Unit VI

Research Direction in Visualizations

• Steps in Designing Visualizations

• Problems in Designing Effective Visualizations

• Issues of Data

• Issues of Cognition, Perception , and Reasoning

• Issues of System Design, Evaluation

• Hardware and Applications

1. Steps in Designing Visualizations

Creating a visualization involves deciding how to map the data fields to

graphical attributes, selecting and implementing methods for modifying views,
and choosing how much data to visualize.

Additional information regarding the data being shown (e.g., labels) and the
mapping (e.g., a color key) are also essential to facilitate interpretation, and
must be integrated into the visualization.

Design stages:
1. Intuitive Mappings from Data to Visualization
2. Selecting and Modifying Views
3. Information Density—When Is It Too Much or Too Little?
4. Keys, Labels, and Legends
5. Using Color with Care
6. The Importance of Aesthetics
Intuitive Mappings from Data to Visualization

To create the most effective visualization for a particular application, it is

critical to consider the semantics of the data and the context of the typical user.

By selecting data-to-graphics mappings that cater to the user’s domain-specific

mental model, the interpretation of the resulting image will be greatly
facilitated.
Intuitive mappings also lead to more rapid interpretation, as translation time is
reduced.

For example, in Figure 12.1, images of planets are used to plot the relationship
between the distance from the planet to the sun and the duration of its orbit.

Using intuitive scatterplot symbols to show the distance from planets to the sun
versus the duration of a single orbit.

Mapping spatial data attributes, such as longitude and latitude, to screen

position is perhaps the most common and intuitive mapping found in
visualizations.

Some of the earliest visualizations took advantage of the ability of humans

to correlate position on the drawing medium with position in the three-
dimensional world
Selecting and Modifying Views:

Except for fairly simple data sets, one view is rarely sufficient to convey all the
information contained in the data.

The key to developing an effective visualization is to be able to anticipate the

types of views and view modifications that will be of most use to the typical
user, and then provide intuitive controls for setting and customizing the views.

View modifications fall into a number of categories, and their inclusion as part
of the functionality should be considered, based on user priorities.
 Scrolling and zooming operations are needed if the entire data set cannot be
presented at the resolution desired by the user.

. Color map control is almost always desirable, minimally supporting a set of

different pallettes, and preferably offering the user control of either individual
colors or the complete pallette.

3. Mapping control allows users to switch between different ways of visualizing

the same data. Features of the data that are hidden in one mapping may stand
out in others (Figure 12.2).

4.Scale control permits the user to modify the range and distribution of values
for a particular data field prior to its mapping. Similarly, clipping and other
forms of filtering allow the user to focus on data subsets.

5.Level-of-detail controls provide the ability to eliminate or highlight detail,

supporting views at different levels of abstraction. Depending on the task at
hand, a user may need to repeatedly switch between several distinct levels
(Figure 12.3).

Three views of the IRIS data set (scatterplot matrix, star glyphs, and parallel
coordinates). (Image from XmdvTool.)
 Levels of detail in maps. (Images courtesy of Google Maps c 2008 Google)

Information Density—When Is It Too Much or Too Little?

One of the key decisions one makes when designing a visualization is

determining how much information to display.

This gives rise to two extreme situations.

The first, which might be called “gratuitous graphics,” occurs when there is
very little information to present.

Many examples of graphics

can be found that convey only two or three distinct values, such as the
percentage of males and females within a particular sample (this actually can
be communicated with one number).
The other extreme, trying to convey too much information, is also a common
problem.
 Excessive information content can lead to confusion, intimidation, and
difficulties in interpretation on the part of the viewer.

Important information contained within the data can be lost or de-emphasized

on a cluttered display, and viewers may have a hard time determining where to
focus their attention.

There are many effective solutions to the problem of excessive information

content in a visualization.

One method is to give the user the option of disabling or enabling different
components of the display

Another solution is to use multiple screens, either as disjoint panes or with

partial occlusions. This method makes better use of screen space, while making
each of the individual pieces of data readily available.

Another common cause of cluttered displays is large or unevenly distributed

data sets.

Keys, Labels, and Legends

A common problem with many visualizations is that insufficient information is

provided to the user to allow unambiguous and accurate interpretation.

This supporting information should begin with a detailed caption indicating the
particular data fields being displayed, and the mappings that were used.
A complex visualization with and without captions/ticks/legends
• grid or tick marks should be displayed to convey the ranges and values of
interest for numeric fields when absolute judgments are important, and all
axes should be labeled with appropriate units.
• Figure 12.4 highlights the importance of this supplementary information.
• The use of grid and tick marks can be both a boon and a curse to the
visualization.
• The actual positions of the markings can also have a bearing on how readily
the data is interpreted.
• Based on the semantics of the data, certain gaps between markings may
make more sense to the user than others.
 • Grid spacings: (a) illogical; (b) logical.

Logical and illogical data ranges

• One final rule of thumb pertains to the use of multiple frames or windows.
• It is important to follow a consistent labeling and gridding scheme.
• Changing the position of labels and keys or the range of values shown (for
the same field) can cause confusion and increase the risk of
misinterpretation.
• If range changes are necessary (e.g., for views that differ in level of detail),
the label, as well as the grid markings, should convey the change.
• The designer must also decide which range of values is to be displayed.
• There is always the risk of misinterpretation when the expected range of
values is not shown.
Using Color with Care:
• One of the most frequently misused parameters in visualization design is
that of color.
• Selecting the wrong color map or attempting to convey too much
quantitative information through color can lead to ineffective or misleading
visualizations.
 • Too many colors versus a moderate number of colors.
• The following guidelines can assist in the effective use of color in
visualization:
1. If the visualization task involves absolute judgment, keep the number of distinct
numeric levels low (see Figure 12.8).
2. Use redundant mappings if possible, e.g., map a particular field to both color and
size , to improve the chances of the data being communicated accurately.
4.In creating a color map for conveying numeric information, make sure that both
hue and lightness are changed for each entry.
5.Include a labeled color key to help user interpret the colors (see the previous
section).
• Color can add significant visual appeal to a visualization, but can also
significantly decrease the effectiveness of the communication process
The Importance of Aesthetics:
• Once we have ensured that our designed visualization conveys the desired
information to the user (function), the final step is to assess the aesthetics
(form) of the results.
• The best visualizations are both informative and pleasing to the eye.
• There are many guidelines for attractive visualization design that can be
drawn from the art and graphic design communities. These include:
• Focus. The viewer’s focus should be drawn towards the part of the
visualization that is most important. If the important components are not
 sufficiently emphasized, viewers don’t have sufficient cues for guiding their
inspection (see Figure 12.11).

• (a) Subdued streamlines vs. (b) highlighted streamlines from OpenDX

• Balance: The screen space should be used effectively, with the most
important components in the center. Emphasis should not be given to any
particular border (Figure 12.12).

• Simplicity. Don’t try to cram too much information in one display (see
Section 12.1.3), and don’t use graphics gimmick simply because they are
available (e.g., using 3D Phong shaded histograms when a bar or line chart
could convey the same information).
• A useful procedure to follow once a visualization has been designed is to
iteratively remove features and measure the loss of information being
conveyed.
•

• Progression from a cluttered chart to a simplified chart

• 2. Problems in Designing Effective Visualizations:
• some of the common problems found in visualizations, which can occur
even if the steps outlined above are followed.
• These problems have a deeper root, and relate to decisions regarding what to
visualize and what is the most appropriate method to use
• Some of the problems involve intentional or inadvertent data distortion,
which can lead to misinterpretation.
• Others involve hiding the real data behind “cleaned” versions or excessive
supporting graphics.
• In all cases, steps can be taken to improve the quality and “honesty” of the
visualization.
• 1. Misleading Visualizations
• 2. Visual Nonsense—Comparing Apples and Oranges.
• 3. Losing Data in the Chart Junk
• 4. Raw versus Derived Data
• 5. Absolute versus Relative Judgment
• Misleading Visualizations
• One of the foremost rules of visualization is that the image should be an
accurate depiction of the data.
• However, throughout history, there are examples of how visualizations from
distorted data have been used to sway opinions and lie to the audience.
• These so-called “viz lies” can be found everywhere, from the most
prestigious journals to company portfolios.
• Data scrubbing:Raw data can often be very rough in form, and the
temptation when creating a visualization is to remove some of the
roughness.
• Outlier removal is a common tactic in this situation.

•
• The problem with data scrubbing: (a) raw data showing lack of correlation;
• (b) scrubbed data revealing false correlation.
• Unbalanced scaling:Scaling is a powerful tool in visualization, since careful
selection of scale factors can reveal patterns and structures not visible in
unscaled views.
• However, scaling can be used to deceive the viewer into believing that a
trend is stronger or weaker than supported by the data. This can lead to what
Tufte refers to as the “lie factor,” which is the ratio between the raw data
change and the change as depicted in the visualization.
•
• Vis Lies: perspective distorts size in favor of closer objects.
• Range distortion:As mentioned in an earlier section, viewers often have an
expectation about the ranges for a particular data dimension; by setting this
range to be significantly different from this expectation, the user may be
deceived into misinterpretation.
• Abusing dimensionality:that errors in interpretation rise with the power of
the dimensionality being portrayed.
• Thus, our errors in judging volume are much worse than those for area,
which in turn are worse than those for length
• Visual Nonsense—Comparing Apples and Oranges:Visualizations are
designed to convey information, and it is important that the information be
meaningful.
• Visualizations are often created by combining data sets from different
sources.
• The visualization designer should attempt to avoid creating nonsense
graphics before they are presented to users.
• Losing Data in the Chart Junk:the importance of including labeled grid or
tick marks on visualizations that require quantitative assessment.
• The excessive use of such markings is an example of what Tufte referred to
as “chart junk.”
• Chart junk can be defined as any supplementary (nondata) graphics in a
visualization that are not necessary for the accurate interpretation of the data.
• In some visualization tasks, users can switch between qualitative overviews
and quantitative analysis.
• Raw versus Derived Data:A common practice is to compute an analytic
model of the data using curve/surface fitting to obtain a more visually
appealing result.
• In some visualizations, it is common practice to throw out all of the raw data
and only show the smooth approximation derived from that data.
•

(a) Raw data plot with fitted curve; (b)

only fitted curve.
• Absolute versus Relative Judgment:This implies that visualizations that
depend too heavily on users performing accurate measurements of graphical
attributes such as position, length, and color will result in problems in
interpretation.
• One means of combating this human limitation is to design visualizations
that either rely on relative rather than absolute judgment, or that are
 restricted to only using a small number of distinct values for each graphical
attribute being used to convey information.
• Bounding boxes, grids, and tick marks are all excellent tools for converting
an absolute judgment task to one that depends more on relative judgment.
3. Issues of Data:
• Many of the current and proposed future research activities are centered on
expanding the characteristics of data that can be effectively visualized. Some
of these are discussed here.
• Scale:Perhaps the most frequently addressed problem dealing with data is
finding solutions to coping with ever-increasing sizes for data sets.
• Static versus dynamic:While most visualization techniques to date have
been developed with the assumption that data is static (e.g., in files or
databases), a growing interest is in the area of visual analysis of dynamic
data [405].
• An increasing number of streaming data sources are being studied in the
database and data mining communities, and efforts to perform visual
analysis on this type of data are starting to emerge.
• Spatial versus nonspatial data:A growing number of application areas for
visualization include both spatial and nonspatial data, including many
scientific and engineering fields.
• To provide analysts with a powerful environment for studying this data,
several recent efforts have focused on the integration of the spatial
visualization techniques normally found in scientific visualization with the
nonspatial techniques that are common in information visualization.
• Nominal versus ordinal:The graphical attributes to which we map data in
our visualizations, such as position, size, and color, are primarily
quantitative in nature, while it is quite common to have data that is not
quantitative, such as the name of a gene or the address of an employee.
• If this nominal data is to be used in the visualization, a mapping is needed.
• However, it is also important to ensure that relationships derived from visual
analysis are truly part of the data, and not an artifact of the mapping.
• Structured versus nonstructured:Data can be classified based on the
degree to which it follows a predictable structure For example, tables of
numbers would be considered highly structured, while newspaper articles
may be regarded as unstructured.
 • In between, we can have semi-structured data, such as an email message that
contains both a structured component (sender, time, receiver) and an
unstructured part (message body).
• Time:Time is a special variable (and attribute of data).
• Time in dynamic data provides one view: a volume visualization over time
deals with a physical representation, and a common interactive visualization
uses time as a control.
• Time here can be handled just as in the volume visualization.
• For example, in a data set involving education and health indicator data, a
useful question is how to identify similar patterns involving not just the data,
but time as well.
• Variable quality:While most visualization systems and techniques assume
the data is complete and reliable, in fact most sources of data do not match
these constraints.
• The quality of the data itself may also be problematic; out-of-date
information can have low certainty associated with it, inaccurate sensors
may produce values with significant variability, and manual entry of data
can be error-prone.
4. Issues of Cognition, Perception , and Reasoning
• Many of the foundational concepts in data and information visualization
have their roots in our understanding of human perception, particularly in
aspects of selecting effective mappings of data to graphical attributes such as
color and size.
• Part of this effort will be directed at raising the level at which visual tasks
are performed.
• For example, rather than using the visualization to identify a cluster or trend,
it might instead be focused on building a mental model of the entire
phenomenon being studied.
• While this task may be comprised of several more primitive subtasks, the
visualization tool will need to support the combination and sequencing of
these tasks to enable the analyst to cope with the scale and complexity of the
data and information being gathered and generated.
• These higher-level discoveries can then be used by the analyst to form,
confirm, or refute hypotheses, expand or correct mental models, and provide
confidence in decision-making processes.
 • Beyond decision making, we can also envision the expansion of
visualization in the process of human learning.
• Different visual tools and mechanisms are likely to be needed to address
these very different styles of learning.
• In both problem solving and learning activities, we can also imagine using
visualizations as a mechanism to expand and support the memory process,
which is critical to both activities.
• The extents to which visualization can be applied to this have yet to be
extensively studied and exploited.
5. Issues of System Design Evaluation
• Issues of Evaluation:
• In the early days of visualization research, rigorous evaluation was rarely
performed;
• the assumption was that some visualization was better than no visualization
for many tasks, and that if a new technique were developed, it was sufficient
to simply place a couple of sets of images side by side and do a qualitative
judgement of the results.
• While many strategies have been developed and tested, there are many
avenues of research towards improving the overall process.
• The numerous competitions, including the VAST challenges, are a step in
the right direction.
• The key point to make is that evaluation is a necessary component in
moving a field from ad hoc methods to a real science, and we can learn from
and build upon the evaluation strategies used in many other fields, such as
human-computer interaction and sociology, to help validate the usefulness
and effectiveness of techniques.
• Issues of System Design:
• One of the most crucial research challenges in developing visualization tools
is determining how best to integrate computational analysis with interactive
visual analysis.
• While many visualization systems support a modest number of
computational tools, such as clustering, statistical modeling, and dimension
reduction, and similarly many computational analysis systems support some
amount of visualization, such as visualizing the analysis results, there have
 been no systems developed to date that provide a truly seamless integration
of visual and computational techniques.
For example, one can use visual overviews of data to decide on appropriate
parameters to filtering and clustering algorithms, after which the analyst could
examine a resulting cluster to help select a computational model that best fits the
characteristics of the data subset
• Both visual and computational methods could then be used to ascertain the
goodness of fit for the model.
• Another key problem in visualization system design is the development of
powerful new interaction paradigms to support the user’s tasks.
• Many researchers believe that existing modes of interaction during visual
analysis are ill-suited to the tasks being performed.
• While exciting and novel interaction tools can be seen in the immersive
environment field, these and other types of interactions are either too
lowlevel or are not designed with a particular high-level task in mind.
6. Issues of Hardware
• Whenever computer technology advances, the applications that employ this
technology must be reassessed to see how the advances can be leveraged.
• For visualization, there are several technologies that can and will have an
impact.
• Hand-held displays:Most people these days carry with them at least one
form of digital display, whether it be mobile phones, PDAs, portable games,
or mini-PCs.
• While most visualization systems have been designed for desktop (or larger)
displays, there are significant opportunities to deliver information and data
on these smaller devices.
• Display walls:At the other extreme, large-scale displays, often involving
multiple panels stretching 10–30 feet in each direction, are becoming more
and more common, not only for control centers, but also for investigating
large data and information spaces.
• A better solution would be to redesign the visual analysis environment to
arrange the displays of different types and different views of information in
a way that supports the analysis.
• Immersive environments:Virtual and augmented reality systems have been
frequently used within the visualization field Virtual walk-throughs and fly-
 throughs have been used in a diversity of fields, including architecture,
medicine, a
• A key problem with this technology is the need to render the visualizations
with minimal latency, which has spawned significant research in algorithm
optimization.nd aeronautics.
• Graphical processing units:The development of special-purpose graphics
hardware has actually exceeded the growth in performance of general
purpose CPUs, primarily driven by the computer game industry.
• There is much active research in the visualization community that focuses on
harnessing this computational horsepower [216, 341].
• Interaction devices:Each new device for user interaction with the computer
opens up a wide range of possibilities for use in visualization.
• Voice/sound input and output have been extensively studied, though they are
rarely an integral component of a visualization system.
7. Issues of Applications
• Many advances in the field of visualization have been driven by the needs of
a particular application domain.
• These advances are then often generalized to be applicable to many other
domains and problems.
• Thus, research issues can be divided between those specific to a given
domain and those involving the extension of techniques into other domains.
• Depth-based innovations:To develop truly useful visualization tools for a
particular domain, it is very important for the visualization designer to
understand the domain and the tasks the domain specialist is trying to
perform domain experts attempt to design and develop visualization tools for
their needs, they often produce tools with poor designs and user interfaces;
• A key research challenge is thus to develop strategies to create effective
collaborations between the two groups. Many professional meetings and
workshops have focused on this issue over the past decade, and much can be
learned from the reports that have been produced.
• Breadth-based innovations:Another direction of research is to broaden the
number of applications in which data and information visualization can be
applied.
• Indeed, it is hard to imagine an area in which visualization would not be
applicable, as all areas of society are experiencing a glut of information,
while at the same time, display devices have become ubiquitous.
• In many applications, visual information presentation is rapidly replacing
much of the textual communication, such as weather reports, stock market
behavior, health statistics, and so on. Daily schedules are often best captured
in a graphical presentation of an hourly or daily calendar.
•
• Graphs are used to capture complex social networks, organizational charts,
process flows, and communication patterns.
• Another critical problem is the conversion of data and information into a
format that is amenable to existing visualization techniques.
• Concerted efforts are needed, both in the visualization of unstructured data,
as well as in automatic or semi-automatic conversion of unstructured data
into structured forms, to tap into these rich sources of new visualization
applications.