AlphaSlider is a query interface that uses a direct manipulation
slider to select words, phrases, or names from an existing list.
This paper introduces a prototype of AlphaSlider, describes the
design issues, reports on an experimental evaluation, and offers
directions for further research. The experiment tested 24 subjects
selecting items from lists of 40, 80, 160, and 320 entries. Mean
selection times only doubled with the 8-fold increase in list
length. Users quickly accommodated to this selection method.
Keywords: keyboard, data entry, touchscreen, direct manipulation,
Contact: Ben Shneiderman, firstname.lastname@example.org
As portable, palm-top, and pocket computers become more popular, the pressure to create and easy-to-use interfaces with non-keyboard input devices has grown. For example, a pocket-sized electronic phone book would be more effective if users did not have to type names, but could select them easily. Similarly, medical image workstations in hospital settings need direct and instant responses for physicians to make prompt and accurate diagnoses. Physicians should not have to take their eyes from the images to operate workstations, and many of them do not want to operate keyboards to retrieve patient data.
Voice recognition and handwriting recognition are often suggested as good possibilities for non-keyboard input. However, both of these options are limited in their practical applications at the present time because recognition and response rates are not fast enough, and end-users also have to adjust their speech or handwriting patterns to make recognition reliable. In contrast, direct manipulation interfaces with graphic objects have been introduced to many places such as bank teller machines, information kiosks of museums and libraries, etc.
We believe that AlphaSlider is practical for small portable devices
the size of a credit card, and also as a component for larger
and more complex applications. Many refinements and extensions
seem possible, but inclusion of similar widgets in graphic user
interface management software and toolkits seems appropriate.
2. Previous Research
Dynamic queries (Ahlberg, Williamson & Shneiderman, 1992; Williamson & Shneiderman, 1992) is a database searching technique using direct manipulation. It provides direct manipulation widgets such as sliders, and gives immediate feedback which inform users how close they are to finding the data they are searching for. Users can adjust the sliders, thereby performing dozens of queries, and see results immediately with no syntax errors. The data used in this application was numerical information, but dynamic queriea are also applicable to text data entry and retrieval.
A touchscreen keyboard, which is one of the solutions for alphanumeric
data entry without physical keyboards, has been explored by Weisner
(1988), and Sears Revis, Swatski, Crittenden & Shneiderman
(1993). The former study showed that a QWERTY touchscreen keyboard
was acceptable for entering limited quantities of alphanumeric
data. The latter study also showed that it was usable, that the
keyboard on a screen could be reduced in size without increasing
errors, and that the performance was 25 words per minute(wpm)
by computer science students, which was better than mouse-activated
keyboard (17wpm), and less than the standard keyboard (58wpm).
Plaisant & Sears (1992) demonstrated that repeated use of
a touchscreen keyboard over a two hour period did not lead to
fatigue and that users could cope with complex typing tasks involving
capitals, numerics, and special characters.
3. 800-Number Yellow Pages: An example of AlphaSlider
We chose a potential consumer electronics application that would involve a small credit card sized pocket computer with travel information such as toll-free telephone numbers for car rental, airlines, hotels, etc. A screen from the prototype of AlphaSlider, to retrieve 800-Number Yellow Page entries is shown in Figure 1. The horizontal bar is 10 pixels high and 320 pixels long. Just underneath this bar the letters 'A' through 'Z' are displayed with spacing between each letter proportional to the number of entries in the database. The slider button is 15 pixels high and 5 pixels wide. Two left and right triangular fine tune buttons are attached to the main slider button. Users can drag the slider button to go from one letter to another. They can also point to any place in the bar so that the slider button will move automatically to the selected location. The two fine tune buttons reappear beneath the slider button when the slider button is released.
Four versions of the 800-Number Yellow pages were created on an Apple Macintosh II fx using Aldus SuperCard, a prototyping package. A mouse was used to manipulate the interface. Response time for clicking and dragging was 17msec, giving the users the perception of smooth movement. All of the four slider bars and slide buttons looked the same, and were manipulated in the same manner. The only difference was the width of each list entry in pixels on the horizontal bar (1, 2, 4, and 8 pixels). Since the total width was fixed at 320 pixels (approximately 12 cm), this corresponds to 320, 160, 80, and 40 entries.
The question field (seen as "Find: xxxxx) and Pause/Resume button were put on the top for experimental purposes. Subjects were required to find entries appearing in the question field, then push the 'Dial' button. If the selected entry matched the question, the system emitted a bell-like sound and the next question appeared. Otherwise, the system emitted a short beep and the question field remained unchanged. Users pushed the 'Pause' button when they wanted to stop working, and then pushed 'Resume' button, to continue.
Figure 1. 800-Number Yellow Pages
4. Theoretical Estimates of End-User Performance
Figure 2 shows how AlphaSlider would be used. Users start searching
by looking at a question field to find the target name. Then they
would click on the slider bar, aiming for the letter of the target
name. If the result box did not contain the target name, they
would either click fine-tune buttons or drag the slider button
to scroll through the list. When they found the targeted name,
they would click the Dial button and complete the question.
Figure 2. Transition diagram of 800-Number Yellow page usage
Estimated time for each action can be calculated using the Model
Human Information Processor (Card, Moran, and Newell, 1983). For
example, the action of clicking to a position nearest the targeted
word consists of :
- eye movement to look at the question teye =330msec
- storing the question in visual working memory tp = 100msec
- recognition of the alphabetical order talphabet
- eye movement to search mouse cursor tsearch
- eye movement to look at the bar teye=330msec
- locate the target tp+tc = 170msec
- move and click the mouse from Dial button to the bar
tm+Tpos1 = 70msec+602msec =672msec
- see the result teye +tp +tc=400msec
Tpos1 is a time estimation applying Fitts's law (Fitts, 1954;
Card, Moran, and Newell, 1983), where IM=200msec/bit, and D/S=33.
D represents a distance between the starting position (on the
Dial button) and a target position (on the bar). S represents
a width of a target. Since the questions appear in a random manner,
the average distance between the starting position and the target
position is calculated as Dmax+Dmin/2. The overall estimated time
for clicking on the slider bar was computed as
= 1802 + talphabet + tsearch msec.
In a similar fashion, the estimated time for fine-tuning the slide
button movement on the slider bar was calculated as
where m represents the number of clicking on the fine tune buttons.
The estimated time for dragging the button across the slider bar
where n represents the number of dragging movement made. The estimated
time for clicking on the "Dial" button is
teye+tp+tc+tm+Tpos4 = 570 msec.
Overall the estimated time to complete one question depends upon
how many times users click on the fine-tune, or drag on the bar.
The shortest time to complete one question is,
if users click the exact target on the bar so that there is no
clicking or dragging. If users click on the bar a point a pixels
far from the target, they have to click a/k times on the fine-tune
buttons or to drag a/k times, where k is the number of pixels
between each data entry on the bar. The estimated time to complete
one question using the dragging strategy is
t = tsearch+talphabet+3102 + (270+talphabet)a/k msec (a/k >1)
t = tsearch+talphabet+2372 msec (a/k =2)
The estimated time to complete one question using the clicking strategy is
t = tsearch+talphabet+3442 + (170+talphabet)a/k msec (a/k >1)
t = tsearch+talphabet+2372 msec (a/k =2)
The goals of this experiment were to (1) measure end-users' speed
in locating entries in the four versions of the 800-Number Yellow
Pages and (2) investigate whether they achieved a significant
improvement in performance across trial blocks.
5.1. Prototype Design
Four versions of 800-Number Yellow Pages were used with 40, 80, 160, and 320 entries, and 8 pixels, 4 pixels, 2 pixels, and 1 pixel between each entry on the bar, respectively.
During the experiments, questions were presented in random order,
and the slider buttons were reset automatically to the middle
of the bar at the beginning of each question, to eliminate sequencing
bias. Times and the usage of the fine tune buttons were automatically
Subjects were trained and had a practice session before the experiment.
Subjects performed 24 tasks using each of the four versions (a
total of 96 tasks), in a counterbalanced ordering.
24 subjects were used for the experiment. 12 were female and 12
were male. Their ages ranged from 25 to 52. 2 of them had little
computer experience, 16 had intermediate levels of experience
(having used 2-3 systems), and 6 of them were advanced computer
users. 21 of them did not speak English as their native language.
All subjects had at least 6-years English education and had lived
in the United States for at least 1 year. One subject, in addition
to the 24 subjects, was trained as an expert AlphaSlider user
so that experienced performance could be estimated.
5.4. Independent Variables and Dependent Variables
Independent variables of this experiment were
(1) Number of entries (number of pixels for each entry)
- 40 entries (8 pixels)
- 80 entries (4 pixels)
- 160 entries (2 pixels)
- 320 entries (1 pixel)
(2) Trial blocks
- first 8 tasks
- second 8 tasks
- third 8 tasks
Dependent variables of this experiment were
(1) Time to accomplish each task, and
(2) Number of times the fine-tune buttons were used in each task.
Subjects were given a brief explanation of the 800-Number Yellow
Pages and procedures for the experiment. Training included clicking
and dragging the slider button, dragging left and right, and clicking
on the slider bar, 'Dial' button, and Pause/Resume buttons. A
practice session was given to help users understand how the actual
tasks would be done and a standard consent agreement was signed.
After the experiment, participants filled out a questionnaire
on their computer skills and educational backgrounds.
The mean times and standard deviations for subjects to select one entry appear in Table 1 and Figure 3. An ANOVA for the number of pixels between entries showed a significant main effect for version (F(3,92)=34.4, p<0.01). The mean times for the experienced AlphaSlider user to select one entry are also shown in Table 1.
The mean number of times required by subjects to press the fine-tune
buttons per question are 4.1, 2.1, 1.1, and 0.23 where the number
of pixels between entries is 1, 2, 4, and 8, respectively.
Figure 3. Graph of mean time to complete one task
The mean times required for subjects to select one entry, grouped
by the time line in each session are shown in Table 2 and Figure
4. Improvements over the three trial blocks were statistically
significant (p < 0.05) for the 2 and 8 pixel cases.
Table 2. Mean times to complete one task grouped by time line
in each trial block (sec)
Figure 4. Graph of mean times (sec) to complete one question grouped
by time line in each trial block
The results of the mean times to select one entry (Table 1 and Figure 3), the mean number of times to push fine-tune buttons in one question, follow the formulae of performance estimation for overall time. Approximate values derived from the results are, a = 4 , tsearch = 2141, and talphabet = 1301. It takes as much as 1.3 seconds for subjects and the expert to recognize the alphabetical order of the entry because each entry has several words. The average number of words per entry is 3.7.
This experiment was run using a computer display and mouse input, but users of the proposed credit card size device might perform more rapidly. Using a hardware slider button would eliminate the search time for mouse cursor on the display and provide better tactile feedback.
Since the average entry had 3.7 words, users' performance might be computed as 17wpm (1 pixel), 23wpm (2 pixels), 29wpm (4 pixels), and 32wpm (8 pixels). This hypothetical performance would increase as the average number of words per entry increase.
The expert user performed 15-37% better than the mean for the
24 subjects suggesting that the AlphaSlider does require some
skill to master. The experiment monitors observed that the 24
subjects often used fine tune buttons while the expert user tried
to drag the slider and avoid using the fine tune buttons. This
suggests that a longer training period is needed for subjects
to master the slider widgets.
7. Conclusion and Future Direction
Results of this experiment suggest that AlphaSlider can give users a new input method to retrieve data, and its performance is at the same level as touchscreen keyboards. They also suggest that further research about AlphaSlider is needed. Important items to examine are:
- slider design when the number of entries exceeds the number of horizontal pixels
- performance when sliders and buttons are made of hardware devices
- combination of AlphaSliders and other information visualization techniquges.
The advantages of this AlphaSlider approach are that there are
no typing errors, that performance increases as the number of
words per entry increases, and that keyboards are not necessary.
Therefore, once data entries are entered to the computer system,
this approach is suitable for retrieving data by personal and
corporate names, technical terms, etc., which have more than 2
words and are often misspelled when typed out on a keyboard. The
disadvantages of the current implementation are that the number
of entries is limited by the number of pixels, and that it also
does not allow users to input words freely. Considering these
advantages and disadvantages, potential applications of AlphaSlider
are personal database retrieval on palmtop computers, corporate
database retrieval on electronic white boards, and patient database
retrieval on medical workstations, where data is already entered
into the systems and users are not allowed to use keyboards to
We thank the members of the Human-Computer Interaction Laboratory,
who made many useful suggestions, and the subjects for their cooperation.
We also appreciate Toshiba Corporation for its financial support
of this research.
Ahlberg, C., Williamson, C., and Shneiderman, B. (1992), "Dynamic Queries for Information Exploration: An Implementation and Evaluation," Human Factors In Computing Systems - CHI '92 Conference Proceedings, 619-626.
Card, S., Moran, T., and Newell, A.(1983), The Psychology of Human-Computer Interaction, Hillsdale, NJ: Lawrence Erlbaum Associates.
Fitts, P. M. (1954), "The information capacity of the human motor system in controlling the amplitude of movement," Journal of Experimental Psychology, 70, 193-242.
Plaisant, C. and Sears, A. (1992), "Touchscreen interfaces for flexible alphanumeric data entry," Proc. of the Human Factors Society - 36th Annual Meeting.
Sears, A., Revis, D., Swatski, J., Crittenden, R., and Shneiderman, B. (1993), "Investigating Touchscreen Typing: The effect of keyboard size on typing speed," Behaviour & Information Technology, 12, 1 (Jan-Feb 1993), 17-22.
Weisner, S. (1988), "A Touch-Only User Interface for a Medical Monitor," Proceedings of the Human Factors Society - 32nd Annual Meeting, 435-439.
Williamson, C. and Shneiderman, B. (1992), "The Dynamic HomeFinder: Evaluating dynamic queries in a real-estate information exploration system", Proc. ACM SIGIR Conference, 339-346.