You are on page 1of 10

Displays 24 (2003) 111–120

www.elsevier.com/locate/displa

The effect of icon spacing and size on the speed of icon


processing in the human visual system
Tomas Lindberg*, Risto Näsänen
Brainwork Laboratory, Finnish Institute of Occupational Health, Topeliuksenkatu 41 aA, Helsinki FIN-00250, Finland
Accepted 12 September 2003

Abstract
Alphanumeric and graphical information needs to be presented in such a way that its perception is accurate, fast and as effortless as
possible. This study investigated the effects of spacing and size of individual interface elements on their perception. Experiment 1
investigated the effect of icon spacing on the speed of visual search for a target icon and determined the perceptual span for icons, that is, the
number of icons that can be processed by one eye fixation. Experiment 2 studied the effect of size, and experiment 3 the subjective
preferences for levels of icon spacing. The results of experiment 1 showed that spacing does not have an effect on search times. On average
the perceptual span for icons was found to be 25 arranged in a 5 £ 5 array. The size of the interface elements, on the other hand, was found to
have a great effect. Icons smaller than 0.78 resulted in significantly raised search times. Experiment 3 revealed that an inter-element spacing
of one icon is to be preferred and a spacing of zero icons is to be avoided.
q 2003 Elsevier B.V. All rights reserved.
Keywords: Visual search; Perceptual span; Stimulus density; Stimulus size; Icons; Visual acuity; Eye movements; Threshold search time

1. Introduction information. Visual search, using the visual search para-


digm [5], is an efficient way of measuring the effects of
When working with computers and other devices with different presentation factors on the speed by which humans
complex user interfaces people receive large amounts of can process large amounts of complex visual information.
alphanumeric and graphical information that they need to Consequently, several studies have applied the visual search
interpret. This information should be presented in such a paradigm in the context of user interfaces [6 – 13].
way that its perception is accurate, fast and as effortless as
possible. As with alphanumeric information the efficiency 1.1. Eye movements in visual search
with which graphical information is processed should
depend on several different factors. For alphanumeric In most cases, visual search involves eye movements.
information these factors have been shown to be, for The reason for this is the fact that the acuity of the human
example, size and contrast [1 – 4]. visual system is highest at the fovea and decreases towards
The processing of graphical information involves locat- the periphery [14]. This is due to a similar decrease of the
ing the correct piece of information, recognising the ganglion cell density on the retina, which is also highest at
physical object and understanding what it means in the the fovea [15]. As a result of this, in order to discern
current context of use. Locating a specific piece of information located outside of the fovea, the eyes need to
information, that is, visually searching for it is most difficult move so that this area is brought into the fovea or close
within a new environment and becomes easier as the enough to it for identification.
environment becomes more familiar. Nevertheless, even a The eye movements consist of fixations and saccades.
familiar environment requires some amount of visual search Fixations are periods when the eyes do not move (in reality
to pin down the exact location of the sought after there exists tremor, drift and micro saccades during a
fixation [16]). Information is only picked up during fixations.
* Corresponding author. The average duration of a fixation is about 225– 275 ms for
E-mail address: tomas.lindberg@ttl.fi (T. Lindberg). reading and 275 ms for visual search [17] although
0141-9382/$ - see front matter q 2003 Elsevier B.V. All rights reserved.
doi:10.1016/S0141-9382(03)00035-0
112 T. Lindberg, R. Näsänen / Displays 24 (2003) 111–120

considerably shorter fixations (approaching 150 ms) have It is not known, however, whether the perceptual span
also been reported [18,19]. Between every fixation there is a grows and shrinks with changes in element spacing and to
saccade, that is, a rapid movement of the eyes from one what extent attention, peripheral visual acuity and crowding
point of fixation to the next. Compared to fixations, limit the perceptual span.
the duration of saccades is much shorter. A saccade of 58,
for example, lasts around 40 ms [16]. The velocities of 1.4. Purpose of the study
saccades are very high and can reach values exceeding 6008
per second [16]. This is one reason why little or no The purpose of the study was to investigate the effects of
information is picked up during saccades. Even eye the density and size of user interface elements, more
movements that appear smooth and continuous to the specifically computer icons, on the speed with which users
observer do in fact consist of fixations and saccades. find what they are looking for using the visual search
However, smooth eye movements (pursuits) occur when paradigm. The study also defines the size of the perceptual
moving objects are viewed. The brain integrates the images span for icons and gives guidelines to designers regarding
at different fixations so that the observer sees a continuous the density of elements and their recommended size.
image. For a comprehensive review on eye movements
see Ref. [17].
Studies of eye movements in visual search have shown 2. Methods
that the observer cannot directly control the fixation
duration. Instead, it has been found to be under the indirect Experiment 1 gave objective measures on the effect of
control of a mechanism that uses estimations of foveal stimulus density and set size on the speed with which
analysis times of previous fixations [20,21]. Because the subjects found the target icon among distractors. Exper-
fixation duration cannot be directly controlled, it follows iment 2, on the other hand, reported on the effects of
that the search time is directly linked to the number of stimulus size. These results can, as will be discussed in the
fixations. Thus, information systems, where information sections below, be used to design effective graphical user
needs to be located quickly, should be designed to minimise interfaces in which the users find the sought after objects
the number of fixations. quickly and efficiently. There is, however, more to the
usability of a user interface than merely the efficiency and
1.2. Perceptual span effectiveness with which it is used. The subjective
preferences and aesthetic impressions of the users also
Perceptual span is the area around the point of fixation play a very big role in the overall usability. Considering
from which useful information can be gathered [17]. interfaces designed for consumers, the subjective
Knowing the size of the perceptual span for a specific impression and ‘feeling’ the interface gives the user, can
type of stimulus can help the designers minimise the number be even more important than overall efficiency. Therefore,
of fixations and thus speed up the visual search. A study by experiment 3 aims to find out which level of density is rated
Näsänen and Ojanpää [18], for example, found the highest.
perceptual span to encompass a total of four faces arranged
in a 2 £ 2 array. Whenever arrays of faces need to be 2.1. Subjects
designed and the speed of visual search is of importance,
designers now know how to optimise the size of the array. Three subjects participated in experiments 1 and 2. Of
The actual shape of the visible area during a single fixation these, one had university level education and two were
was found by Ojanpää, Näsänen and Kojo [19] to be roughly university students with sixth form college education. All
elliptical and larger in the horizontal direction. three subjects were of the same age group with two being 25
years old and one 26 and they all had normal or corrected to
1.3. Stimulus density normal vision. At the start of the experiments one of the
three subjects was naı̈ve, one slightly practised and one
Vlaskamp, Hooge and Over [22] measured search considerably practised. Since experiment 2 took place after
performance, defined as search time divided by number of experiment 1, all the subjects were practised when
elements as subjects searched for an ‘O’ among ‘C’s. The performing it.
results of the study indicate that inter-element distances A total of fourteen subjects participated in experiment 3,
larger than 1.58 do not have an impact on search six men and eight women. Of the men three were in the age
performance, whereas for distances smaller than this, search range 20 –25, one 31 –35, one 41 –45 and one 46– 50. Four
performance ‘decreased dramatically’. As an explanation of the six male subjects had university level education and
for the decrease of performance for dense arrays Vlaskamp two had sixth form college levels with one of these two also
et al. [22] offered crowding, that is, that humans are better at reporting vocational school levels of education. Of the
identifying letters presented alone than when surrounded by women taking part in the experiment one was in the age
other letters in close proximity [23]. range 20 –25, two 31 – 35, three 41– 45, one 46– 50 and
T. Lindberg, R. Näsänen / Displays 24 (2003) 111–120 113

finally one 51– 55. The educational background of the


women varied somewhat more than that of the men. Five
had a university education, one had a vocational high school
education, one had graduated from an educational institute
of healthcare and one had comprehensive school education.
All of the subjects had normal, or corrected to normal vision
and were naı̈ve as to the purpose of the study.

2.2. Stimuli

The stimuli in experiment 1 consisted of 25 different icon


matrix configurations of five different sizes (2 £ 2, 3 £ 3, Fig. 2. Examples of masks shown before and after stimuli.
5 £ 5, 7 £ 7 and 10 £ 10) and five different density-levels
(inter-element distances being zero icons (0 pixels), 1/4 icon
See Fig. 2 for example of masks. Like the stimuli,
(8 pixels), 1/2 icon (16 pixels), one icon (32 pixels) and two
the masks were also saved in bitmap format with a colour
icons (64 pixels)). The total size of the matrices thus ranged
depth of 24 bits.
from 64 £ 64 pixels for a 2 £ 2 matrix with inter-element
The stimuli were presented using an SGI 1600SW LCD
distance of zero icons to 896 £ 896 pixels for a 10 £ 10
monitor with a #9 Revolution IV digital graphics adapter
matrix with inter-element distance of two icons. For each
connected to a PC computer with a 450 MHz Pentium III
combination of size and density 100 different matrices
processor running Windows2000. The pixel sizes of the
containing the target and 100 matrices not containing the
monitor were 0.0231 £ 0.0231 cm 2 corresponding to
target were generated. The matrices were made up of 219
0.02318 of the visual angle with an observation distance of
different colour icons in total, 218 distractors and one target.
57 cm used in the experiment.
All icons used were found in the Windows2000 operating
The stimuli of experiment 3 were printed on an HP C
system and were 32 £ 32 pixels in size. The stimuli were
LaserJet 4500 PS printer with a resolution of 600 dpi on
saved in bitmap format with a colour depth of 24 bits. The
Canon High Grade 100 g/m2 coated paper. The paper
background colour used in the matrix was grey representing
presentation was done in order to enable the subjects to
the RGB values of 185 for each component, corresponding
easily arrange the stimuli in the order they preferred and to
to half the maximum luminance of the display. Fig. 1 shows
view them at the same time spread out on a table. The
examples of icon matrices used in the experiments.
stimuli selected for the experiment were of all five different
The stimuli used in experiment 2 were the 10 £ 10 icon
density levels arranged in 5 £ 5 matrices.
matrices with an inter-element spacing of one icon from
experiment 1. The reason for selecting the 10 £ 10 matrix
2.3. Procedure
size was that it is large enough to allow eye movements. The
icon spacing of one icon was selected because it had the best
In experiment 1, we varied the set size and inter-element
subjective impression as reported in the results of
spacing. Of the three subjects one performed the visual
experiment 3 below.
search on all combinations of set size and density. The other
Mask stimuli were shown before and after the stimulus
two subjects used all densities but only set sizes 3 £ 3,
matrices to mask any possible residual images on the retina
5 £ 5, 7 £ 7 and 10 £ 10. The reason for this being the
that could affect the search results. The masks used
extremely low threshold measured by subject TL for set size
were pixelated noise, with a check size of 10 £ 10 pixels.
2 £ 2.
Experiment 2 varied the viewing distance, corresponding
to changes in icon size. The distances used were 28, 57, 114
and 228 cm. The distances corresponded to icon sizes in
degrees of visual angle of 1.48, 0.74, 0.37 and 0.18,
respectively.
In experiments 1 and 2, the subjects were asked to locate
a target icon from a set of distractors. The target was either
present or absent, both with a probability of 0.5. The
subjects were asked to indicate if he/she found the target or
not by clicking one of two buttons located to the left of the
stimuli. If the target was present and the subject found it,
Fig. 1. Examples of icon matrices used in the experiments. The figures from
he/she clicked a button with an image of the target, and if
left to right have the following set size and inter-element spacing, 3 £ 3 and not present or not found a button with the label ‘No’. A
0 icon width, 2 £ 2 and 1 icon width, 5 £ 5 and 1/2 icon width. wrong selection was indicated with a sound. See Fig. 3 for
114 T. Lindberg, R. Näsänen / Displays 24 (2003) 111–120

best, that is, which they would prefer to use in an actual user
interface. The instructions were printed on paper to ensure
consistency of instruction across subjects. The subjects were
asked for their initial impressions, that is, they arranged the
density levels into their preferred order without having seen
the stimuli beforehand. The illumination was adjusted so
that the luminance corresponded to that of the stimulus
presented on the LCD screen used in experiment 1
(approximately 62 cd/m2). After the stimuli were arranged
into the preferred order, the subjects were asked to motivate
why they had placed a certain density level last and another
one first.

2.4. Eye movement recordings

Eye movements were recorded in parallel with the


Fig. 3. Example of the search display with a 10 £ 10 matrix with inter-
element distance zero icons.
threshold measurements for experiment 1. No eye move-
ments were recorded for experiments 2 and 3. The
an example of the search display. Before and after stimulus equipment used was an SMI (SensoMotoric Instruments
presentation a mask, as described in the section above, was Inc.) Eye Link video eye tracker. The equipment enables the
shown. The delay between clicking a selection button and measurement of both eyes by illuminating them with
showing the next stimulus was 500 ms. The subjects were infrared LEDs located in miniature infrared cameras
not given a fixation point from which to start the search. positioned one below each eye. The system has a sampling
Instead, they were free to choose their own pattern of eye rate of 250 Hz and is controlled by a separate PC running
movements while searching for the target. The subjects MS-DOS. It is connected to the computer presenting the
were, however, instructed to immediately move their eyes to stimuli by an Ethernet cable.
the buttons on the left-hand side of the screen when the Software bundled with the eye tracking system auto-
target was found, then move the mouse cursor to the correct matically analyses the eye movements recognising saccades
button and only after having moved their eyes back to the and fixations. A sample belongs to a saccade if one of two
mask click the button. To ensure that the subject stayed as criteria is met: acceleration needs to exceed 95008/s2 or
still as possible and that the observation distance remained velocity needs to exceed 358/s. If neither of these two
constant at 57 cm a chin rest was used. criteria is met, the software interprets the sample as
Following the procedure described above, the threshold belonging to a fixation. Because measurements were made
search time, that is, the time in which the subject finds the for both eyes, the results represent the mean of the results for
target with a certain level of probability was measured. For the left and the right eye. As with the thresholds, eye data
every combination of density and set size in experiment 1 was only recorded after the subjects had made two errors.
and for element sizes in experiment 2 four different The eye movement recordings and the presentation of
thresholds were measured and the final threshold was stimulus were connected so that the recording of eye data
obtained as the mean value of these. After the subject had began when the stimulus was presented to the subjects. The
answered correctly on three stimuli in succession the recordings were stopped either when the subjects clicked a
presentation time of the stimuli was reduced by a factor of button, moved their eyes away from the matrix to the
1.26 and in the case of an incorrect answer the presentation buttons or when the stimulus presentation was ended and
time was increased by the same factor. The starting replaced with a mask. This was done in order to minimise
presentation time was 8000 ms. After eight reversals, that recording eye movements after the subject had found the
is, after eight changes from decreasing the presentation time target, which would otherwise affect the results.
to increasing it or vice versa, an estimate of the threshold at
a probability level of 0.79 of correct responses was found
[24]. Counting reversals was not started until the subjects 3. Results
had made two errors.
In experiment 3 the subjective impressions on the density 3.1. Experiment 1
of elements were examined with high quality paper print
outs of different density levels in a single matrix size. Only Fig. 4 shows the combined threshold search times as a
subjects with normal or corrected to normal vision were function of inter-element spacing for all subjects. Examin-
accepted as subjects. The subjects were asked to arrange the ing the threshold search times as a function of inter-element
pictures, presented initially in a randomly arranged pile, spacing showed no significant effects when combined over
according to their subjective opinion of which looked or felt subjects. More specifically, a Friedman’s rank test for k
T. Lindberg, R. Näsänen / Displays 24 (2003) 111–120 115

Fig. 4. The combined threshold search times of all subjects as a function of


icon spacing. The error bars indicate the standard error of the mean.

correlated samples [25] showed no significant effects for


set sizes 10 £ 10 (xF ð4Þ ¼ 9:13; p ¼ 0:058; ns), 7 £ 7
( xF ð4Þ ¼ 3; p ¼ 0:558; ns) and 5 £ 5( xF ð4Þ ¼ 4:60;
p . 0:331; ns). The test did, however, show an effect for
the 3 £ 3 set size (xF ð4Þ ¼ 13:52; p , 0:05). Even though
the threshold search time was shown to depend on the inter-
element spacing for set size 3 £ 3, the actual value of the
threshold varies very little and the number of fixations at the
set size, as will be shown below, is one for all subjects
independent of inter-element spacing. Due to these reasons
the dependence shown for that set size has no significance in
practice. There existed, on the other hand, individual
differences between subjects, see Fig. 5. The results for
subject KL showed a systematic tendency of increase at low
and high values of icon spacing. The results for the other
two subjects did not show similar systematic effects.
Fig. 6 shows the threshold search times of the subjects as
a function of set size. The threshold search times can be seen
to increase with increasing set size. There were, however,
again differences between subjects with subject TL having
clearly lower thresholds than subjects KL and KH. The
Fig. 5. The threshold search times as a function of icon spacing. The error
experiment showed, in other words, a clear effect of set size bars indicate the standard error of the mean.
on the threshold search times for all densities.
Fig. 7 below shows the fixation durations of the subjects Subject KH required one fixation for the 3 £ 3 set size and
for set size 10 £ 10. The fixation duration for the largest set only about 1.5 fixations for set size 5 £ 5. Subject TL, on the
size varied considerably between subjects. Subject TL had other hand, found the target with only one fixation in set
the shortest fixations whereas subject KL had the longest, sizes 2 £ 2, 3 £ 3 and 5 £ 5 independent of inter-element
even over twice as long as subject TL. Subject TL showed spacing. Larger set sizes, however, required more fixations
an increase in fixation duration for the inter-element and a difference between different levels of inter-element
distance of two icons compared to the other inter-element spacing at a specific set size also became evident. Finally,
distances, which was in contrast to subjects KH and KL who subject KL also found the target with one fixation at set sizes
showed an increase for the inter-element distance of zero 3 £ 3 and 5 £ 5. In addition to this, subject KL also required
icons. Subject KH also showed an increase for the icon only one fixation at set size 7 £ 7 for some densities. In
spacing 1=4 icon. other words, there existed differences between subjects as
Fig. 8 shows the numbers of fixations for specific inter- expected, but the set size that could be seen with one
element distances as a function of set size. Whereas the fixation seems to be 5 £ 5 on average.
threshold search time was already significantly elevated at a Finally, Fig. 9 shows the number of fixations as a
set size of 5 £ 5, compared to smaller set sizes, the numbers function of degrees of visual angle. The fact that set sizes of
of fixations showed no such results for two out of the three 5 £ 5 (or even more under some circumstances) could be
subjects. The one subject, KH, that showed an elevated seen with a single fixation independent of element spacing
number of fixations for set sizes 5 £ 5 and larger, showed, means that the actual area, in degrees of visual angle, that
however, a very moderate elevation for the 5 £ 5 set size. was seen with a single fixation increased as the set size and
116 T. Lindberg, R. Näsänen / Displays 24 (2003) 111–120

Fig. 6. The threshold search times as a function of set size. The error bars
indicate the standard error of the mean.
Fig. 8. The number of fixations as a function of set size. No systematic
effects at the largest set size exist. The error bars indicate the standard error
inter-element spacing increased. The 5 £ 5 set size with an of the mean.
inter-element distance of zero icons for example occupied
an area of about 3.5 £ 3.58 squared whereas a set size of about 9.5 £ 9.58 squared. Both were still seen with one (or
5 £ 5 with inter-element spacing of two icons occupied close to one) fixation. On the other hand, the 10 £ 10 matrix
with an inter-element distance of zero icons occupied a
somewhat smaller area (7.4 £ 7.48 squared) than the 5 £ 5
matrix with the inter-element spacing of two icons
(9.5 £ 9.58 squared) but the number of fixations was
significantly higher. The results thus showed that the
number of fixations seemed to depend on the number of
elements, not the area they covered.

3.2. Experiment 2

As can be seen from Fig. 10, the threshold search times


increased sharply as the icon size decreased from about 0.48
Fig. 7. Fixation durations as a function of icon spacing for set size 10 £ 10. to about 0.28. Also the increase in the threshold search time
The error bars indicate the standard error of the mean. from an icon size of around 0.78 to about 0.48 was clear
T. Lindberg, R. Näsänen / Displays 24 (2003) 111–120 117

although not nearly as drastic. The change in threshold


search time from icon size 1.5– 0.78 was very small. In other
words, as the icon size decreased from about 0.78, the
threshold search time increased for all subjects. As in
experiment 1, small differences existed between subjects.

3.3. Experiment 3

Table 1 shows the rankings of the density levels for the


different subjects. A Friedman’s rank test for k correlated
samples [25] performed on the ranking of the density levels,
see Eq. 1, revealed that the distribution of rankings was not
random but depended on the level of density. The equation
for the Friedman’s rank test can be written as [25]:

12 Xk
x 2F ¼ R2 2 3Nðk þ 1Þ ð1Þ
Nkðk þ 1Þ i¼1 i

where N is the number of subjects, k the number of density


levels and Ri the sum of the ranks for the different density
levels. Entering the values into Eq. 1 gave x2F ¼ 35:6 on 4
df. Comparing this to x20:001 ð4Þ ¼ 18:467 H0 ; the null
hypothesis that the rankings are randomly distributed,
could be rejected and as mentioned above the rankings
depend statistically highly significantly on the density level.
From the ranking sums it can be seen that the spacing of
one icon width was the most preferred while the spacing of
zero icons was clearly the least preferred. Following the
spacing of one icon width, the next best density level,
measured by the sum of the rankings, was the spacing of 1/2
icon width. The last two-icon spacing levels, 1/4 icon width
and two icon widths were quite closely matched. The
spacing level of two icon widths was judged to be only
marginally better than the spacing level of 1/4 icon width.

Table 1
The ranking of the different icon spacing levels by subjects
Fig. 9. The number of fixations as a function of degrees of visual angle. The
error bars indicate the standard error of the mean. Subject Icon spacing

0 Icons 1/4 Icon 1/2 Icon 1 Icon 2 Icons


(8 pixels) (16 pixels) (32 pixels) (64 pixels)

Subject 1 5 4 2 1 3
Subject 2 5 4 3 1 2
Subject 3 5 4 3 1 2
Subject 4 5 4 2 1 3
Subject 5 5 3 2 1 4
Subject 6 5 4 3 1 2
Subject 7 5 4 2 1 3
Subject 8 5 4 3 1 2
Subject 9 5 3 2 1 4
Subject 10 5 4 3 1 2
Subject 11 5 1 2 3 4
Subject 12 5 4 2 3 1
Subject 13 5 1 2 3 4
Subject 14 4 2 1 3 5
Fig. 10. The effect of icon size on the threshold search time. The error bars Sum 69 46 32 22 41
indicate the standard error of the mean.
118 T. Lindberg, R. Näsänen / Displays 24 (2003) 111–120

According to the comments of the subjects the indicating that different search strategies may be equally
motivations for selecting the icon spacing of one icon as efficient.
the most preferred were that ‘all the information can be seen Finally experiment 1 showed that the threshold search
at a glance’ and that ‘the icons were adequately spaced times increase as the set size increases. This is in line with
resulting in it being easy to find targets’. ‘They were easily the theory by Treisman and Gelade [5], which states that
distinguishable’. One comment about the density level was search for a conjunction of features is serial. The target icon
that ‘it was manageable and easily perceived’. The image could not be detected based on any one single feature, in
was in one word, clear. The same comments applied even to other words it did not pop-out but required instead the
the other icon spacing levels that were reported to be most deployment of attention before it could be found from
preferred with the additional comment on the 1/4 spacing of among the distractors.
it being calm, not uneasy or nervous. It would seem that In experiment 2, the reason for the decreased search
these are the attributes of a good layout. Subjects just found performance as the size of the icons decreased might be the
them in slightly different levels of density. low-pass filtering performed by the human visual system.
Because the density levels preferred the least were the As a result of this any high-frequency information (fine
ones from both extremes, the comments for why they were detail) in the icons is filtered away and the discrimination of
not liked did not match. The subject who found a spacing of the icons has to be made based on low-frequency
two icons to be the worst said the reason to be that ‘they information (areas with low detail). As a result of this the
were spaced too far apart, the visual field became too large’. icons used in the experiment become more difficult to tell
The subjects who reported the zero icon spacing level to be apart, resulting in longer search times.
the least preferred to a large extent commented on it being From this it follows that small icons should be designed
‘too small and crowded’. ‘The individual icons became hard separately from large ones. Whereas large icons may utilise
to find’, ‘they were getting mixed together’ and ‘it was more high spatial frequency information relative to icon size
difficult to tell where one ended and the next began’. One
(more fine detail) small icons should be designed so that
subject said it was ‘oppressive’ and one said she ‘got
they are easy to recognise based on low spatial frequency
irritated looking at it’ and that it was ‘uneasy’.
information (low detail).
The experiment only used icons as large as 1.488 and can
3.4. Discussion
thus not draw conclusions about the possible effect on
search time of even larger elements. However, an icon size
The results showed that the effect of inter-element
of 1.488 is already twice as large as the normal icon size on a
spacing on search times is negligible at most. This would
Windows desktop viewed at normal distance. Taking this
thus indicate that crowding [23] does not play a part when
into consideration, the use of even larger elements in
icons are concerned. It is, however, likely that considerably
applications designed for the average user is unlikely.
larger inter-element distances than used here would have a
negative effect on search times because of the decrease of Applications designed for devices with small displays are,
visual acuity with eccentricity. on the contrary, more likely to use even smaller icons in
The important finding regarding the perceptual span for order to save space.
icons is that it seems not to be limited by target eccentricity The implications of the results obtained in the exper-
at least for the range of inter-element distances used here. iment are that the size of the elements used in interfaces is
About 25 elements arranged in a 5 £ 5 array could be important and can have a highly significant impact on
processed during one fixation independent of inter-element performance. This fact should be taken into consideration in
distance. Larger arrays with more elements could cover a the design of applications. Likewise, the spatial frequencies
smaller area, in degrees of visual angle, than a 5 £ 5 array of the icons do have a difference. Small icons should use low
but still required more than one fixation. spatial frequencies relative to icon size, or in other words,
The results of experiment 1 are in agreement with the small icons should be simple.
finding by Motter and Belky [26] that the perceptual span From the results of experiment 3, it is very clear that the
could have a constant diameter when expressed indepen- inter-element distance of zero icons is not suitable for use in
dently of stimulus density. The study by Motter and Belky applications if people’s subjective opinions are taken into
[26] was performed on rhesus monkeys and is thus not account. Put simply, people do not seem to like it when the
necessarily directly compatible with results obtained by interface is too crowded and complex. The display becomes
human subjects. nervous and even oppressive to some.
Another interesting finding that the eye movement The inter-element distance that is preferred by the
recordings reveal is a difference in search strategies between subjects is one icon. If for some reason the one icon spacing
the subjects. Subject TL, for example, used many short is not usable the results reported above suggest that the next
fixations whereas subject KL used few long fixations. The best icon spacing is 1/2 icon. These alternatives seem to be
differences in performance for the 10 £ 10 set size for best balanced between a too crowded display with too much
subjects KL and TL were not very different, however, information and a sparse one where the interface elements
T. Lindberg, R. Näsänen / Displays 24 (2003) 111–120 119

are too far apart and the interface contains too little 4. Conclusions
information.
One thing that must be taken into account, however, is Based on the results of the experiments guidelines can
that as mentioned in the methods section the subjects were now be given to interface designers on the density and size
asked for their initial impressions of the density levels. of icons (interface elements) used in graphical user
Given this, the results might be significantly different if the interfaces:
subjects had been using actual applications with the
different density levels for some time before stating their † The preferred spacing to use between icons (interface
opinion. elements) is 1=2 2 1 icon (interface element) width.
Independent of what inter-element spacing the user † Icons (interface elements) should not be placed next to
prefers, it does not have a negative effect on efficiency of each other without any spacing between them.
use measured in threshold search time. Ideally then, users † The minimum size of icons (interface elements) should
should be able to modify the inter-element spacing to not be less than 0.78 of visual angle, which corresponds
their own liking. to about 0.5 cm at a viewing distance of 40 cm, and about
What is the real-world validity of findings based on a 0.9 cm at a viewing distance of 70 cm. If smaller icons
visual search task? An unfamiliar icon in an unfamiliar (interface elements) need to be used they should be
surrounding has to be searched from a large area, which designed carefully using mostly low spatial frequency
corresponds to large set sizes in a visual search information (low level of detail).
experiment. However, the approximate position of a † The perceptual span for icons is 5 £ 5, that is, they can be
familiar icon in a familiar surrounding is known to the seen with one fixation. This knowledge can be used to
user and, therefore, has a small search area, which minimise the number of fixations for an interface.
corresponds to small set sizes in the visual search
experiments. One of the basic advantages of a graphical
user interface is that it reduces the memory load of a
user. Users do not have to remember the exact location References
of the vast number of elements in user interfaces she/he
[1] G.E. Legge, D.G. Pelli, G.S. Rubin, M.M. Schleske, Psychophysics of
is using more or less regularly. Therefore, we believe
reading—I. Normal vision, Vision Res. 25 (1985) 239–252.
that locating information nearly always requires at least [2] G.E. Legge, G.S. Rubin, A. Luebker, Psychophysics of reading—V.
some amount of visual search. The role of contrast in normal vision, Vision Res. 27 (1987)
The icons used in the experiments were set up in 1165– 1177.
symmetrical arrays and they all had an arrow in the lower [3] R. Näsänen, J. Karlsson, H. Ojanpää, Display quality and the speed of
visual letter search, Displays 22 (2001) 107 –113.
left corner (see Fig. 1 for examples). The set-up can be
[4] R. Näsänen, H. Ojanpää, I. Kojo, Effect of stimulus contrast on
considered to represent a fairly typical situation. Oftentimes performance and eye movements in visual search, Vision Res. 41
the icons on the desktop, the mobile phone or PDA are set (2001) 1817– 1824.
up in this way by default. To determine whether the [5] A.M. Treisman, G. Gelade, A feature-integration theory of attention,
existence of the arrows on the icons had an impact on the Cogn. Psychol. 12 (1980) 97 –136.
results, a control experiment was performed with the same [6] I. Benbasat, P. Todd, An experimental investigation of interface
design alternatives: icon vs. text and direct manipulation vs. menus,
icons without the arrows. The control experiment showed Int. J. Man–Machine Stud. 38 (1993) 369–402.
that the arrows did not have any effect. The actual icons [7] W.L. Bewley, T.L. Roberts, D. Schroit, W.L. Verplank. Human
themselves were real icons found on a computer running the factors testing in the design of Xerox’s 8010 ‘Star’ office workstation.
Windows2000 operating system. Proceedings of 1983 Conference on Human Factors in Computing
Even though the study used icons as stimuli, we believe Systems, 1983, pp. 72 –77.
[8] M.D. Byrne. Using icons to find documents: simplicity is critical.
that the results are applicable to other forms of graphical Proceedings of the Conference on Human Factors in Computer
interface elements as well. The guidelines listed in the Systems, 1993, pp. 446 –453.
section below should thus, be used every time a new [9] J. Moyes. When users do and don’t rely on icon shape. Proceedings of
interface is designed or an existing one is modified. The the CHI ‘94 Conference Companion on Human Factors in Computing
guidelines can be seen as a complement to the ISO 9241-12 Systems, 1994, pp. 283 –284.
[10] M. Niemelä, J. Saarinen, Visual search for grouped versus ungrouped
standard [27] for information presentation and they state
icons in a computer interface, Hum. Factors 42 (2000) 630– 635.
how the density and size of interface elements is to be taken [11] L. Repokari, T. Saarela, I. Kurkela. Visual search on a mobile phone
into consideration when designing graphical user interfaces. display. In SAICSIT 2002 Enablement Through Technology Annual
Following the guidelines should result in applications that Conference of the South African Institute of Computer Scientists and
are more efficient to use while retaining a high level of user Information Technologists, South Africa, 2002.
[12] T. Tanskanen, Visuaalinen haku graafisessa käyttöliittymässä,
satisfaction. The results obtained in the study are also
Progradu, University of Helsinki, Helsinki, 1999.
applicable to all areas of interface design ranging from [13] S. Wiedenbeck, The use of icons and labels in an end user application
applications controlling complex industrial processes to program: an empirical study of learning and retention, Behav. Inf.
mobile applications used occasionally by novices. Technol. 18 (1999) 68–82.
120 T. Lindberg, R. Näsänen / Displays 24 (2003) 111–120

[14] T.H. Wertheim, Peripheral visual acuity, Am. J. Optom. Physiol. Opt. [21] I.T. Hooge, C.J. Erkelens, Adjustment of fixation duration in visual
57 (1980) 915 –924. search, Vision Res. 38 (1998) 1295–1302.
[15] C.A. Curcio, K.A. Allen, Topography of ganglion cells in human [22] B. Vlaskamp, I. Hooge, E. Over, Stimulus density does not affect
retina, J. Comp. Neurol. 300 (1990) 5–25. saccadic search performance, Perception (2002) 31.
[16] K.J. Ciuffreda, B. Tannen, Eye Movement Basics for the Clinician, [23] H. Bouma, Interaction effects in parafoveal letter recognition, Nature
Mosby, St Louis, MO, USA, 1995. 226 (1970) 177–178.
[17] K. Rayner, Eye movements in reading and information processing: 20 [24] G.B. Wetherill, H. Levitt, Sequential estimation of points on
years of research, Psychol. Bull. 124 (1998) 372–422. a psychometric function, Br. J. Math. Stat. Psychol. 18 (1965)
[18] R. Näsänen, H. Ojanpää. How many faces can be processed during a 1–10.
single eye fixation? In 11th European Conference on Eye Movements, [25] D.C. Howell, Statistical Methods for Psychology, Fifth ed., Duxbury,
Turku, Finland, 2002. Pacific Grove, USA, 2002.
[19] H. Ojanpää, R. Näsänen, I. Kojo, Eye movements in the visual search [26] B.C. Motter, E.J. Belky, The zone of focal attention during active
of word lists, Vision Res. 42 (2002) 1499–1512. visual search, Vision Res. 38 (1998) 1007–1022.
[20] I.T. Hooge, C.J. Erkelens, Control of fixation duration in a simple [27] ISO 9241-12, Ergonomic requirements for office work with visual
search task, Percept. Psychophy. 58 (1996) 969 –976. display terminals (VDTs)—Part 12: Presentation of information.

You might also like