mouse button the Alphaslider returned
to the middle granularity, to avoid modes that the Alphaslider
could be left in. While this was not detected as a design flaw
in the design process and in the pilot experiment, during the
experiment it became obvious that this design caused frustration
for subjects - especially those who frequently released the mouse
button by mistake.
From both the forced choice ratings and the QUIS
analysis it is obvious that subjects preferred the Scrollbar interface.
One explanation for this is that the slide area and thumb part
of the Alphaslider was similar to other scrollbars subjects had
previously used. The particular feature of the Scrollbar interface
that subjects liked was the arrow buttons. One subject stated
about the Scrollbar interface: "This is the interface type
I am most familiar with, and thus I was able to apply many of
my personal strategies to it. It was neither as fast nor as intuitive
as #3 (Position interface), however".
Subjects' reactions to the Acceleration interface
were interesting. One subject stated: "Why accelerate at
all, as you can just click and go to a particular place directly?"
- he avoided the acceleration by clicking on the bar and then
moving the slider thumb slowly. Reflecting the opposite opinion,
one subject stated: "It's much easier than the other interfaces,
you just need to move the mouse [as opposed to other more complicated
schemes]". Subjects appreciated the stability of the Position
interface; "With this interface I can exactly determine by
what speed I'm going to move".
GUIDELINES FOR DESIGNERS
In the light of the experiment described above, a
redesigned Alphaslider can be proposed. It was obvious from the
experiment results that the arrow buttons of the Scrollbar interface
were helpful to users.
Figure 8: Redesigned Alphaslider
The Position interface performed well as it allowed
subjects to move directly to a particular part of the Alphaslider,
the value could be set by just moving the mouse, and it still
allowed coarse movement with the thumb.
The Alphaslider in [Figure 8] would allow subjects
to select either coarse or fine movement by selecting different
parts of the thumb. Fine tuning can also be done by clicking on
the arrow buttons.
The Alphaslider is a widget that makes it possible
to rapidly select items from long lists without a keyboard using
minimal screen space. Four different designs of an Alphaslider
were evaluated in a controlled experiment.
Lessons learned from the study tell implementors
and designers that Alphasliders are ready to be included in interactive
systems and user interface management systems. With good use of
feedback techniques, the Alphaslider is a powerful, compact, and
rapid way of selecting items from lists. The University of Maryland
is seeking to patent the Alphaslider.
We appreciate support from Chalmerska Forskningsfonden, Adlerbertska Forskningsfonden, and Kungliga Hvitfeldska Stipendieinrättningen which made this research possible.
This research was done in the "Widget Carvers
of College Park" group - Rich Chimera, Catherine Plaisant,
Ninad Jog, Harsha Kumar, and Marko Teittinen, who all contributed
at lively design meetings, late night hacking sessions, and early
morning email reports.
1. Ahlberg, C., Williamson, C., Shneiderman, B.,
Dynamic Queries for Information Exploration: An Implementation
and Evaluation, Proceedings ACM CHI'92 Conference, 619-626,
2. Apple Computer Inc., Human Interface Guidelines:
The Apple Desktop Interface, Reading, MA: Addison-Wesley,
3. Backs, R. W., Walrath, L. C., and Hancock, G.
A., Comparison of horizontal and vertical menu formats, Proceedings
of the Human Factors Society Conference 31st Annual Meeting,
4. Beck, D., Elkerton, J., Development and evaluation of direct manipulation lists, ACM SIGCHI Bulletin,
20 (3), 72-78, 1989
5. Card, S., User perceptual mechanisms in the search
of computer command menus, Proceedings ACM CHI'82 Conference,
6. Chin, J., Diehl, V., Norman, K., Development of
an instrument measuring user satisfaction of the human-computer
interface, Proceedings ACM CHI'92 Conference, 213-218,
7. Doughty, R., Kelso, J., An evaluation of menu
width and depth on user performance, unpublished student paper,
described in Shneiderman, B., Designing the User Interface:
Strategies for Effective Human-Computer Interaction, 1st ed.,
8. Elkerton, J., Williges, R., Information retrieval
strategies in a file-search environment, Human-Factors,
26, 171-184, 1984
9. Gillan, D., Holden, K., Adam, S., Rudisill, M.,
Magee, L., How does Fitts' law fit pointing and dragging?, Proceedings
ACM CHI'90 Conference, 227-234, 1990
10. Harpster, J., Random versus Ordered Menus in
Self-Terminating Menu Searches, Proceedings of the Human Factors
Society 31st Annual Meeting, 718-721, 1987
11. Juola, J., Ward, N., McNamara, T., Visual search
and reading of rapid, serial presentation of letter strings, words,
and text, Journal of Experimental Psychology: General,
111, 208-227, 1982
12. Kang, J., Muter, P., Reading dynamically displayed
text, Behaviour and Information Technology, 8(1), 33-42,
13. Landauer, T., Nachbar, D., Selection from alphabetic
and numeric menu trees using a touch screen: breadth, depth, and
width, Proceedings ACM CHI'85 Conference, 73-78, 1985
14. MacKenzie, S., Abigail, S., Buxton, W., A comparison
of input devices in elemental pointing and dragging tasks, Proceedings
ACM CHI'92 Conference, 161-166, 1992
15. Masson, M., Conceptual processing of text during
skimming and rapid sequential reading, Memory & Cognition,
11, 262-274, 1983
16. Norman, K, The Psychology of Menu Selection:
Designing Cognitive Control at the Human/Computer Interface,
17. Open Software Foundation, OSF/Motif Style
Guide, Englewood Cliffs, NJ: Prentice Hall, 1990
18. Osada, M., Liao, H., Shneiderman, B., AlphaSlider:
searching textual lists with sliders, Proceedings of the Ninth
Annual Japanese Conference on Human Interface, October 1993
19. Sears, A., Kochavy, Y., Shneiderman, B., Touchscreen
field specification for public access database queries: let your
fingers do the walking, Proceedings of ACM CSC'90 Conference,
20. Shneiderman, B., Designing the User Interface:
Strategies for Effective Human-Computer Interaction, 2nd ed.,
21. Welty, C., Correcting user errors in SQL, International
Journal of Man-Machine Studie, 22(4), 463-477, 1985
22. Williamson, C., Shneiderman, B., The Dynamic
HomeFinder: Evaluating dynamic queries in a real-estate information
exploration system, Proceedings of ACM SIGIR Conference,