Content uploaded by Alistair Edwards
Author content
All content in this area was uploaded by Alistair Edwards on Mar 18, 2014
Content may be subject to copyright.
Unpublished paper
alistair@minster.york.ac.uk
How many ways can you use one button? Timing data for
button presses
Alistair D N Edwards, Yanyu Li
alistair@minster.york.ac.uk
Department of Computer Science
University of York
Heslington
York
YO10 5DD
Abstract
Modern small devices such as phones and watches inevitably rely to a great extent on
the input that can be achieved using a small number of simple buttons. To maximize
the range of input it is necessary to use the buttons in as many ways as possible:
single-clicks, double-clicks and so on. In designing for a mass market, it is really
necessary to know the timing parameters that should be applied to such types of
button press. That was the objective of this study. Data was collected from a number
of people with a wide range of ages and levels of computer experience. Detailed
results are presented for ‘average’ users. This might be used in interaction designs or
as the basis of further, more detailed research. A major conclusion, though, is that
there is a wide range of performance in the population at large and that systems
should adapt to individuals.
1. Introduction
Miniaturization of computer devices is making it possible to build devices which are
as small – and hence as portable – as the designer may choose so that it is often
ergonomic factors which dictate how small a device can be (Edwards, 2001). The
most elementary input device is the button – in terms of its construction and in the
sense that it essentially communicates a single bit per press. On simple, small devices
the designer may be limited to the use of a small number of buttons. To make the
most of them, it is necessary to use these buttons is a variety of ways. There are
variations on the elementary pressing of the button (double-clicking etc) which can be
used to broaden the input, but there is a question as to how many different button
presses are practical. The objective of the study described herein was to begin to
investigate the set of viable presses in terms of the timing of the presses.
There are many examples of small devices with button-based interfaces. Modern
watches often do more than telling the time, but size limitations restrict them to four
or five buttons for control. Phones have numerical keypads of perhaps 20 or so
buttons. That number is sufficient for the traditional use of the phone – dialling
numbers – but the same keypad is in practice over-loaded with the role of inputting
text as well, for sending Short Message Service (SMS) messages as well as writing
names in phone books etc. To input 52 upper- and lower-case letters, plus 10 digits,
punctuation, spaces and controls through 20 buttons is only possible if the buttons are
used in a variety of ways.
Edwards How many ways can you use one button?
2
Given that there is a continuing pressure to miniaturize devices, to make them more
portable, the problems with maximizing the use of simple buttons will always be a
challenge, limited by space and ergonomics. Other styles of input may become more
common, but there will always be a role for the button. (See Edwards, op. cit., for
further discussion).
A number of input styles and strategies have been developed to achieve this richness
of input. Most styles involve interaction in the sense that the input interpretation relies
on the output (e.g. stop pressing the button when the desired letter is visible on the
screen). However, the present study was designed to start at a more fundamental level.
The objectives were to ascertain the performance of users carrying out a selection of
button-pressing inputs and to see whether there were important differences in the
performance of different groups of users. Timing information was collected which can
be used as guidance as to what button-pressing timing constraints should be – and
whether they should be different for different groups of users.
Choosing appropriate time parameters is important, as there are trade-offs to be made.
Enforced delays (e.g. the time between two clicks or the duration of a single ‘long’
click) retards user input and may cause impatience and frustration. However, to
reduce such delays requires faster reactions which might result on more errors. Errors
can themselves be frustrating and time spent repairing them adds to the time of the
interaction.
Within this study we chose to investigate four forms of button-press:
• the single-click,
• the double-click,
• the triple-click and
• the long-click.
The idea was to collect data from a variety of users to ascertain the times they
achieved for each of these inputs. The objective was to collect data on the average
times it takes typical users to perform the different types of click. The results are
compared with the theoretical predictions of the Model Human Processor
It may be significant that there are so many devices in common use which rely on
these simple button interactions, and yet there is no published data on human
performance. Or perhaps that is simply an artefact of the fact that most such devices
are commercial products so that the work has been done but never openly published.
In the next section the types of button presses investigated are defined. The following
section describes the method of the experiment. The data collected had to be analysed
first in general terms so that decisions could be made, such as who can be considered
as ‘average’ users. Then their results are discussed in detail. The paper concludes with
a discussion of the meaning of the results, a presentation of some limitations of the
study and some conclusions, including suggestions for follow-up work.
Edwards How many ways can you use one button?
3
2. Characterization of button presses
2.1 The single-click
The single-click is a simple press, transmitting 1 bit. The only parameter is c, the time
that the button is held down, as in Figure 1.
Button up
Button down
c
Time
Figure 1. The single-click: characterized by the time, c, that the button is
depressed.
2.2 The long-click
The long-click differs from the single-click only in the time that the button is held
down, l (Figure 2). Experiments will be required to ascertain what values of l are
sufficient to be reliably distinguished from c. For instance, if it is found that there is a
wide variance in c for different people, then it may be necessary to set l to a value
much greater than the longest likely value of c, so that single-clicks are not mis-read
as longs.
l
Figure 2. The long-click, which is identical to the single-click, except
that is duration, l, must be longer than c.
It seems likely that there will be different values of l with and without feedback. In
the with-feedback condition, the user would be given auditory or visual feedback
when the depression time was (sufficiently) greater than c. Without feedback, it is up
to the user to judge when they have depressed for long enough, that they have held it
for at least the maximum allowed value of c. The hypothesis would be that without
feedback, users will hold the button for longer, ‘to be on the safe side’.
Edwards How many ways can you use one button?
4
2.3 The extended click
An extension of the long-click is the extended click, whereby the duration of the press
(beyond l) becomes significant. In other words a press of t ms has a different meaning
from one of t + ? ms, which in turn is different from t + 2?, and so on. Conventionally
such presses are often interpreted as multiple single-clicks; the button ‘auto-repeats’
and generates single-clicks until released (think of holding the button on a scroll bar
arrow). The extended click was not investigated in this study because its application
depends on feedback – some means of indicating when sufficient work has been done
and the button can be released.
2.4 Double-click
A double-click is quite distinct from the other clicks. In other words, it is not two
single-clicks – in timing, and in meaning.
1 2 3
2
ddd
1
D
Figure 3. The double-click. This resembles two single-clicks, and must
be distinguishable.
Three times can be measured: the duration of the clicks (d1 and d3 in Figure 3) and the
separation between them (d2). It seems likely that it would be appropriate that d1= d3
= c, since presumably c is set at the minimum. That can be verified by experiment. It
seems likely that the critical parameter is d2. On the one hand, the shorter this is the
less likely it is that a double-click will be mis-interpreted as two single-clicks, but on
the other, if it is too short, it may be too difficult for some, less dextrous users to
achieve.
It may be that there are other, better ways to characterize the double-click. For
instance the total duration (effectively d1 + d2 + d3) is important, and two presses
within that total duration should be counted as a double-click. In other words,
symmetry becomes unimportant and it does not matter (for instance) if d1 ? d2.
2.5 Triple-click
The viability of the triple-click is questionable (Figure 4). Anecdotal evidence
suggests that producing it reliably seems to be difficult, with much scope for mis-
interpretation. Nevertheless, it is worth carrying out some experiments to measure the
tolerances. Some of the same considerations as for the double-click may be relevant
(e.g. using the total time).
Edwards How many ways can you use one button?
5
1 2 3
23
ttttt
4 5
1
Figure 4. The triple-click.
2.6 Combinations
One of the problems is distinguishing combinations of clicks. In the example in
Figure 5 the intention is to perform a single-click followed by a double-click, so it is
necessary that c2 be sufficiently long that the first two clicks are not interpreted as a
double-click.
cddd
123
c
1 2
123
Figure 5. Combination clicks. In this case the intention was for a single-
click followed by a double-click, which might be characterized by the
separation between them, c2 being sufficiently long.
If clicks 1 and 2 are interpreted as a double-click then presumably the system will be
put into a mode, and click 3 interpreted as a single-click within that mode. In other
words, the state of the system will be very different from what the user expected.
The values of the times, ci and di will probably also affect the user’s ability to detect
and repair errors. In other words, if c1 + c2 + d1 is sufficiently long, then the user
might have time to recognize that clicks 1 and 2 have been (mis-) interpreted as a
double-click and hold back from issuing click 3. However, this seems unlikely. Since
we would want to minimize these times, it is likely that momentum will push the user
into making click 3, even if and when the error has been recognized. It may be that
the cognitive models such as the Model Human Processor (Card, Moran et al., 1983)
or Interacting Cognitive Subsystems (Barnard, 1985; Barnard and May, 1994) might
give us some data as to how quickly users might react. However, such analyses are
beyond the scope of this paper (though the Model Human Processor is discussed
further in Section 6).
Edwards How many ways can you use one button?
6
Other combinations might be required (e.g. long+double, double+single etc.) but we
did not consider these at this stage, for the sake of simplicity.
2.7 Morse Code
Morse Code is a communication medium based solely on the use of a single switch.
As such it is rather different from the use of a switch explored in this study: Morse
messages are long, typically comprising hundreds or thousands of button presses and
the people who generate them are highly trained. However, it is useful to note that a
typical duration of a ‘dot’ of Morse Code (corresponding to a single-click) is 62.5 ms
(known as 1 baud), and that a ‘dash’ (which might be likened to a long-click) is
defined to be three times as long as a dot (i.e. 3 bauds), or 187.5ms. The gap between
the dots and dashes making up a letter is 1 baud, between the letters of a word, 3
bauds and between words 5 bauds (312.5 ms). These times are all calculated using
data to be found in Morsh and Stannard, 1948.
3. Method
The focus of the present study was the ‘natural’ behaviour of people. That is to say
that minimal assumptions were made or imposed on users about the characteristics of
the press types, as discussed above. One objective was to collect this data which will
then be used in future experiments which will be more prescriptive. For instance, if it
emerges that the mean single-click time is found to be c ms, then a future experiment
might require users to perform all single-clicks in c ms, and then the number of errors
will be counted. (In truth, future experiments are likely to be more complex than this,
as described in Section 8).
Participants were introduced to the experiment and its objectives and told that they
would be asked to perform each of the click types, described as follows:
Single-click
Press the button once, normally.
Long-click
Press the button and hold it however long you want.
Double-click
Press the same button twice quickly.
Triple-click
Press the button three times quickly.
It was important to collect data from a wide population, in terms of age and of
experience of computer usage. To that end, participants were recruited from a wide
variety of sources, including vacation residents in university accommodation, children
waiting to play in a sports centre and tourists on York railway station! The age of the
participant was recorded, as was their level of computer experience. For the purpose
of analysis, eight age groups were identified, as summarized in Table 1, and the
categories of computer experience are summarized in Table 2.
Edwards How many ways can you use one button?
7
Group Age range No of
participants
A <10 12
B 11–20 18
C 21–30 15
D 31–40 12
E 41–50 13
F 51–60 12
G 61–70 10
H >70 8
Table 1. Age characteristics of the test participants.
0 Never.
1 Fewer than 12 times per year.
2 Fewer than 3 times per month.
3 One per week.
4 Approximately 3 times per week
5 Daily.
6 More than once per day.
Table 2. Categories used to classify computer experience. Participants
were asked to estimate their current average frequency of computer use.
Participants were asked to perform their clicks using the Caps Lock key on a
computer keyboard. This key was specified simply because this was the only key for
which it was possible to collect the timing data needed, as it is the only key that
would not ‘auto-repeat’ to the Java software used, when held down. The experiment
consisted of the participant being presented a prompt for the kind of click required, as
illustrated in Figure 6 and then producing the required click(s). The different types of
clicks were interspersed and presented in a random order. Each type of click was
requested 12 times of each participant. The test took approximately 3 minutes.
Edwards How many ways can you use one button?
8
Please Click Caps Lock Key:
Please make a triple-click.
Thank you!
Figure 6. Appearance of the screen used in the experiment.
The experiment was carried out using a Compaq Armada 1590DT laptop computer,
with a 133 Mhz Pentium processor, 16 Mbyte of memory and a 12-inch colour
monitor. The software was developed in Java using the Java Development Kit 1.1.1.
4. Overall results
The numbers of older people it was possible to recruit were small, so that it is not
possible to treat their results in any statistical depth. It is clear that their times were
generally slower, so it is probably safe to conclude that systems to me used by older
people should be designed to be more tolerant of slower responses – though not at the
expense of younger and quicker users.1
The results for children are similarly inconclusive. Not only is there clear variation in
the data, but there was evident contamination in their performance in that they had to
be prompted and assisted by the Experimenter and/or parent. Again, though, the
important conclusion is that the behaviour of children is different and the
software/interface designer should bear this in mind.
The data for the long-click is also inconclusive. This is mainly because the definition
of the long-click had been left (deliberately) vague. The experiment was also a little
unrealistic in that no feedback was given on the completion of a long-click. In a real
application long-clicks usually involve the user holding the button until feedback is
generated (visually or as an auditory signal). What we have collected is some
behavioural information about people’s perceptions of the length of a long-click.
Other data collected in the study can be used in future for further investigation of the
long-click. That is to say that the important characteristic of a long-click is that it must
1 The need to accommodate the needs of the growing number of older people could be the topic of
a paper in itself. We will not follow that side-line herein. The interested reader is pointed to the
study which can be found at http://www.trace.wisc.edu/docs/pacbell_ud/agpd.htm, in which more
tolerant time constraints introduced for older customers were found to improve service for ALL
users.
Edwards How many ways can you use one button?
9
be longer than the single-click, to minimize mis-classifications. The data on the
single-click can therefore be used to design the long-click.
Table 3 presents the statistics for the single-click experiments. The first feature to note
is that all the means, modes and medians are approximately equal. This is a
characteristic of the normal distribution and therefore in all subsequent treatments we
make the assumption that they are, indeed normal.
From the table, it is also evident that there were a wide range of timings, with the
slowest average time being 356 ms (Group H) and the fastest being 127 ms (Group
C). This immediately suggests that a single time parameter for single-clicks for all
users would not be appropriate. Variance is important too. It is evident that the
distributions for groups A, G and H are somewhat broad. It is therefore suggested that
the ‘average’ user is represented by Groups B to F (i.e. age range 11–60) and
therefore the following discussion will concentrate on an amalgamation of these
groups.
It could be argued that selection according to age is inappropriate and that computer
experience might be more relevant. The performance of (say) a 30-year-old regular
computer user might be much quicker than that of someone of similar age with no
experience. The former might be rather more typical, while to include the latter’s data
could distort the results. The problem is that it proved very hard to find people with
no experience of using computers. Furthermore, the participants identified as having
no computer experience were predominantly among the older ones. (Eight of the 10
members of Group G). We would contend, therefore that the results presented herein
are relevant to people in the age group identified, regardless of their level of computer
experience, but that the typical, more experienced user might be expected to be
slightly quicker than the averages presented. As an indication, averages of those few
participants with little computer experience are compared in Section 5.5.
Group
A B C D E F G H
Age < 10 10–20
21–30
31–40
41–50
51-60 61–70
> 70
Mean 290
146
131
138
201
215
242
348
Mode 220
110
110
110
110
220
220
270
Median 270
110
110
110
160
170
220
330
Interquartile range 213
50
100
60
110
110
120
170
Standard deviation 177
110
103
77
107
188
156
200
Table 3. Summary data for single-clicks (ms).
5. Detailed results
It might be presumed that the click timings would be normal in distribution. It is not
clear that this is the case, on examination of the data in Table 3. In a normal
distribution the mean, mode and median will be equal. Standard deviations are given
in the table and are somewhat broad in some cases. As an alternative indication of
width of the distribution the interquartile range is also given. For the purposes of this
Edwards How many ways can you use one button?
10
study, though, the type of distribution is not important. The assumption that will be
made is that the data collected is representative of the population, and all
recommendations made on the basis of the distributions observed, rather than an
idealized normal distribution.
The data collected can be used to define what will be categorized as each kind of
click. It must be borne in mind that different systems may react in different ways. For
instance, for mouse buttons it is often the button release which initiates the action.
This means that the user is not generally committed by the button press; by holding
the button down and moving the cursor off the current widget the action can be
cancelled, the action of the widget will not be executed. This may affect the way
presses are treated. For instance in the discussion below it may be suggested that a
double-click can be recognized as soon as the button is depressed for the second time
(within a certain time), but in practice the associated action may not be initiated until
the second release of the button.
5.1 Single-clicks
One conclusion of this study is to suggest timings which can be used to define the
different click types. In the case of the single-click that amounts to saying that any
click briefer than some value c will be treated as a single-click, and any longer than c
will be long-clicks.
Table 4 and Figure 7 presents the statistics for the single-click for Groups B to F. No
minimum time will be imposed on the single-click; any click less than the threshold
will be counted. From Table 4 we can see that if the maximum time for a single-click
were set at 199ms, as many as 45% of attempts at single-clicks would be in error
(longer than 199ms). Setting the limit at 399ms would reduce the errors to 6%, or
nearly one in 20. Whether a 6% error rate is appropriate, it is not possible to say
without further experiments. This is discussed further below, in Section 6.
Edwards How many ways can you use one button?
11
Time (ms) Frequency
Cumulative
frequency
0-49 0 0%
50-99 81 10%
100-149 79 20%
150-199 286 55%
200-249 168 76%
250-299 88 87%
300-349 44 92%
350-399 18 94%
400-449 12 96%
450-499 9 97%
500-549 8 98%
550-599 2 98%
> 600 14 100%
Total 809
Table 4. Single-click time distribution for ‘average’ participants (Groups
B to F). Mean = 164 ms, SD = 142. (This data is also represented in
Figure 7). The ‘Cumulative frequency’ is the percentage of click times
less than the specified time. For instance, 55% of the clicks were in a
time less than or equal to 199ms.
0
50
100
150
200
250
300
350
50 100 150 200 250 300 350 400 450 500 550 600
Time (ms)
Frequency
Figure 7. Distribution of times (milliseconds) for single-clicks for
average people (Groups B to F). Marked on the graph is the 350ms
point. If that were chosen as the limit for a single-click, then the
approximately 6% of clicks to the right of that line would be errors – not
counted as single-clicks.
Edwards How many ways can you use one button?
12
It is not surprising to note that the Morse Code operator’s ‘single-click’ time of
62.5ms is somewhat less than the mean in this experiment, given the level of training
of such an operator.
5.2 Long-clicks
A long-click is one which is held down for longer than a single-click. The results
above suggest that if a single-click is up to 600ms and anything longer than that
counted as a long-click, then normally no single-clicks will be mis-interpreted as
long-clicks. Of course the price to be paid is that the user must hold the button for a
relatively long time. That is to say that a long-click would be 436ms longer than an
average single-click – or around half-a-second. If one were to allow for 6% errors,
then a long-click could be defined as 400ms or longer, 236ms longer than the average
single-click.
This is much shorter than the mean long-click measured for these participants, which
was 1,432ms. That is a positive result inasmuch as it implies that users will not have
to waste as much time just waiting for their long-clicks to be recognized. However, as
discussed previously, the long-click experiment was unreliable and so too much
cannot be read into this.
The Morse Code operator’s ‘long-click’ (the dash) has a duration of 187.5ms, but that
would be much too fast, according to the results in Table 4. Being close to the mean
duration of a single-click it would imply too many false recognitions. Once again, this
is not a surprise, though.
5.3 Double-clicks
It is observed (see Figure 8) that within double-clicks d1 ˜ d3. The means (for Groups
B to F) are 129ms and 127ms respectively. (See Table 5 for further details). It is
interesting to note, however, that these times are rather shorter than the single-click
(mean = 164ms). This implies that one way of detecting a double-click would be to
measure the duration of a single-click. If it is less than some threshold (say 140ms)
then it could be assumed to be the first part of a double-click. This would not be very
practical, though. By definition there must be a second click involved and a user who
was capable of very rapid single-clicks would observe very confusing effects.
Edwards How many ways can you use one button?
13
0
50
100
150
200
250
300
350
400
40
80
120
160
200
240
280
320
360
400
440
480
520
560
600
d1
d3
Time (ms)
Frequency
Figure 8. Frequency of times (ms) for Groups B to F for the two button
depressions, d1 and d3 of a double-click (see Figure 3).
d1 d2 d3 D =
d1+ d2 + d3
Mean 129
150
127
406
Table 5. Timing data (ms) for the three components of the double-click.
Time (ms) Frequency Cumulative
frequency
0–39 0 0%
40–79 0 0%
80–119 121 15%
120–159 377 63%
160–199 89 74%
200–239 74 83%
240–279 72 92%
280–319 22 95%
320–359 0 95%
360–399 8 96%
= 400 6 100%
Table 6. Frequency distribution of d2 timings.
Other measurements that might be used to classify the double-click are d1 + d2 or
d1 + d2+ d3, but the simplest is d2. In other words if a single-click is detected then the
system must wait for a time and if another button-down occurs then a double-click is
recorded. The mean value for d2 was 150ms. If one was to adopt a 5% error rate Table
6), then the setting for d2 would be 319ms). This does imply that the minimum time
Edwards How many ways can you use one button?
14
delay for a single-click (i.e. the time to wait to ascertain whether it is the first of two
clicks of a double-click) will be718ms (= 399 + 319).
5.4 Triple-clicks
Anecdotal evidence suggests that triple-clicks are quite error-prone. With five
components, an error at any time may cause a mis-recognition. For instance, an
intended triple-click might be interpreted as a double-click followed by a single-click.
However, in the experiment the times were very uniform (as in Table 7). The overall
means of t1, t3 and t5 is 125ms, which is very similar to the button-down times, d1 and
d3 of the double-click (mean 128ms).
t1 t2 t3 t4 t5
Mean 127 153 117 160 125
Interquartile range 50 60 100 60 60
Table 7. Timing data (ms) for the five components of the triple-click.
A triple-click might be detected as a double-click, followed by a third button press
within time t4. Table 8 shows the distribution of t4times.
Time (ms) Frequency Cumulative
frequency
0–39 0 0%
40–79 0 0%
80–119 399 56%
120–159 239 89%
160–199 36 94%
= 200 43 100%
Table 8. Distribution of the times of the t4 component of the triple-click.
5.5 Novices
As discussed earlier, it was difficult to identify and separate novice participants.
However, the single-click data of all those participants in Groups B to F who had
declared that they had no computer experience was extracted. They showed a mean of
221ms. It comes as no surprise that this is slower and of wider variance than overall
group. It is also comparable with the performance of the discarded groups, confirming
the assertion that it is difficult to separate the effects of age and of computer
experience.
5.6 Errors
Participants made errors several times during the experiment. These took the form of
the person making the ‘wrong’ kind of click. This was most apparent when the
prompt requested a double-click but for some reason the person clicked just once.
They would then be waiting for a new prompt, and when one did not appear they
Edwards How many ways can you use one button?
15
would re-read the current one and realize their error. Where such errors were
observed, they were noted and the data subsequently discarded.
6. Discussion
The main objective of this experiment was to generate data that would be useful for
designers in setting time limits on button clicks and that has been achieved. Others
may want to use this data in a variety of ways. For our part, we make a number of
suggestions in the next section for further work that we would like to carry out on this
basis.
For example, further experiments will be required to assess what is the appropriate
acceptable error level. It is suggested above that this might be 5%. This would imply
that on average 1 click in 20 will be mis-classified, but that is taken over the whole
population. Some individuals will be able to achieve near 100% correct clicks and
such dextrous people might find the delays imposed unacceptable. For instance, if the
double-click limit (d1 + d2) is set at 829ms then someone who performs a single-click
in the average time of 164ms will have to wait for a delay of 665ms before the system
responds (having ascertained that the user has not made a double-click). Such a delay,
of over half a second, might make the system seem slow and might lead to frustration
– which in turn might cause more errors.
6.1 Comparison with the Model Human Processor
The data shows good agreement with the Model Human Processor of Card, Moran et
al., 1983. According to the model, the time required to perform a button-pressing
action is
Tp + Tc + Tm
Where Tp is the cycle time of the Perceptual Processor and Tc, Tm are the cycle times
of the Cognitive Processor and Motor Processor respectively. It is assumed that a
single-click requires one cycle of the Perceptual Processor to perceive the prompt, no
cognitive processing and a single cycle of the Motor Processor (i.e. the press and
release are not processed separately). Thus
c = Tp + Tm
Card et al. give the following values for an average user (so-called Middleman),
Tp = 100ms, Tm = 70ms. This gives a prediction of 170ms, which is very close to the
measured average of 164ms.
The double-click is harder to analyse. The time might be given by
D = 2Tp + Tc + 2Tm
if the following assumptions are made:
• two cycles of the Motor Processor are required;
• that the Cognitive Processor will be involved in counting the clicks and
Edwards How many ways can you use one button?
16
• two cycles of the perceptual processor are required, one to perceive the prompt
(as above) and one to perceive the registration of the first click.
On that basis,
D = 200 + 70 + 140 = 410ms
This is very close to the observed mean of 406ms. However, it would be unwise to
rely on this too much without further investigation of the validity of the above
assumptions.
7. Limitations
There are a number of limitations of the experiments which should be borne in mind
in using their results. One is that the task was an artificial one. In a real application the
user is not directly prompted by the interface as to what kind of click to make. Rather
the current context and user’s knowledge are drawn upon. For instance, in a word
processor, the user may know that a double-click will select a word rather than a
point, or in text entry to a telephone the user will be prompted by labels on the keypad
as to how many clicks are required to access the required letter.
There was also a limitation in the software used. For a reason that was never
uncovered, some short timings were not measured accurately and a time of 110ms
was returned when evidently the actual elapsed time was shorter. Therefore, some of
the data reported is slightly skewed and should be shorter.
8. Further work
The experiment that should immediately follow this is to make people perform the
different clicks within times implied by the results of this experiment. The first
decision is as to what level of errors is acceptable (e.g. 6%). Then, using the tables
above, it will be possible to set time limits for the different click types. First such an
experiment would ascertain whether the error rate prediction is correct. Then it would
be possible to find out how acceptable that is, subjectively, to users. Two levels of
acceptance could be gauged: error rate and system response.
Ergonomic factors have not been addressed in this study. That is to say that only one
design of button was used. It would be important to repeat this experiment with other
types of button. Simple parameters, such as the distance of travel of the button might
make a significant different. Indeed, buttons with zero travel (e.g. on a membrane
keyboard) might yield completely different data.
There is a great variance in the performance of users. This seems to depend partly on
experience. It seems appropriate therefore that systems might adapt to individuals: as
a user gets more dextrous, timing limits might become shorter, so making the
interaction more efficient. Experiments should be conducted to find how best to
implement such adaptation and to see whether it does indeed improve the interaction.
Another way of broadening the forms of input available might be to exploit people’s
innate musical rhythmic abilities. For instance, the rhythm of the opening bar of
Beethoven’s Fifth might be assigned a particular meaning. Experiments could be
carried out to ascertain the viability of such an approach.
Edwards How many ways can you use one button?
17
9. Conclusions
This study has generated data that can be used as the basis of the design of button-
based interactions. It provides the basis for further studies, some of which have been
suggested. Although most of the analysis has concentrated on the ‘average’
participant, it is clear that there is wide variance across the population and one
inevitable conclusion is that one size does not fit all. In other words, it is inappropriate
to impose the same time constraints on all users. So, where possible, click speeds
should be adjustable or should adapt automatically to individuals.
10. Acknowledgement
Many thanks to Peter Wright for his most valuable comments on a draft of this paper.
11. References
Barnard, P. and May, J. (1994). Interactions with Advanced Graphical Interfaces and
the Deployment of Latent Human Knowledge. (in) Interactive Systems:
Design, Specification and Verification. F. Paterno (Ed.) Heidelberg, Springer-
Verlag.
Barnard, P. J. (1985). Interacting Cognitive Subsystems: A psycholinguistic approach
to short term memory. (in) Progress in the Psychology of Language. A. Ellis
(Ed.) London, Lawrence Erlbaum Associates. 2: pp. 197-258.
Card, S. K., Moran, T. P. and Newell, A. (1983). The Psychology of Human-
Computer Interaction. Hillsdale, New Jersey:, Lawrence Erlbaum.
Edwards, A. D. N. (2001). Does size matter? (in preparation):,
Morsh, J. E. and Stannard, A. F. B. (1948). Studies in international Morse code VI:
Speed calculations for international Morse code. Canadian Journal of
Psychology 2: pp. 62-70,
