The Eye-tracking Study of the Line Charts in Dashboards Design

Pavel A. Orlov

1,2

, Tatiana Ermolova

, Vladimir Laptev

, Alexey Mitrofanov

and Vladimir Ivanov

Department of Engineering Graphics and Design, Peter the Great St. Petersburg Polytechnic University,

Polytechnicheskaya 29, St. Petersburg, 195251, Russia

School of Computing, University of Eastern Finland, PO Box 111, FI-80101, Joensuu, Finland

Keywords:

Eye Tracking, Eye Movement, Dashboards, Information Graphics, Line Graphs.

Abstract:

Dashboards are an important ﬁeld for an investigation as they are the visual part of the management informa-

tion systems. Our study aimed to ﬁnd out the effects of the changes in the types and numbers of line graphs

that can be displayed on a single screen simultaneously. Two laboratory experiments were conducted using an

eye-tracker to ﬁnd out how subjects’ perception of line graphs on dashboards changes with increase in graph

numbers, changes in sizes and increase of the overall area taken up by the graphs on the screen. We show that

if the graphs take up the same area, the subjects perceive the line graphs displayed simultaneously in a similar

manner. If the subjects are shown an increasing number of graphs of the same size, the subjects take longer to

respond to tasks and have higher ﬁxation count per each stimuli. The study revealed that there is no correlation

between graph’s slope (trend) and subjects’ perception.

1 INTRODUCTION

Dashboards (sometimes referred to as instrument pan-

els, control panels) are a tool for data visualisation.

Dashboards are a vital part of everyday life, for exam-

ple for NASDAQ traders (American/Canadian stock

exchange traders). One faces a considerable amount

of information both during work and during leisure.

For working with large amounts of data, especially if

managing companies, it is vital to have all the data,

trends and correlations ’at the ﬁnger tips’. Visualis-

ing all the required data on a single computer screen

poses a challenge for both programmers and design-

ers.

Providing effective data visualisation is critical to

dashboard design. There are many ways of visualis-

ing the same information. There are, however, rela-

tively few studies on how the number of graphs dis-

played at the same time affects the person working

with the dashboard. In our study, we will test the ef-

fects of the amount of graphs displayed on the screen

simultaneously on the subjects’ perception of the in-

formation. Because the same number of graphs can be

displayed on the screen differently, we will also test

the effect of the graphs’ size on the subjects’ percep-

tion. This research is important for dashboard design

as it presents the basics of any design, which is often

neglected during dashboard development. Instead of

using scientiﬁcally proven methods for designing the

dashboards, designers sometimes opt for trial and er-

ror method, which is a tedious process that may not

yield any results.

With this in mind, it is difﬁcult to address all the

challenges that are presented before the dashboard de-

signers. They can be, however, addressed separately.

This process of deconstructing dashboards into differ-

ent design aspects and starting from the more simple

ones, would give more solid grounds for further, more

complicated work. This approach would also limit the

amount of variables that are being studied at one time.

In our case, line graphs were studied. After all, they

are the most frequently used basic way to visualise

a considerable amount of comparable information on

websites, user interfaces and dashboards.

This paper aims to start the process of system-

atic analysis of different aspects of dashboards design.

This will give a clearer picture about the perception

of information (which contributes to the scientiﬁc do-

main of information perception studies) and help to

develop recommendations on better ways of informa-

tion visualisation not only on dashboards, but also on

websites and on other types of user interfaces.

The study was carried out on the effect of the

graph sizes and graph count on immediate reactions

and answers of the users. Many dashboards are de-

signed and used for quick and easy perception of in-

formation. Usually this information requires imme-

diate action. The dashboards also have to visualise a

Orlov, P., Ermolova, T., Laptev, V., Mitrofanov, A. and Ivanov, V.

The Eye-tracking Study of the Line Charts in Dashboards Design.

DOI: 10.5220/0005692102030211

In Proceedings of the 11th Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2016) - Volume 2: IVAPP, pages 205-213

ISBN: 978-989-758-175-5

205

considerable amount of information. The tasks given

to the subjects during our study were designed to be

quick, easy and yielding the quickest responses in or-

der to model a simpliﬁed dashboard environment.

2 BACKGROUND RESEARCH

2.1 Information Graphics

Two of the most popular graph types of information

graphics (graphs) are line and bar graphs. Line graphs

are used to display data or information that changes

over time, thus they are used for conveying data trends

and distributions. Researchers and designers for a

number of reasons often criticize line graphs. Nev-

ertheless, they remain the most popular tool for data

visualisation due to their simplicity.

(Cleveland and McGill, 1985) pointed out that

”today graphs are a vital part of communication in

science and technology, business, education, and the

mass media” (p. 530). Even though William Playfair

introduced graphs nearly 200 years ago, there remain

many questions about their proper use, design and ef-

fects on data perception. There are many studies of

information graphs (e.g. Tufte, 1983), but a limited

amount of scientiﬁc data exists on the subject.

Studies had been carried out in order to determine

graph perception. A number of researchers addressed

the problem of ﬁnding the best task- or person-

orientated visualisations (Newman and Scholl, 2012).

Small multiples is a term introduced by Edward

Tufte in 1983 to describe similar graphs and charts

with the same axes and scales that can be compared

easily. As Tufte puts it: ”for a wide range of problems

in data presentation, small multiples are the best de-

sign solution.” (Tufte, Envisioning Information, 1990,

p. 67) It was found that small multiples allow users

to compare all different alternatives available to them

and to choose the one that interests them (van den

Elzen and van Wijk, 2013). This means that the best

way to develop a task that consists of ﬁnding the right

graph among other graphs would be to use small mul-

tiples.

Small multiples are often used during dashboard

design. Theoretically, however, small multiples have

to be placed in a logical order (Tufte, 1983), which is

a very hard task, especially if the dashboard is set to

display varying information that changes over time.

2.2 Modern Dashboard Design

Challenges

Dashboards are visual displays and the information

on a dashboard is presented visually, combining text

and graphic (Few, 2006). The idea of digital dash-

boards was devised in the mi ddle of 1970s. When

the dashboards gained popularity in 2000s, each com-

pany took care of their own dashboards. Development

and popularity of the web technology now permits the

companies to buy dashboards from third-party com-

panies.

There are several deﬁnitions of the word ”dash-

board”. Dashboard is a tool that helps to identify

trends, patterns and anomalies for effective decision-

making (Brath and Peters, 2004). This tool is in-

tuitive, ﬂexible and easy to use (Harel and Sitko,

2003). Designed for rapid monitoring, dashboards

display the most important information on a single

screen (Brath and Peters, 2004).

There is a number of recommendations on dash-

board design. However, as (Few, 2006) puts it, the

process of dashboard design is made too complicated

due to the lack of solid design recommendations.

Dundas Data Visualization suggests several char-

acteristics of a good dashboard (2013): ”[a dash-

board] communicates with clarity; quickly, and com-

pellingly [...] it applies the latest understanding of

human visual perception to the visual presentation of

information”.

There is a question about how many different el-

ements and what amount of vital information can be

displayed on the same screen simultaneously. Few

suggests that ”across the entire dashboard, non-data

pixels- any pixels that are not used to display data, ex-

cluding a blank background – should be reduced to a

reasonable minimum” (Few, 2006) (p 97). This corre-

sponds with Edward Tufte’s (1983) ’Data Ink Ratio’:

”A large share of ink on a graphic should present data-

information, the ink changing as the data changes”

(p.96).

Following the Data Ink Ratio allows designers to

create effective single information graphs. Neverthe-

less, what is a ”reasonable minimum” and how much

information can be displayed on the same screen

without the user having problems with data percep-

tion (Few, 2006). When the data displayed to the sub-

ject is greater than subject’s working memory a cog-

nitive overload occurs (Mayer, 1989) and the user ex-

periences problems with decision-making and fails to

respond adequately to the data.

Completion and accuracy of a task may indicate

the degree of graph effectiveness; however, a more

in-depth study is required (Goldberg and Helfman,

IVAPP 2016 - International Conference on Information Visualization Theory and Applications

206

2010). One of the tools used to understand how the

subject perceives graphs is an eye-tracking technol-

ogy.

There had been a number of discussions on

whether graph trends also affect the information per-

ception. This idea is important to consider whilst

studying the graph’s effects on the subjects. (Newman

and Scholl, 2012) tested the above assumption and

determined that there are indeed some differences in

how rising and declining graphs are perceived by the

subjects. The above research indicated that subjects

tend to regard the rising graphs in a positive manner,

whilst the declining graphs are seen in a highly nega-

tive manner. That is to say, the graph trends affect the

subjects’ perception.

2.3 Eye-tracking

Eye tracking is used to understand the eye movements

and gaze positions (Orlov et al., 2014). Based on the

eye movements, eye-tracking can be used to deter-

mine whether the type of diagram affects how the sub-

ject reads it (Goldberg and Helfman, 2010) or whether

the subject experiences a cognitive overload whilst

working with a distracting dashboard (Bera, 2014).

Two metrics are commonly used in eye tracking:

eye ﬁxations and eye saccades (Jacob and Karn, 2003;

Duchowski, 2007). The human eye scans the scene by

rapid eye movements (called saccades), followed by

ﬁxations (relatively stable movements). During sac-

cades, the vision is essentially suppressed.

During ﬁxations, eyes remain almost motionless.

A ﬁxation lasts typically for about 200-300 millisec-

onds (Yarbus, 1965) and is considered to show where

viewer’s attention is directed (Rayner, 2009). Fixa-

tion count is a value revealing the level of cognitive

processing of a subject (Bridgeman et al., 1994; Jacob

and Karn, 2003). Some studies show that a dashboard

with excessive information increases the overall ﬁxa-

tion time and count (Bera, 2014).

3 EXPERIMENT 1

3.1 Method

Hypothesis: The number of graphs simultaneously

displayed on the screen and their type (rising, declin-

ing and stable) affects the process of obtaining visual

information. From our hypothesis, we formed 6 re-

search questions:

• RQ 1. Will the number of graphs inﬂuence the

ﬁxation duration?

• RQ 2. Will the number of graphs inﬂuence the

task solving time?

• RQ 3. Will the number of graphs inﬂuence the

number of ﬁxations?

• RQ 4. Will the type of graphs inﬂuence the ﬁxa-

tion duration?

• RQ 5. Will the type of graphs inﬂuence the task

solving time?

• RQ 6. Will the type of graphs inﬂuence the num-

ber of ﬁxations?

Rising, declining and stable line graphs were used

(Figure 1) as stimuli. To eliminate the human fac-

tor from data generation, a Python script was writ-

ten to generate graph values procedurally. The values

were synthetically generated on purpose, as using real

data would have brought certain difﬁculties and extra

variables into our experiments. Even, if a consider-

able amount of required data were obtained, it would

still generate uneven graphs that would not have clear

trends. Using speciﬁc data would also mean that we

would limit the effects of line graphs to a speciﬁc do-

main. All the data used for graph generation did not

carry any speciﬁc information, such as house prices,

or food prices, so that the subjects would pay atten-

tion to the trends, rather than what effects those trends

would have on them in this speciﬁc domain.

According to Cleveland’s theory, graphs, display-

ing slow processes (such as wind speed, water temper-

ature etc.) should follow the banking 45

◦

rule (Cleve-

land and McGill, 1987). Our graphs were generated

in accordance with this rule. The generated values

varied, but the difference between sequential values

was strictly limited to no bigger than three. The stim-

uli graphs were even and controlled with no ﬂuctua-

tion.

All the graphs were given the same basic title.

The graphs were generated in a way similar to that

of (van den Elzen and van Wijk, 2013): the small

multiples were generated using the same parameters

for all instances except one single variable (van den

Elzen and van Wijk, 2013). Small multiples are of-

ten used in dashboards. However, they limit the way

the graphs have to be displayed: theoretically, small

multiples must be placed in a logical order. For our

study, we will adopt the small multiple principle, but

we will deal with a more complicated scheme. The

graphs will be virtually identical but for one variable,

but they will have a random order on the screen. Our

experiment does not aim to test the beneﬁts of small

multiples. Instead, we look at information perception

in a simpliﬁed dashboard environment.

Yost pointed out a number of different actions sub-

jects can carry out whilst examining information dis-

The Eye-tracking Study of the Line Charts in Dashboards Design

207

Figure 2: An example of a 5x5 line graph grid.

Figure 1: Rising, declining and stable graph visualizations

used in the tutorial.

played on the small multiples: identifying, exploring,

comparing and even clustering information (Yost and

North, 2006). This, however, requires keeping the

graphs constant during the entire experiment. How-

ever, in order to answer all the research questions

listed above, we would have to change two important

variables (graphs sizes and active area sizes). For re-

sults that are more reliable only one process (a process

of identifying a particular graph type) can be intro-

duced.

The graphs were placed side by side forming a

grid. The subjects were shown a 2x2, 3x3, 4x4 and,

ﬁnally graph 5x5 grid for each task. The 5x5 grid was

chosen as a reasonable maximum number of graphs

(25), see Figure 2.

For each set, the subjects were given a task of ﬁnd-

ing one type of graph (either rising, declining or stable

graph). The process was split into two parts: subjects

were ﬁrst shown line graphs and, once they found the

correct answer, they switched to the next screen where

the graph grid was copied using grey rectangles. On

the second screen, the subjects were told to click on

the place, where the correct answer to their task was

on the previous screen. This method separated the

data about the subjects’ ﬁxations and the action of tar-

get click.

Before the experiment, all the subjects were

shown a tutorial, where they were told what kinds of

graphs they would be shown and what task they need

to carry out (the task of ﬁnding one particular graph

type in each set of tasks was explained using images

and examples).

After the subjects clicked on their answer on the

screen with grey rectangles, an image of an animal

was displayed. This powerful semiotic image dis-

tracted the subjects. This break introduced a new area

of interest, so the subjects would not have to carry out

the next task, while focusing on the last.

Before carrying out the experiment, a test exper-

iment was performed in order to check our method

against real results and determine any difﬁculties the

subjects may experience during the experiment. For

this experiment, the graph grid took up the same area

on each screen. The number of graphs for each set

of graphs gradually grew (4, 9, 16 and 25 graphs on

a single screen). As the number of graphs grew, the

graphs became smaller. The subjects were asked to

Figure 3: A layout for the 2x2, 3x3 and 4x4 grid for Exper-

iment 1.

IVAPP 2016 - International Conference on Information Visualization Theory and Applications

208

ﬁnd one particular type of graph in each set of graphs

(at ﬁrst they looked for rising graphs, then declining,

then stable). After carrying out the test experiment,

some minor changes were introduced, such as a more

detailed explanation of the tasks in the tutorial.

3.2 Apparatus

The stimulus was presented on a PC with Intel(R)

Core(TM) 2 Duo CPU E8400 @ 3.00G Hz, 3.25GB

memory and the Microsoft Windows XP operating

system. The graphics system consisted of High-

performance monitor BENQ XL2411 24” (53cm x

30cm), 1920 x 1080px with 144Hz. Eye movements

were registered with an eye tracking system SMI

RED250 on a different computer. Fixation detection

was done by SMI iViewX software. The eye track-

ing system was calibrated to the subject’s eyes, using

ﬁve points on a display. The SMI iViewX software

saves the coordinates of the points and all the follow-

ing calculations were done based on the calibration

data (Orlov et al., 2014).

3.3 Procedure

37 subjects were given a task of ﬁnding a certain

type of graph in three sets of graphs. The age of

the subjects varied from 18 to 26 years. The sub-

jects were both undergraduate and postgraduate stu-

dents. We did not perform any vision tests on the sub-

jects. We did however ask the subjects about their

eye-sight. All the subjects reported having normal

vision or wore lenses/glasses, thus making their vi-

sion normal. None of the subjects reported colour-

blindness or any other vision defects. All the students

were designers/studying design, so any vision defects

would have been well known to them and their tutors.

Graphs within the grid were randomized for each

subject; as a result, the answer to the task was ran-

domly positioned within the grid. This allowed us to

eliminate the effects of the graph positioning on graph

perception, as each subject had to look for the correct

answer in a randomized point on the screen.

Regardless of the tutorial, the subjects still took

time to adjust to the task, so, consequently, they re-

quired more time to go through the ﬁrst set of graphs.

To eliminate this delay in our study, we randomized

tasks. Unlike the test experiment, rise-decline-stable

task sequence was no longer retained and each sub-

ject had a different sequence of tasks (i.e. decline-

rise-stable, stable-decline-rise etc.).

3.4 Results

In our study, eye movement data was recorded: ﬁxa-

tion duration, number of ﬁxations per stimuli and total

duration. We also noted whether the answer given by

the subject was correct. The task was simple, so none

of the subject made any mistakes during the experi-

ment.

The number of graphs simultaneously displayed

was determined to have an effect on the ﬁxation dura-

tions (Kruskal-Wallis rank sum test (for none normal

distributed data, like ﬁxations duration (Duchowski,

2007)): p − value = .000 < .05; chi − squared =

21.163). The increase in graph count resulted in a

decrease in ﬁxation durations. From Figure 4 we can

determine that when subjects looked at four graphs,

they had longer ﬁxations, compared to that of the av-

erage length of the ﬁxation for 16 graphs.

However, no correlation between the number of

graphs displayed on the screen and the number of

ﬁxations per stimuli (p − value = .69 > .05;chi −

squared = 1.465) was found. On average, the sub-

jects made around 10 ﬁxations per stimuli.

Figure 4: Experiment 1. Total ﬁxation duration.

Surprisingly, the response time (total duration of

task solving) is not affected by the number of graphs

displayed (p − value = .44 > .05;chi − squared =

2.702).

The data was analysed in order to determine

whether the type (rising, declining and stable) of

graph the subjects were looking for affected the per-

ception of data. The determined that the type of graph

does not affect the number of ﬁxations (F(2, 80) =

.65, p = .524) and the overall time for task solv-

ing (F(1, 81) = .932, p = .337) .The Kruskal-Wallis

rank sum test determined that the type of graph does

not affect the ﬁxation duration (p − value = .265 >

.05;chi − squared = 2.654). None of these differ-

ences was statistically signiﬁcant.

The Eye-tracking Study of the Line Charts in Dashboards Design

209

3.5 Discussion

The ﬁrst set of questions is aimed to determine the

inﬂuence of the number of graphs on perception. It

was found that this factor inﬂuences the ﬁxation du-

ration (RQ1). We can conclude that our ﬁndings

(see Figure 4) correspond with concepts of attention

landscapes by (Velichkovsky, 2006). Longer ﬁxa-

tions (with shorter saccades) used for the answering

the What question corresponds with the time for ob-

taining the meaning of the visual zone. An oppo-

site happens with shorter ﬁxations (with longer sac-

cades). Shorter ﬁxations are typically used for the

searching tasks (the Where question) (Burgert et al.,

2007). From this we can conclude, that subjects’ vi-

sual searching strategy switched when the number of

graphs reached 16.

No correlation between the number of graphs and

the number of ﬁxations (RQ 3) was found. In our

research, no inﬂuence of graph types was determined

(RQ 4, RQ 5 and RQ 6). This result may be explained

by the fact that subjects perceive different graph types

in a similar manner.

The number of graphs does not inﬂuence the task

solving duration (RQ2). This might be explained by

the fact that the physical size of the graphs was dif-

ferent: the smaller the number of graphs, the bigger

size of the each graph. In order to exclude the size

factor from the experiment conditions the second ex-

periment was conducted.

4 EXPERIMENT 2

4.1 Procedure

38 undergraduate and postgraduate students took part

in the second experiment. In order to eliminate the

bias of learning, new subjects were introduced into

the experiment. The subjects were from 18 to 28 years

of age. The subjects reported about their vision and

vision defects. The subjects, who did not have 20/20

vision, were asked to wear their prescribed glasses

and lenses. This means that in experiments one and

two all the subjects had normal vision without any

strong or evident vision defects.

The same Method and Hypothesis was used. We

kept the idea of graph randomization and task se-

quence randomization. The basic set up of the exper-

iment was the same as for the ﬁrst experiment. This

time, however, we took a different approach to how

the graphs were displayed on the screen. Unlike the

previous experiment, the size of the graphs remained

constant (256x153 px), consequently the area taken

up the graphs on the screen changed noticeably (from

512x306 px for the 2x2 grid up to 1280x765 px for

the 5x5 grid), see Figure 5.

Figure 5: A layout for the 2x2, 3x3 and 4x4 grid for Exper-

iment 2.

4.2 Results

We found that the number of graphs affects the num-

ber of ﬁxations (ANOVA: F(1, 774) = 126.38, p =

.000). Figure 6 (left) provides the mean of the ﬁxa-

tion numbers. Generally, a smaller number of graphs

had a smaller number of ﬁxations.

Further analysis showed that the time required for

task completion is also inﬂuenced by the number of

graphs (ANOVA: F(1, 94) = 11.024, p = .001). Fig-

ure 6 (right) shows the positive correlation between

the number of graphs and the total task solving time.

We did not determine the effect of the number of

graphs on the ﬁxation duration (The Kruskal-Wallis

rank sum test: p − value = .07 > .05; chi −squared =

7.047).

The overall ﬁxation duration, however, almost

doubled during the second experiment, going from

the mean of 13.317 for the 2x2 grid to 21.72 for the

5x5 grid.

It was determined that there was no correlation be-

tween the types of graphs the subjects were tasked

to ﬁnd on the screen and ﬁxation duration (The

Kruskal-Wallis rank sum test: p − value = .829 >

.05;chi − squared = 0.375), ﬁxation count (ANOVA:

F(2, 93) = .419, p = .659) and time for task comple-

tion (ANOVA: F(2, 93) = .635, p = .532).

4.3 Discussion

This experiment was designed in order to determine

the importance of the graph count and type in dash-

board design. We found that the number of graphs

inﬂuences the number of ﬁxations. The number of

graphs (RQ 2, RQ 3) also inﬂuences the total task

solving time. The number of graphs (RQ 1) does not

affect the ﬁxation durations.

IVAPP 2016 - International Conference on Information Visualization Theory and Applications

210

Figure 6: Experiment 2. Left: Number of ﬁxations per each graph grid type. Right: Total time required for the subjects to

respond to the task.

When compared to the Experiment 1, we can con-

clude that the ﬁxation durations are affected greatly

by the graph sizes. There are several possible ex-

planations for this result. It could be because when

a greater amount of details was available to the sub-

jects, subjects spent more time studying the details on

the graph rather than solving the task. This could also

be explained by considering that, during perception,

each graph’s importance and the amount of informa-

tion it displays for the user depends heavily on the

area it occupies on the screen.

Unlike the ﬁrst experiment with no signiﬁcant

change in ﬁxation count, during the second experi-

ment mean ﬁxation count almost doubled from 5.542

ﬁxations per 2x2 stimuli to 10.292 ﬁxation for 5x5

stimuli during the second experiment.

The second group of research questions focuses

on the graph’s type. As in an Experiment 1, here we

also did not ﬁnd any inﬂuences of this factor on the

ﬁxation duration, task solving time and the number of

ﬁxations (RQ 4, RQ 5, and RQ 6).

5 GENERAL DISCUSSION

The results display an evident difference between the

effects of the two possible approaches to line graph

distributions on the screen.

Subjects studied bigger graphs with longer ﬁxa-

tions. When working with a graph that takes up a

relatively big area on the screen, the subjects stud-

ied it in more detail. For the smaller graphs, the sub-

jects opted for a ’glancing over’. In the ﬁrst experi-

ment, the subjects perceived a graph grid of four line

graphs and a grid of 25 graphs with the same speed.

In the second experiment, when the graphs were all

of the same size, four graphs were perceived faster

than 25. This means that the useful area of the screen

(not the graph count) correlates to the time required

to solve the task. During dashboard design, the de-

signers should consider this ﬁnding: the size of the

graph depends heavily on the level of detail that is

needed for a speciﬁc task. The information for imme-

diate reaction and quick overall assessment should be

displayed on a small area relative to the screen.

Our study shows that if four graphs are being dis-

played, their total size is recommended to be 16% of

the total active screen and 36% of the screen for 9

graphs. Graphs that convey a more in-depth infor-

mation have to be larger, as larger graphs are always

studied in more detail.

It is not necessary to ﬁll the screen with large

graphs. Designers have to sacriﬁce the useful screen

area in order to increase the speed of problem solv-

ing and information assessment. Nevertheless, it is

important that the rest of the available space must re-

main empty, as adding images and other graphics to

the dashboard results in so called ”distracting dash-

board” (Bera, 2014).

It was determined that the graph type does not af-

fect the subjects’ perception. There is no evidence

showing that the subjects perceive more ’positive’ ris-

ing graphs better than the declining graphs, or ﬁnd

stable graphs more readily. This remains true for both

experiment setups. (Goldberg and Helfman, 2010)

determined that line graphs do not promote left to

right scanning of the graph’s body, instead the sub-

jects opt for a ’back and forth’ approach. This be-

haviour supports our results: if the subject studies the

graph using a back and forth eye movement, there

should be no actual preference to which graph types

the subject ﬁnds faster. If the graphs were to be

scanned from left to right, declining graphs would

have been easier to detect, as their lines start at the

upper left corner of the axis.

On the other hand, the study by (Newman and

Scholl, 2012) shows that the rising and declining

graphs, in his case bar graphs, are still perceived in

The Eye-tracking Study of the Line Charts in Dashboards Design

211

a different manner: rising graphs are seen by the sub-

jects to be more optimistic than they really are, while

the subjects interpret a declining graph in a nega-

tive manner. As our subjects were not questioned on

whether they found it easier to ﬁnd a certain graph

type, we cannot conclude whether the graph types

have absolutely no effect on users in terms of positive

or negative perception and correct decision-making.

6 CONCLUSION

Our ﬁndings can be applied to a whole number of

different domains e.g. web design. While not be-

ing classic dashboards, web pages, as well as other

user interfaces, often carry out the same tasks as dash-

boards (display information graphs, allow users to

make immediate decisions, convey important infor-

mation). The paper could be used as one of the guide-

lines for both web and dashboard designers, when

making choices about graph placements and graph

sizes relative to the screen. Many websites could ben-

eﬁt from the idea that the amount of graphs that is

being displayed simultaneously is not as important as

the size of those graphs (or any other images, text-

blocks or information).

When designing the dashboard (or a dashboard

screen) designers should note that the graphs that oc-

cupy a relatively small amount of the active screen

convey information quicker than larger ones.

It also important to note that during the ﬁrst ex-

periment a change in a visual searching strategy by

the subjects was noted. This is an interesting concept

that can be explored further in future works.

The paper deals with the idea of whether bigger

graphs are better for perception speed and whether

a greater number of graphs makes information per-

ception difﬁcult. Our ﬁndings showed that the useful

screen area corresponds to the speed of the problem

solving. This means that dashboard designers should

sacriﬁce screen area, use smaller graphs and leave

empty spaces if the dashboard (dashboard screen they

are designing) is used for basic immediate assess-

ment. If a more in-depth assessment is required,

larger graphs are recommended.

The paper also contributes to the studies of infor-

mation perception in terms of both user interfaces and

infographics. The paper may serve as the base for a

more in-depth study of information perception. As

a follow-up of the experiments, an eye-tracking study

should be carried out on other types of graphs, such as

bar graphs or pie charts, in order to ﬁnd out whether

our conclusions are true for other types of information

graphs. Adding graphs that are more complicated or

information from a speciﬁc domain, greater distract-

ing points of interest or longer and more demanding

tasks (such as exploring, concluding, comparing and

clustering (Yost and North, 2006)) could also be in-

troduced. This would help to understand how design-

ers should act, when their dashboard needs to convey

more information (not just help to identify a graph

with the correct trend relative to the task) or when

only brief and immediate information needs to be dis-

played.

REFERENCES

Bera, P. (2014). Do Distracting Dashboards Matter? Ev-

idence from an Eye Tracking Study. In Information

Systems: Education, Applications, Research, pages

65–74. Springer.

Brath, R. and Peters, M. (2004). Dashboard Design: Why

Design is Important. Data Mining Review/Data Min-

ing Direct, (October).

Bridgeman, B., Van der Heijden, A. H. C., and

Velichkovsky, B. M. (1994). A theory of visual sta-

bility across saccadic eye movements. Behavioral and

Brain Sciences, 17(02):247–258.

Burgert, O.,

Orn, V., Velichkovsky, B. M., Gessat, M.,

Joos, M., Strauss, G., Tietjen, C., Preim, B., and Her-

tel, I. (2007). Evaluation of Perception Performance

in Neck Dissection Planning using Eye Tracking and

Attention Landscapes. In Jiang, Y. and Sahiner, B.,

editors, SPIE Proceedings, page 9 pages. Society of

Photo-Optical Instrumentation Engineers.

Cleveland, W. S. and McGill, R. (1985). Graphical per-

ception and graphical methods for analyzing scientiﬁc

data. Science, 229(4716):828–833.

Cleveland, W. S. and McGill, R. (1987). Graphical per-

ception: The visual decoding of quantitative infor-

mation on graphical displays of data. Journal of the

Royal Statistical Society. Series A (General), pages

192–229.

Duchowski, A. (2007). Eye Tracking Methodology.

Springer Science+Business Media, London.

Few, S. (2006). Information dashboard design. O’Reilly.

Goldberg, J. H. and Helfman, J. I. (2010). Comparing In-

formation Graphics: A Critical Look at Eye Tracking.

In 3rd BELIV’10 Workshop: Beyond time and errors:

Novel evaluation methods for information visualiza-

tion., pages 71–78. ACM.

Harel, E. C. and Sitko, T. D. (2003). Driving Higher Edu-

cation Decisions. Research Bulletin, 1(19):2–10.

Jacob, R. J. K. and Karn, K. S. (2003). Eye tracking in

HumanComputer interaction and usability research:

Ready to deliver the promises. The Mind’s Eye:

Cognitive and Applied Aspects of Eye Movement Re-

search, pages 573–605.

Newman, G. E. and Scholl, B. J. (2012). Bar graphs de-

picting averages are perceptually misinterpreted: The

IVAPP 2016 - International Conference on Information Visualization Theory and Applications

212

within-the-bar bias. Psychonomic Bulletin & Review,

19(4):601–607.

Orlov, P. A., Laptev, V. V., and Ivanov, V. M. (2014).

For the question about eye-tracking applying. St. Pe-

tersburg State Polytechnical University Journal. Com-

puter Science. Telecommunication and Control Sys-

tems., 5(205):84–94.

Rayner, K. (2009). Eye movements and attention in reading,

scene perception, and visual search. Quarterly journal

of experimental psychology (2006), 62(8):1457–1506.

van den Elzen, S. and van Wijk, J. J. (2013). Small multi-

ples, large singles: A new approach for visual data ex-

ploration. In Computer Graphics Forum, volume 32,

pages 191–200. Wiley Online Library.

Velichkovsky, B. M. (2006). Cognitive Science: Founda-

tions of epistemic psychology. [Kognitivnaya nauka :

Osnovy psihologii poznaniya], volume 2. Smusl: Pub-

lisher center Akademia, Moscow.

Yarbus, A. L. (1965). Eye movements and vision [Rol dvi-

jenyi glaz v procecce zrenia]. ”Nauka”, Moscow.

Yost, B. and North, C. (2006). The perceptual scalability

of visualization. IEEE Transactions on Visualization

and Computer Graphics, 12(5):837–844.

The Eye-tracking Study of the Line Charts in Dashboards Design

213