Lecture Notes in Computer Science 4950

Commenced Publication in 1973

Founding and Former Series Editors:
Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen

Editorial Board
David Hutchison
Lancaster University, UK
Takeo Kanade
Carnegie Mellon University, Pittsburgh, PA, USA
Josef Kittler
University of Surrey, Guildford, UK
Jon M. Kleinberg
Cornell University, Ithaca, NY, USA
Alfred Kobsa
University of California, Irvine, CA, USA
Friedemann Mattern
ETH Zurich, Switzerland
John C. Mitchell
Stanford University, CA, USA
Moni Naor
Weizmann Institute of Science, Rehovot, Israel
Oscar Nierstrasz
University of Bern, Switzerland
C. Pandu Rangan
Indian Institute of Technology, Madras, India
Bernhard Steffen
University of Dortmund, Germany
Madhu Sudan
Massachusetts Institute of Technology, MA, USA
Demetri Terzopoulos
University of California, Los Angeles, CA, USA
Doug Tygar
University of California, Berkeley, CA, USA
Gerhard Weikum
Max-Planck Institute of Computer Science, Saarbruecken, Germany
Andreas Kerren John T. Stasko
Jean-Daniel Fekete Chris North (Eds.)

Information
Visualization
Human-Centered Issues and Perspectives

With 15 colored illustrations

13
Volume Editors

Andreas Kerren
Växjö University
School of Mathematics and Systems Engineering
Computer Science Department
Vejdes Plats 7, 351 95 Växjö, Sweden
E-mail: [email protected]

John T. Stasko
Georgia Institute of Technology
School of Interactive Computing and GVU Center
85 5th St., NW, Atlanta, GA 30332-0760, USA
E-mail: [email protected]

Jean-Daniel Fekete
INRIA Saclay - Île-de-France Research Centre
Bat. 490, Université Paris-Sud
91405 Orsay Cedex, France
E-mail: [email protected]

Chris North
Virginia Tech, Department of Computer Science
and Center for Human-Computer Interaction
2202 Kraft Drive, Blacksburg, VA 24061-0106, USA
E-mail: [email protected]

Library of Congress Control Number: 2008931479

CR Subject Classification (1998): H.2.8, H.5, I.2.4, I.3

LNCS Sublibrary: SL 3 – Information Systems and Application, incl. Internet/Web

and HCI

ISSN 0302-9743
ISBN-10 3-540-70955-X Springer Berlin Heidelberg New York
ISBN-13 978-3-540-70955-8 Springer Berlin Heidelberg New York

This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, re-use of illustrations, recitation, broadcasting,
reproduction on microfilms or in any other way, and storage in data banks. Duplication of this publication
or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965,
in its current version, and permission for use must always be obtained from Springer. Violations are liable
to prosecution under the German Copyright Law.
Springer is a part of Springer Science+Business Media
springer.com
© Springer-Verlag Berlin Heidelberg 2008
Printed in Germany
Typesetting: Camera-ready by author, data conversion by Scientific Publishing Services, Chennai, India
Printed on acid-free paper SPIN: 12438815 06/3180 543210
 Preface

From May 28 to June 1, 2007, a seminar on “Information Visualization – Human-

Centered Issues in Visual Representation, Interaction, and Evaluation” (
07221)
took place at the International Conference and Research Center for Computer
Science, Dagstuhl Castle, Germany. The center was initiated by the German
government to promote informatics research at an international level. It seeks
to foster dialog among the computer science research community, advance aca-
demic education and professional development, and transfer knowledge between
academia and industry.
The primary feature of Dagstuhl is its series of week-long seminars on var-
ious topics in computer science. Dagstuhl seminars are frequently described as
being the most productive academic events that the participant researchers have
ever experienced. The informal and friendly atmosphere fostered at the center
promotes personal interaction between the guests. Traditionally, there is no set
program followed at Dagstuhl seminars. Instead, the pace and the procedure are
determined by the presentations offered during the seminar and the discussion
results. Further general information about Dagstuhl seminars can be found on
the Dagstuhl Castle webpage1 .
Information visualization (InfoVis) is a relatively new research area, which
focuses on the use of visualization techniques to help people understand and
analyze data. While related fields such as scientific visualization involve the pre-
sentation of data that has some physical or geometric correspondence, informa-
tion visualization centers on abstract information without such correspondences,
i.e., information that cannot be mapped into the physical world in most cases.
Examples of such abstract data are symbolic, tabular, networked, hierarchical,
or textual information sources. The ever increasing amount of data generated
or made available every day amplifies the urgent need for InfoVis tools. To give
visualization a firm fundament, InfoVis combines several aspects of different re-
search areas, such as scientific visualization, human-computer interaction, data
mining, information design, cognitive psychology, visual perception, cartography,
graph drawing, and computer graphics.
One main goal of our seminar on information visualization was to bring
together researchers and practitioners from the above-mentioned research areas
as well as from application areas, such as bioinformatics, finance, the geosciences,
software engineering, and telecommunications. Several international conferences
include information visualization topics, each with a slightly different high-level
objective. Another goal of the Dagstuhl seminar was to consolidate these diverse
areas in one joint meeting. The seminar allowed critical reflection on actual
research efforts, the state of the field, evaluation challenges, and future directions.
A detailed report on participation, program, and discussions was published in
the journal Information Visualization, published by Palgrave Macmillan Ltd.
(6(3):189-196, 2007).
1
http://www.dagstuhl.de/en/about-dagstuhl/
VI Preface

This book is the outcome of Dagstuhl Seminar
07221. It documents and

extends the findings and discussions of the various sessions in detail. During the
last day of the seminar, the most important topics for publication were identified
and assigned to those interested. The resulting author groups worked together
to write papers on the chosen topics. In the following, we briefly present the
different papers.

Book Structure
Paper 1 “The Value of Information Visualization”, provides a discussion of
issues surrounding the utility and benefits of InfoVis. The paper identifies
why communicating the value of InfoVis is more difficult than in many other
areas, and it provides a number of arguments and examples that help to
illustrate InfoVis’ value.
Paper 2 “Evaluating Information Visualizations”, discusses the challenges as-
sociated with the evaluation of InfoVis tools and approaches. Different types
of evaluation are described as well as the advantages and disadvantages of
different empirical methodologies.
Paper 3 “Theoretical Foundations of Information Visualization”, addresses an
important issue: InfoVis, being related to many other diverse disciplines, suf-
fers from not being based on a clear underlying theory. Drawing on theories
within associated disciplines, three different approaches to theoretical foun-
dations of information visualization are presented: data-centric predictive
theory, information theory, and scientific modeling.
Paper 4 “Teaching Information Visualization”, presents the results of a survey
about InfoVis-related courses that was distributed to the Dagstuhl atten-
dees during the seminar. It also summarizes the discussions about teaching
held by the attendees during a special session on that topic. The paper in-
cludes the perspectives of three seminar participants in relation to their own
InfoVis-related teaching experiences.
Paper 5 “Creation and Collaboration: Engaging New Audiences for Informa-
tion Visualization”, discusses creation and collaboration tools for interactive
visualization. The paper characterizes the increasingly diverse audience for
visualization technology, and it formulates a design space for new creative
and collaborative tools to support these users.
Paper 6 “Process and Pitfalls in Writing Information Visualization Research
Papers”, identifies a set of pitfalls and problems that recur in many InfoVis-
related papers, using a chronological model of the research process. The
aim of this paper is to help authors avoid these pitfalls and write better
papers. Reviewers might also find these pitfalls interesting to consider when
evaluating the merits of a paper.
Paper 7 “Visual Analytics: Definition, Process, and Challenges”, describes the
related and growing field of visual analytics. The paper explains the perceived
difference between visual analytics and InfoVis, and it identifies the technical
challenges faced by visual analytics researchers. The paper concludes by
describing a number of visual analytics applications.
 Preface VII

Acknowledgments

We would like to thank all those who participated in the seminar for the lively
discussions as well as the scientific directorate of Dagstuhl Castle for giving us
the opportunity to organize this event. The abstracts and talks can be found on
the Dagstuhl website for this seminar2 . In addition, we are also grateful to all
the authors for their valuable time and contributions to the book. Last but not
least, the seminar would not have been possible without the great help of the
staff of Dagstuhl Castle. We would like to thank all of them for their assistance.

March 2008 Andreas Kerren

John T. Stasko
Jean-Daniel Fekete
Chris North

2
http://www.dagstuhl.de/07221
 Table of Contents

Part I. General Reflections

The Value of Information Visualization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Jean-Daniel Fekete, Jarke J van Wijk, John T. Stasko, Chris North

Evaluating Information Visualizations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

Sheelagh Carpendale

Theoretical Foundations of Information Visualization . . . . . . . . . . . . . . . . . . 46

Helen C. Purchase, Natalia Andrienko, T.J. Jankun-Kelly,
Matthew Ward

Teaching Information Visualization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

Andreas Kerren, John T. Stasko, Jason Dykes

Part II. Specific Aspects

Creation and Collaboration: Engaging New Audiences for Information
Visualization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Jeffrey Heer, Frank van Ham, Sheelagh Carpendale, Chris Weaver,
Petra Isenberg

Process and Pitfalls in Writing Information Visualization Research Papers 134

Tamara Munzner

Visual Analytics: Definition, Process, and Challenges . . . . . . . . . . . . . . . . . . 154

Daniel Keim, Gennady Andrienko, Jean-Daniel Fekete,
Carsten Görg, Jörn Kohlhammer, Guy Melançon

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

 The Value of Information Visualization

Jean-Daniel Fekete1 , Jarke J. van Wijk2 , John T. Stasko3 , and Chris North4
1
Université Paris-Sud, INRIA, Bât 490,
F-91405 Orsay Cedex, France,
[email protected],
http://www.aviz.fr/~ fekete/
2
Department of Mathematics and Computing Science,
Eindhoven University of Technology, P.O. Box 513,
5600 MB EINDHOVEN, The Netherlands,
[email protected],
http://www.win.tue.nl/~ vanwijk/
3
School of Interactive Computing, College of Computing & GVU Center,
Georgia Institute of Technology, 85 5th St., NW,
Atlanta, GA 30332-0760, USA,
[email protected],
http://www.cc.gatech.edu/~ john.stasko
4
Dept of Computer Science, 2202 Kraft Drive,
Virginia Tech, Blacksburg, VA 24061-0106, USA,
[email protected],
http://people.cs.vt.edu/~ north/

Abstract. Researchers and users of Information Visualization are con-

vinced that it has value. This value can easily be communicated to others
in a face-to-face setting, such that this value is experienced in practice.
To convince broader audiences, and also, to understand the intrinsic
qualities of visualization is more difficult, however. In this paper we con-
sider information visualization from different points of view, and gather
arguments to explain the value of our field.

1 Problems and Challenges

This paper provides a discussion of issues surrounding the value of Information
Visualization (InfoVis). The very existence of the paper should alert the reader
that challenges do exist in both recognizing and communicating the field’s value.
After all, if the value would be clear and undisputed, there would be no need
to write the paper! Unfortunately, the current situation is far from that. By its
very focus and purpose, InfoVis is a discipline that makes the recognition of
value extremely difficult, a point that will be expanded below.
Why is showing value important? Well, today’s research environment places
great importance on evaluation involving quantifiable metrics that can be as-
sessed and judged with clarity and accuracy. Organizations sponsoring research
and corporations that serve to benefit from it want to know that the monetary
investments they make are being well-spent. Researchers are being challenged

A. Kerren et al. (Eds.): Information Visualization, LNCS 4950, pp. 1–18, 2008.


c Springer-Verlag Berlin Heidelberg 2008
2 J.-D. Fekete et al.

to show that their inventions are measurably better than the existing state of
the art.
In broad analytic fields, of which we include InfoVis as a member, the exis-
tence of a ground truth for a problem can greatly facilitate evaluations of value.
For instance, consider the field of computer vision and algorithms for identifying
objects from scenes. It is very easy to create a library of images upon which new
algorithms can be tested. From that, one can measure how well each algorithm
performs and compare results precisely. The TREC [29] and MUC [3] Contests
are examples of this type of evaluation.
Even with a human in the loop, certain fields lend themselves very well to
quantifiable evaluations. Consider systems that support search for particular
documents or facts. Even though different people will perform differently using
a system, researchers can run repeated search trials and measure how often a
person is able to find the target and how long the search took. Averaged over a
large number of human participants, this task yields quantifiable results that can
be measured and communicated quite easily. People or organizations then using
the technology can make well-informed judgments about the value of new tools.
So why is identifying the value of InfoVis so difficult? To help answer that
question, let us turn to what is probably the most accepted definition of InfoVis,
one that comes from Card, Mackinlay, and Shneiderman and that actually is
their definition for “visualization.” They describe visualization as “the use of
computer-supported, interactive visual representations of data to amplify cog-
nition.” [2] The last three words of their definition communicate the ultimate
purpose of visualization, to amplify cognition. So, returning to our discussion
above, is the amplification of cognition something with a ground truth that is
easily and precisely measurable? Clearly it is not and so results the key challenge
in communicating the value of InfoVis.
Further examining the use and purpose of InfoVis helps understand why
communicating its value is so difficult. InfoVis systems are best applied for ex-
ploratory tasks, ones that involve browsing a large information space. Frequently,
the person using the InfoVis system may not have a specific goal or question in
mind. Instead, the person simply may be examining the data to learn more about
it, to make new discoveries, or to gain insight about it. The exploratory process
itself may influence the questions and tasks that arise.
Conversely, one might argue that when a person does have a specific question
to be answered, InfoVis systems are often not the best tools to use. Instead, the
person may formulate his or her question into a query that can be dispatched
to a database or to a search engine that is likely to provide the answer to that
precise question quickly and accurately.
InfoVis systems, on the other hand, appear to be most useful when a person
simply does not know what questions to ask about the data or when the person
wants to ask better, more meaningful questions. InfoVis systems help people to
rapidly narrow in from a large space and find parts of the data to study more
carefully.
Unfortunately, however, activities like exploration, browsing, gaining insight,
and asking better questions are not ones that are easily amenable to establishing
 The Value of Information Visualization 3

and measuring a ground truth. This realization is at the core of all the issues
involved in communicating the value of InfoVis. By its very nature, by its very
purpose, InfoVis presents fundamental challenges for identifying and measuring
value. For instance, how does one measure insight? How does one quantify the
benefits of an InfoVis system used for exploring an information space to gain a
broad understanding of it? For these reasons and others, InfoVis is fundamentally
challenging to evaluate [17].
If we accept that InfoVis may be most valuable as an exploratory aid, then
identifying situations where browsing is useful can help to determine scenarios
most likely to illustrate InfoVis’ value. Lin [11] describes a number of conditions
in which browsing is useful:

– When there is a good underlying structure so that items close to one another
can be inferred to be similar
– When users are unfamiliar with a collection’s contents
– When users have limited understanding of how a system is organized and
prefer a less cognitively loaded method of exploration
– When users have difficulty verbalizing the underlying information need
– When information is easier to recognize than describe

These conditions serve as good criteria for determining situations in which the
value of InfoVis may be most evident.

1.1 Epistemological Issues

Natural sciences are about understanding how nature works. Mathematics is

about truth and systems of verifiable inferences. Human sciences are about un-
derstanding Man in various perspectives. Information Visualization is about de-
veloping insights from collected data, not about understanding a specific domain.
Its object is unique and therefore raises interest and skepticism.
Science has focused on producing results: the goal was essentially the creation
and validation of new theories compatible with collected facts. The importance
of the process — coined as the Method — was raised by the development of
epistemology in the 20th century, in particular with the work of Karl R. Popper
(1902–1994) [18]. It showed that the Method was paramount to the activity of
science.
Karl Popper has explained that a scientific theory cannot be proved true, it
can only be falsified. Therefore, a scientific domain searches for theories that are
as compatible as possible with empirical facts. The good theories are the ones
that have been selected by domain experts among a set of competing theories
in regard of the facts that they should describe. Popper considers science as a
Darwinian selection process among competing theories.
Still, no other scientific domain has argued that generating insights was im-
portant for science. Popper does not explain how a new theory emerges; he only
explains how it is selected when it emerges. Furthermore, Popper has demon-
strated in an article called “The Problem of Induction” that new theories cannot
4 J.-D. Fekete et al.

rationally emerge from empirical data: it is impossible to justify a law by obser-

vation or experiment, since it ’transcends experience’.
Information Visualization is still an inductive method in the sense that it is
meant at generating new insights and ideas that are the seeds of theories, but
it does it by using human perception as a very fast filter: if vision perceives
some pattern, there might be a pattern in the data that reveals a structure.
Drilling down allows the same perception system to confirm or infirm the pattern
very quickly. Therefore, information visualization is meant at “speeding up” the
process of filtering among competing theories regarding collected data by relying
on the speed of the perception system. Therefore, it plays a special role in the
sciences as an insight generating method. It is not only compatible with Popper’s
epistemology system but it furthermore provides a mechanism for accelerating
its otherwise painful Darwinian selection process.

1.2 Moving Forward

It is clear that InfoVis researchers and practitioners face an important challenge
in communicating the value of InfoVis. In the remainder of the paper we explore
this challenge more deeply and we provide several answers to the questions “How
and why is InfoVis useful?”. Since there are several audiences to convince, we
present a number of different sections that are each facets of argumentation to
explain why InfoVis is useful and effective as a mean of understanding complex
datasets and developing insights. The contents of the sections are gathered from
practitioners who already attested that the arguments developed were convinc-
ing. We hope they will be useful to you as well.

2 Cognitive and Perceptual Arguments

Several famous historical figures have argued that the eye was the main sense to
help us understand nature.
The eye. . .
the window of the soul,
is the principal means
by which the central sense
can most completely and
abundantly appreciate
the infinite works of nature.
Leonardo da Vinci (1452 – 1519)
Leonardo’s words are inspirational and they are echoed in everyday expressions
that we are all familiar with such as, “Seeing is believing” and “A picture is worth
a thousand words.” Is there actual support for these sentiments, however?
Let us first consider the case of the phrase, “A picture is worth a thousand
words.” While people may agree or disagree with the sentiments behind that
cliché, specific examples can help support the claim. Consider, for instance, the
 The Value of Information Visualization 5

(a) A thousand words (b) A picture

Fig. 1. “A picture is worth a thousand words”

example shown in Figure 1. Part (a) shows a spreadsheet with data for the 50
states and the District of Columbia in the U.S. Also shown are the percentage
of citizens of each state with a college degree and the per capita income of the
states’ citizens.
Given just the spreadsheet, answering a question such as, “Which state has
the highest average income?” is not too difficult. A simple scan of the income
column likely will produce the correct answer in a few seconds. More complex
questions can be quite challenging given just the data, however. For example,
are the college degree percentage and income correlated? If they are correlated,
are there particular states that are outliers to the correlation? These questions
are much more difficult to answer using only the spreadsheet.
Now, let us turn to a graphical visualization of the data. If we simply draw
the data in a scatterplot as shown in part (b), the questions now become much
easier to answer. Specifically, there does appear to be an overall correlation
between the two attributes and states such as Nevada and Utah are outliers on
the correlation. The simple act of plotting the spreadsheet data in this more
meaningfully communicative form makes these kinds of analytic queries easier
to answer correctly and more rapidly.
Note that the spreadsheet itself is a visual representation of the data that
facilitates queries as well. Consider how difficult the three questions would be if
the data for each state was recorded on a separate piece of paper or webpage.
Or worse yet, what if the data values were read to you and you had to answer
6 J.-D. Fekete et al.

the questions? In this case, already challenging questions become practically

impossible.

2.1 Cognitive Benefits

While the states example illustrates that visualizations can help people under-
stand data better, how do visuals facilitate this process? The core of the benefits
provided by visuals seems to hinge upon their acting as a frame of reference or
as a temporary storage area for human cognitive processes. Visuals augment hu-
man memory to provide a larger working set for thinking and analysis and thus
become external cognition aids. Consider the process of multiplying two long in-
tegers in your head versus then having a pencil and paper available. The visual
representations of the numbers on paper acts as a memory aid while performing
the series of multiplication operations.
Performing a multiplication requires the processing of symbolic data, which
is arguably different than the processing of visual features and shapes. In “Ab-
stract Planning and Perceptual Chunks: Elements of Expertise in Geometry” [8],
Koedinger and Anderson show that experts in geometry effectively use their vi-
sion to infer geometrical properties (parallelism, connectivity, relative positions)
on diagrams; they solve simple problems quickly and accurately, several of mag-
nitude faster than if they had to apply symbolic inference rules.
Larkin and Simon, in their landmark paper “Why a diagram is (sometimes)
worth 10,000 words” [10], discuss how graphical visualization can support more
efficient task performance by allowing substitution of rapid perceptual inferences
for difficult logical inferences and by reducing the search for information required
for task completion. They do note that text can be better than graphics for
certain tasks, however.
Don Norman provides many illustrative examples where visuals can greatly
assist task performance and efficiency [15]. He argues that it is vital to match the
representation used in a visualization to the task it is addressing. The examples
he cites show how judicious visuals can aid information access and computation.
Earlier, we noted how the definition of visualization from Card, Mackinlay
and Shneiderman [2] focused on the use of visuals to “amplify cognition.” Fol-
lowing that definition, the authors listed a number of key ways that the visuals
can amplify cognition:
– Increasing memory and processing resources available
– Reducing search for information
– Enhancing the recognition of patterns
– Enabling perceptual inference operations
– Using perceptual attention mechanisms for monitoring
– Encoding info in a manipulable medium

2.2 Perceptual Support

Most lectures on Information Visualization argue about theoretical properties of
the visual system or more broadly to the perception abilities of humans. Rational
arguments rely on information theory [21] and psychological findings.
 The Value of Information Visualization 7

According to Information Theory, vision is the sense that has the largest
bandwidth: 100 Mb/s [30]. Audition only has around 100 b/s. In that respect,
the visual canal is the best suited to carrying information to the brain.
According to Ware [30], there are two main psychological theories that explain
how vision can be used effectively to perceive features and shapes. At the low
level, Preattentive processing theory [23] explains what visual features can be
effectively processed. At a higher cognitive level, the Gestalt theory [9] describes
some principles used by our brain to understand an image.
Preattentive processing theory explains that some visual features can be per-
ceived very rapidly and accurately by our low-level visual system. For example,
when looking at the group of blue circles in Figure 2, it takes no time and no
effort to see the red circle in the middle. It would be as easy and fast to see that
there is no red circle, or to evaluate the relative quantity of red and blue cir-
cles. Color is one type of feature that can be processed preattentively, but only
for some tasks and within some limits. For example, if there were more than
seven colors used in Figure 2, answering the question could not be done with
preattentive processing and would require sequential scanning, a much longer
process.

Fig. 2. Example of preattentively processed task: finding if there is a red circle among
the blue circles

There is a long list of visual features that can be preattentively processed for
some tasks, including line orientation, line length or width, closure, curvature,
color and many more. Information visualization relies on this theory to choose
the visual encoding used to display data to allow the most interesting visual
queries to be done preattentively.
Gestalt theory explains important principles followed by the visual system
when it tries to understand an image. According to Ware [30], it is based on the
following principles:
8 J.-D. Fekete et al.

Proximity Things that are close together are perceptually grouped together;
Similarity Similar elements tend to be grouped together;
Continuity Visual elements that are smoothly connected or continuous tend
to be grouped;
Symmetry Two symmetrically arranged visual elements are more likely to be
perceived as a whole;
Closure A closed contour tends to be seen as an object;
Relative Size Smaller components of a pattern tend to be perceived as objects
whereas large ones as a background.

Information Visualization experts design visual representations that try to follow

these principles. For example, graph layout algorithms such as [14] designed to
find communities in social networks adhere to the proximity principle by placing
nodes that are connected to a dense group close together and push away nodes
that are connected to another dense group. The Treemap algorithm [5] uses the
closure principle to layout a tree: children of a node are placed inside their parent
node.

3 Success Stories

Information Visualization is much easier to explain using demonstrations than

words. However, to be understood, the data used should be familiar to the au-
dience and interesting. Preparing demonstrations targeted at all the possible
audiences is not possible but there are some datasets that interest most audi-
ences and greatly help make the point. Several striking static examples can be
found in Tufte’s books [24,25,26].
To better explain the value of visualization, demonstrations should start using
a simple question, show that a good representation answers the question at once
and then argue about additional benefits, i.e. questions the users did not knew
they had. From the users perspective, a good representation will confirm what
they already know, let them answer at once the question asked and show them
several insights, leading to the so-called “a-ha” moments when they feel like they
understand the dataset.

3.1 Striking Examples

Static examples used by most InfoVis courses include the map of Napoleon’s
1812 March on Moscow drawn in 1869 by M. Minard (Figure 3) and the map of
London in 1854 overlaid with marks positioning cholera victims that led John
Snow to discovering the origin of the epidemic: infected water extracted with a
water pump at the center of the marks (Figure 4).
In general, good examples show known facts (although sometimes forgotten)
and reveal several unexpected insights at once. Minard’s map can help answer the
question: “What were the casualties of Napoleon’s Russian invasion in 1812?”.
The map reveals at once the magnitude of casualties (from 400,000 to 10,000
 The Value of Information Visualization 9

Fig. 3. Napoleon’s March on Moscow depicted by M. Minard [12]. Width indicates the
number of soldiers. Temperature during the retreat is presented below the map. Image
courtesy of École Nationale des Ponts et Chaussées.

soldiers) as well as the devastating effect of crossing the Berezina river (50,000
soldiers before, 25,000 after). The depiction confirms that Napoleon lost the
invasion (a well known fact) and reveals many other facts, such as the continuous
death rate due to disease and the “scorched earth” tactics of Russia instead of
specific death tolls of large battles.
John Snow’s map was made to answer the question: “What is the origin of the
London cholera epidemics?”. Contrary to the previous map, the answer requires
some thinking. Black rectangles indicate location of deaths. At the center of the
infected zone lies a water pump that John Snow found to be responsible for the
infection. Once again, choosing the right representation was essential for finding
the answer. As a side-effect, the map reveals the magnitude of the epidemic.
Figure 1 answers the question: “Is there a relationship between income and
college degree?” by showing a scatter plot of income by degree for each US state.
The answer is the obvious: yes, but there is much more. There seems to be a
linear correlation between them and some outliers such as Nevada (likely due to
Las Vegas) and Utah do exist, raising new unexpected questions.
Information Visualization couples interaction and visual representation so its
power is better demonstrated interactively. The simplest demonstration suited
to the largest audience is probably the Dynamic HomeFinder5 [32] . It shows the
map of the Washington D.C. area overlaid with all the homes for sale (Figure 5).
Dynamic queries implemented by sliders and check-boxes interactively filter-out
homes that do not fit specific criteria such as cost or number of bedrooms.
Using the interactive controls, it becomes easy to find homes with the desired
attributes or understand how attributes’ constraints should be relaxed to find
some matching homes. Unexpectedly, the Dynamic HomeFinder also reveals the
5
http://www.cs.umd.edu/hcil/pubs/dq-home.zip
10 J.-D. Fekete et al.

Fig. 4. Illustration of John Snow’s deduction that a cholera epidemic was caused by a
bad water pump, circa 1854 [4]. Black rectangles indicate location of deaths.

unpopular neighborhoods around Washington D.C. since they are places where
the homes are cheaper, and the wealthy ones where the houses are more expen-
sive.
Many more examples can be found to demonstrate that InfoVis is effective.
The Map of the Market6 , represented by a squarified treemap, is interesting
for people holding stocks or interested by economic matters. InfoZoom video
on the analysis of Formula 1 results7 is interesting for car racing amateurs.
The video8 comparing two large biological classification trees is interesting to
some biologists. The Baby Name Wizard’s NameVoyager9 is useful for persons
searching a name for their baby to come and a large number of other persons as
witnessed by [31].
With the advent of Social InfoVis through web sites such as Swivel10 or IBM’s
Many-Eyes11, more examples can be found to convince specific audiences. Still,
the process of explaining how InfoVis works remains the same: ask a question
that interests people, show the right representation, let the audience understand
the representation, answer the question and realize how many more unexpected
findings and questions arise.
6
http://www.smartmoney.com/marketmap/
7
http://www.infozoom.com/enu/infozoom/video.htm
8
http://www.fit.fraunhofer.de/~cici/InfoVis2003/StandardForm/Flash/
InfoZoomTrees.html
9
http://babynamewizard.com/namevoyager/
10
http://www.swivel.com
11
http://www.many-eyes.com
 The Value of Information Visualization 11

Fig. 5. Dynamic HomeFinder showing the Washington D.C. area with homes available
for sale and controls to filter them according to several criterion.

3.2 Testimonials
One effective line of argumentation about the value of InfoVis is through re-
porting the success of projects that used InfoVis techniques. These stories exist
but have not been advertised in general scientific publications until recently
[20,16,13]. One problem with trying to report on the success of a project is that
visualization is rarely the only method used to reach the success. For example,
in biological research, the insights gained by an InfoVis system can lead to an
important discovery that is difficult to attribute mainly to the visualization since
it also required months of experimentation to verify the theory formulated from
the insights. In fact, most good human-computer interaction systems allow users
to forget about the system and focus on their task only, which is probably one
reason why success stories are not so common in the InfoVis literature.
Besides these stories that are empirical evidence of the utility of information
visualization, there are strong theoretical arguments to how and why information
visualization works.

4 Information Visualization vs. Automatic Analysis

Several scientific domains are concerned by understanding complex data. Statis-
tics is the oldest, but Data Mining — a subfield of Artificial Intelligence — is
also concerned with automatically understanding the structure of data. There-
fore, InfoVis practitioners frequently need to explain what InfoVis can do that
statistics and data mining cannot.
12 J.-D. Fekete et al.

4.1 Statistics

Statistics is a well grounded field but is composed of several subfields such

as descriptive statistics, classical statistics (also called confirmatory statistics),
Bayesian statistics and Exploratory Data Analysis. Information Visualization
is sometimes considered as a descendant and expansion of Exploratory Data
Analysis.
The differences between these subfields are the methods and the nature of
the answer they seek. All of them start with a problem and gathered data that is
related to the problem to solve. Classical analysis starts by designing a model of
the data, then uses mathematical analysis to test whether the model is refuted
or not by the data to conclude positively or negatively. The main challenge for
classical statistics is to find a model.
Exploratory Data Analysis performs an analysis using visual methods to
acquire insights of what the data looks like, usually to find a model. It uses
visual exploration methods to get the insights.
So why is visualization useful before the modeling? Because, there are cases
when we have no clear idea on the nature of the data and have no model.
To show why visualization can help finding a model, Anscombe in [1] has
designed four datasets that exhibit the same statistical profile but are quite dif-
ferent in shape, as shown in Figure 6. They have the following characteristics12:

– mean of the x values = 9.0

– mean of the y values = 7.5
– equation of the least-squared regression line is: y = 3 + 0.5x
– sums of squared errors (about the mean) = 110.0
– regression sums of squared errors (variance accounted for by x) = 27.5
– residual sums of squared errors (about the regression line) = 13.75
– correlation coefficient = 0.82
– coefficient of determination = 0.67.

Visualization is much more effective at showing the differences between these

datasets than statistics. Although the datasets are synthetic, Anscombe’s Quar-
tet demonstrates that looking at the shape of the data is sometimes better than
relying on statistical characterizations alone.

4.2 Data Mining

More than statistics, the goal of data mining is to automatically find interesting
facts in large datasets. It is thus legitimate to wonder whether data mining, as a
competitor of InfoVis, can overcome and replace the visual capacity of humans.
This question has been addressed by Spence and Garrison in [22] where
they describe a simple plot called the Hertzsprung Russell Diagram (Figure 7a).
It represents the temperature of stars on the X axis and their magnitude on
the Y axis. Asking a person to summarize the diagram produces Figure 7b. It
12
See http://astro.swarthmore.edu/astro121/anscombe.html for details
 The Value of Information Visualization 13

I II III IV
x y x y x y x y
10.0 8.04 10.0 9.14 10.0 7.46 8.0 6.58
8.0 6.95 8.0 8.14 8.0 6.77 8.0 5.76
13.0 7.58 13.0 8.74 13.0 12.74 8.0 7.71
9.0 8.81 9.0 8.77 9.0 7.11 8.0 8.84
11.0 8.33 11.0 9.26 11.0 7.81 8.0 8.47
14.0 9.96 14.0 8.10 14.0 8.84 8.0 7.04
6.0 7.24 6.0 6.13 6.0 6.08 8.0 5.25
4.0 4.26 4.0 3.10 4.0 5.39 19.0 12.50
12.0 10.84 12.0 9.13 12.0 8.15 8.0 5.56
7.0 4.82 7.0 7.26 7.0 6.42 8.0 7.91
5.0 5.68 5.0 4.74 5.0 5.73 8.0 6.89
(a) Four datasets with different values and (b) Dot Plot of the four datasets
the same statistical profile

Fig. 6. Anscombe’s Quartet

(a) Hertzsprung Russell Diagram (b) Standard interpretation of the

Hertzsprung Russell Diagram

Fig. 7. Hertzsprung Russell Diagram and its standard interpretation

turns out that no automatic analysis method has been able to find the same
summarization, due to the noise and artifacts on the data such as the vertical
bands.
Our vision system has evolved with the human specie to help us survive
in a hostile world. We train it to avoid obstacles since we learn how to walk.
It remains remarkably effective at filtering-out noise from useful data, a very
important capability for hunters in deep forests to distinguish the prey moving
14 J.-D. Fekete et al.

behind leaves. We have relied on it and trained it to survive for millennia and it
still surpasses automatic data mining methods to spot interesting patterns. Data
mining still needs to improve to match these pattern matching capabilities.

4.3 Automating or Not?

Is there a competition between confirmatory, automated and exploratory meth-

ods? No, they answer different questions. When a model is known in advance or
expected, using statistics is the right method. When a dataset becomes too large
to be visualized directly, automating some analysis is required. When exploring
a dataset in search of insights, information visualization should be used, possibly
in conjunction with data mining techniques if the dataset is too large.
Furthermore, combining data mining with visualization is the central issue of
Visual Analytics, described by the paper Visual Analytics: Definition, Process,
and Challenges in this book [6].

5 An Economical Model of Value

One important question is how to assess the value of visualization, ranging from
the evaluation of one specific use-case to the discipline in general. If we know
how to do this, then this might lead to an assessment of the current status as
well as the identification of success factors. An attempt was given by Van Wijk
[27] and is summarized here. After a short overview of his model, we discuss how
this model can be applied for InfoVis.
Visualization can be considered as a technology, a collection of methods,
techniques, and tools developed and applied to satisfy a need. Hence, standard
technological measures apply: Visualization has to be effective and efficient. To
measure these, an economic point of view is adopted. Instead of trying to un-
derstand why visualization works (see previous sections), here visualization is
considered from the outside, and an attempt is made to measure its profit. The
profit of visualization is defined as the difference between the value of the increase
in knowledge and the costs made to obtain this insight. Obviously, in practice
these are hard to quantify, but it is illuminating to attempt so. A schematic
model is considered: One visualization method V is used by n users to visual-
ize a data set m times each, where each session takes k explorative steps. The
value of an increase in knowledge (or insight) has to be judged by the user.
Users can be satisfied intrinsically by new knowledge, as an enrichment of their
understanding of the world. A more pragmatic and operational point of view
is to consider if the new knowledge influences decisions, leads to actions, and,
hopefully, improves the quality of these. The overall gain now is nm(W (∆K)),
where W (∆K)) represents the value of the increase in knowledge.
Concerning the costs for the use of (a specific) visualization V , these can
be split into various factors. Initial research and development costs Ci have to
be made; a user has to make initial costs Cu , because he has to spend time to
select and acquire V , and understand how to use it; per session initial costs Cs
 The Value of Information Visualization 15

have to be made, such as conversion of the data; and finally during a session a
user makes costs Ce , because he has to spend time to watch and understand the
visualization, and interactively explore the data set. The overall profit now is

F = nm(W (∆K) − Cs − kCe ) − Ci − nCu .

In other words, this leads to the obvious insight that a great visualization method
is used by many people, who use it routinely to obtain highly valuable knowledge,
while having to spend little time and money on hardware, software, and effort.
And also, no alternatives that are more cost-effective should be available.
In the original paper a number of examples of more or less successful visual-
ization methods are given, viewed in terms of this model. One InfoVis application
was considered: SequoiaView, a tool to visualize the contents of a hard disk, us-
ing cushion treemaps [28]. The popularity of this tool can be explained from
the concrete and useful insights obtained, as well as the low costs in all respects
associated with its application.
When we consider InfoVis in general, we can also come to positive conclu-
sions for almost all parameters, and hence predict a bright future for our field.
The number of potential users is very large. Data in the form of tables, hier-
archies, and networks is ubiquitous, as well as the need to get insight in these.
This holds for professional applications, but also for private use at home. Many
people have a need to get an overview of their email, financial transfers, media
collections, and to search in external data bases, for instance to find a house,
vacation destination, or another product that meets their needs. Methods and
techniques from InfoVis, in the form of separate tools or integrated in custom
applications, can be highly effective here to provide such overviews. Also, many
of these activities will be repeated regularly, hence both n and m are high. The
growing field of Casual InfoVis [19] further illustrates how InfoVis techniques
are becoming more common in people’s everyday lives.
The costs Ce that have to be made to understand visualizations depend on
the prior experience of the users as well as the complexity of the imagery shown.
On the positive side, the use of graphics to show data is highly familiar, and
bar-charts, pie-charts, and other forms of business graphics are ubiquitous. On
the other hand, one should not overestimate familiarity. The scatterplot seems
to be at the boundary: Considered as trivial in the InfoVis community, but too
hard to understand (if the horizontal axis does not represent time) by a lay-
audience, according to Matthew Ericson, deputy graphics director of the New
York Times in his keynote presentation at IEEE InfoVis 2007. Visual literacy is
an area where more work can be done, but on the other hand, InfoVis does have
a strong edge compared to non-visual methods here. And, there are examples of
areas where complex visual coding has been a great success, with the invention
of the script as prime example.
The costs Cs per session and Cu per user can be reduced by tight integration
with applications. The average user will not be interested in producing visual-
izations, her focus will be on solving her own problem, where visualization is one
of the means to this end. Separate InfoVis tools are useful for specialists, which
16 J.-D. Fekete et al.

use them on a day-to-day basis. For many other users, integration within their
favourite tool is much more effective. An example of an environment that offers
such a tight integration is the ubiquitous spreadsheet, where storage, manipu-
lation, and presentation of data are offered; or the graphs and maps shown on
many web sites (and newspapers!) to show data. From an InfoVis point of view,
the presentations offered here can often be improved, and also, the interaction
provided is often limited. Nevertheless, all these examples acknowledge the value
of visualization for many applications.
The initial costs Ci for new InfoVis methods and techniques roughly fall into
two categories: Research and Development. Research costs can be high, because
it is often hard to improve on the state of the art, and because many experiments
(ranging from the development of prototypes to user experiments) are needed.
On the other hand, when problems are addressed with many potential usages,
these costs are still quite limited. Development costs can also be high. It takes
time and effort to produce software that is stable and useful under all conditions,
and that is tightly integrated with its context, but here also one has to take
advantage of the large potential market. Development and availability of suitable
middleware, for instance as libraries or plug-ins that can easily customized for
the problem at hand is an obvious route here.
One intriguing aspect here is how much customization is needed to solve the
problem concerned. On one hand, in many applications one of the standard data
types of InfoVis is central (table, tree, graph, text), and when the number of
items is not too high, the problem is not too hard to solve. On the other hand,
for large numbers of items one typically has to exploit all a priori knowledge of
the data set and tune the visualization accordingly; also, for applications such
as software visualizations all these data types pop up simultaneously, which also
strongly increases the complexity of the problem. So, for the time being, research
and innovation will be needed to come up with solutions for such problems as
well.
In conclusion, graphics has been adopted already on a large scale to com-
municate and present abstract data, which shows that its value has been ac-
knowledged, and we expect that due to the increase in size and complexity of
data available, the need for more powerful and effective information visualization
methods and techniques will only grow.

6 Conclusion
In this paper we have described the challenges in identifying and communicat-
ing the value of InfoVis. We have cited and posed a number of answers to the
questions, “How and why is InfoVis useful?” Hopefully, the examples shown
in the paper provide convincing arguments about InfoVis’ value as an analytic
tool. Ultimately, however, we believe that it is up to the community of InfoVis
researchers and practitioners to create techniques and systems that clearly il-
lustrate the value of the field. When someone has an InfoVis system that they
use in meaningful and important ways, this person likely will not need to be
convinced of the value of InfoVis.
 The Value of Information Visualization 17

References

1. Anscombe, F.: Graphs in statistical analysis. American Statistician 27(1), 17–21

(1973)
2. Card, S.K., Mackinlay, J., Shneiderman, B. (eds.): Readings in Information Visu-
alization – Using Vision to Think. Morgan Kaufmann, San Francisco (1998)
3. Chincor, N., Lewis, D., Hirschman, L.: Evaluating message understanding sys-
tems: An analysis of the third message understanding conference (MUC-3). Com-
putational Linguistics 19(3), 409–449 (1993)
4. Gilbert, E.W.: Pioneer Maps of Health and Disease in England. Geographical
Journal 124, 172–183 (1958)
5. Johnson, B., Shneiderman, B.: Tree-maps: a space-filling approach to the visu-
alization of hierarchical information structures. In: VIS ’91: Proceedings of the
2nd conference on Visualization ’91, Los Alamitos, CA, USA, pp. 284–291. IEEE
Computer Society Press, Los Alamitos (1991)
6. Keim, D., Andrienko, G., Fekete, J.-D., Görg, C., Kohlhammer, J., Melançon,
G.: Visual Analytics: Definition, Process, and Challenges. In: Kerren, A., Stasko,
J.T., Fekete, J.-D., North, C.J. (eds.) Information Visualization. LNCS, vol. 4950,
Springer, Heidelberg (2008)
7. Kerren, A., Stasko, J.T., Fekete, J.-D., North, C.J. (eds.): Information Visualiza-
tion. LNCS, vol. 4950. Springer, Heidelberg (2008)
8. Koedinger, K.R., Anderson, J.R.: Abstract planning and perceptual chunks: Ele-
ments of expertise in geometry. Cognitive Science 14(4), 511–550 (1990)
9. Koffa, K.: Principles of Gestalt Psychology. Routledge & Kegan Paul Ltd, London
(1935)
10. Larkin, J.H., Simon, H.A.: Why a diagram is (sometimes) worth 10,000 words.
Cognitive Science 11, 65–100 (1987)
11. Lin, X.: Map displays for information retrieval. Journal of the American Society
for Information Science 48(1), 40–54 (1997)
12. Marey, E.: La Méthode Graphique. Paris (1885)
13. McLachlan, P., Munzner, T., Koutsofios, E., North, S.: Liverac: Interactive visual
exploration of system management time-series data. In: SIGCHI Conference on
Human Factors in Computing Systems (CHI 2008), ACM Press, New York (2008)
14. Noack, A.: Energy-based clustering of graphs with nonuniform degrees. In: Healy,
P., Nikolov, N.S. (eds.) GD 2005. LNCS, vol. 3843, pp. 309–320. Springer, Hei-
delberg (2006)
15. Norman, D.A.: Things That Make Us Smart: Defending Human Attributes in
the Age of the Machine. Addison-Wesley Longman Publishing Co., Inc., Boston
(1993)
16. Perer, A., Shneiderman, B.: Integrating statistics and visualization: Case studies
of gaining clarity during exploratory data analysis. In: SIGCHI Conference on
Human Factors in Computing Systems (CHI 2008), ACM Press, New York (2008)
17. Plaisant, C.: The challenge of information visualization evaluation. In: AVI ’04:
Proceedings of the working conference on Advanced visual interfaces, pp. 109–116.
ACM Press, New York (2004)
18. Popper, K.R.: The Logic of Scientific Discovery. Basic Books, New York (1959)
19. Pousman, Z., Stasko, J., Mateas, M.: Casual information visualization: Depic-
tions of data in everyday life. IEEE Transactions on Visualization and Computer
Graphics 13(6), 1145–1152 (2007)
18 J.-D. Fekete et al.

20. Saraiya, P., North, C., Lam, V., Duca, K.: An insight-based longitudinal study
of visual analytics. IEEE Transactions on Visualization and Computer Graph-
ics 12(6), 1511–1522 (2006)
21. Shannon, C.E., Weaver, W.: A Mathematical Theory of Communication. Univer-
sity of Illinois Press, Champaign (1963)
22. Spence, I., Garrison, R.F.: A remarkable scatterplot. The American Statistician,
12–19 (1993)
23. Triesman, A.: Preattentive processing in vision. Computer Vision, Graphics, and
Image Processing 31(2), 156–177 (1985)
24. Tufte, E.R.: The Visual Display of Quantitative Information. Graphics Press,
Cheshire (1983)
25. Tufte, E.R.: Envisioning Information. Graphics Press, Cheshire (1990)
26. Tufte, E.R.: Visual Explanations: Images and Quantities, Evidence and Narrative.
Graphics Press, Cheshire (1997)
27. van Wijk, J.J.: The value of visualization. In: Proceedings IEEE Visualization
2005, pp. 79–86 (2005)
28. van Wijk, J.J., van de Wetering, H.: Cushion treemaps. In: Proceedings 1999
IEEE Symposium on Information Visualization (InfoVis’99), pp. 73–78. IEEE
Computer Society Press, Los Alamitos (1999)
29. Voorhees, E., Harman, D.: Overview of the sixth Text Retrieval Conference. In-
formation Processing and Management 36(1), 3–35 (2000)
30. Ware, C.: Information Visualization: Perception for Design. Morgan Kaufmann
Publishers Inc., San Francisco (2004)
31. Wattenberg, M., Kriss, J.: Designing for social data analysis. IEEE Transactions
on Visualization and Computer Graphics 12(4), 549–557 (2006)
32. Williamson, C., Shneiderman, B.: The dynamic homefinder: evaluating dynamic
queries in a real-estate information exploration system. In: SIGIR ’92: Proceed-
ings of the 15th annual international ACM SIGIR conference on Research and
development in information retrieval, pp. 338–346. ACM Press, New York (1992)
 Evaluating Information Visualizations

Sheelagh Carpendale

Department of Computer Science, University of Calgary,

2500 University Dr. NW, Calgary, AB, Canada T2N 1N4
[email protected]

1 Introduction
Information visualization research is becoming more established, and as a result, it is
becoming increasingly important that research in this field is validated. With the gen-
eral increase in information visualization research there has also been an increase,
albeit disproportionately small, in the amount of empirical work directly focused on
information visualization. The purpose of this paper is to increase awareness of em-
pirical research in general, of its relationship to information visualization in particu-
lar; to emphasize its importance; and to encourage thoughtful application of a greater
variety of evaluative research methodologies in information visualization.
One reason that it may be important to discuss the evaluation of information visu-
alization, in general, is that it has been suggested that current evaluations are not con-
vincing enough to encourage widespread adoption of information visualization tools
[57]. Reasons given include that information visualizations are often evaluated using
small datasets, with university student participants, and using simple tasks. To en-
courage interest by potential adopters, information visualizations need to be tested
with real users, real tasks, and also with large and complex datasets. For instance, it is
not sufficient to know that an information visualization is usable with 100 data items
if 20,000 is more likely to be the real-world case. Running evaluations with full data
sets, domain specific tasks, and domain experts as participants will help develop
much more concrete and realistic evidence of the effectiveness of a given information
visualization. However, choosing such a realistic setting will make it difficult to get a
large enough participant sample, to control for extraneous variables, or to get precise
measurements. This makes it difficult to make definite statements or generalize from
the results. Rather than looking to a single methodology to provide an answer, it will
probably will take a variety of evaluative methodologies that together may start to
approach the kind of answers sought.
The paper is organized as follows. Section 2 discusses the challenges in evaluating
information visualizations. Section 3 outlines different types of evaluations and dis-
cusses the advantages and disadvantages of different empirical methodologies and the
trade-offs among them. Section 4 focuses on empirical laboratory experiments and the
generation of quantitative results. Section 5 discusses qualitative approaches and the
different kinds of advantages offered by pursuing this type of empirical research.
Section 6 concludes the paper.

A. Kerren et al. (Eds.): Information Visualization, LNCS 4950, pp. 19 – 45, 2008.
© Springer-Verlag Berlin Heidelberg 2008
20 S. Carpendale

2 Challenges in Evaluating Information Visualizations

Much has already been written about the challenges facing empirical research in in-
formation visualization [2, 12, 53, 57]. Many of these challenges are common to all
empirical research. For example, in all empirical research it is difficult to pick the
right focus and to ask the right questions. Given interesting questions, it is difficult to
choose the right methodology and to be sufficiently rigorous in procedure and data
collection. Given all of the above, appropriate data analysis is still difficult and per-
haps most difficult of all is relating a new set of results to previous research and to
existing theory. However, information visualization research is not alone in these
difficulties; the many other research fields that also face these challenges can offer a
wealth of pertinent experience and advice.
In particular, empirical research in information visualization relates to human com-
puter interaction (HCI) empirical research, perceptual psychology empirical research,
and cognitive reasoning empirical research. The relationship to empirical research in
HCI is evident in that many of the tasks of interest are interface interaction tasks, such
as zooming, filtering, and accessing data details [66]. The aspects of these interactive
tasks that provide access to the visual representation and its underlying dataset often
relate to the usability of a system. Other challenges that are shared with HCI empirical
research include the difficulty of obtaining an appropriate sample of participants. If
the visualization is intended for domain experts it can be hard to obtain their time.
Also, when evaluating complex visualization software, it may not be clear whether
the results are due to a particular underlying technique or the overall system solution.
If an existing piece of software is to be used as a benchmark against which to com-
pare an interactive visualization technique, it is likely that participants may be much
more familiar with the existing software and that this may skew the results. This prob-
lem becomes more extreme the more novel a given visualization technique is. Re-
search prototypes are not normally as smooth to operate as well established software,
creating further possibilities for affecting study results and leading to controversy
about testing research prototypes against the best competitive solution. Greenberg and
Buxton [27] discuss this problem in terms of interaction sketches, encouraging cau-
tion when thinking about conducting usability testing on new ideas and new interface
sketches in order to avoid interfering with the development of new ideas. In addition,
research software does not often reach a stage in which it can support a full set of
possible tasks or be fully deployable in real-world scenarios [57].
In addition to usability questions, perceptual and comprehensibility questions such
as those considered in perceptual psychology are important in assessing the appropri-
ateness of a representational encoding and the readability of visuals [30, 79]. Also, in
information visualization, there are a great variety of cognitive reasoning tasks that
vary with data type and character, from low-level detailed tasks to complex high-level
tasks. Some of these tasks are not clearly defined, particularly those that hold some
aspect of gaining new insight into the data, and may be more challenging to test em-
pirically. Examples of low-level detailed tasks include such tasks as compare, con-
trast, associate, distinguish, rank, cluster, correlate, or categorize [57]; higher-level
and more complex cognitive tasks include developing an understanding of data trends,
uncertainties, causal relationships, predicting the future, or learning a domain [1]. Many
important tasks can require weeks or months to complete. The success of information
 Evaluating Information Visualizations 21

visualization is often an interplay between an expert’s meta-knowledge and knowl-

edge of other sources as well as information from the visualization in use.
While all of the above are important, a question that lies at the heart of the success
of a given information visualization is whether it sheds light on or promotes insight
into the data [55, 63]. Often, the information processing and analysis tasks are com-
plex and ill-defined, such as discovering the unexpected, and are often long term or
on-going. What exactly insight is probably varies from person to person and instance
to instance; thus it is hard to define, and consequently hard to measure. Plaisant [57]
describes this challenge as “answering questions you didn’t know you had.” While it
is possible to ask participants what they have learned about a dataset after use of an
information visualization tool, it strongly depends on the participants’ motivation,
their previous knowledge about the domain, and their interest in the dataset [55, 63].
Development of insight is difficult to measure because in a realistic work setting it is
not always possible to trace whether a successful discovery was made through the use
of an information visualization since many factors might have played a role in the
discovery. Insight is also temporally elusive in that insight triggered by a given visu-
alization may occur hours, days, or even weeks after the actual interaction with the
visualization. In addition, these information processing tasks frequently involve
teamwork and include social factors, political considerations and external pressures
such as in emergency response scenarios. However, there are other fields of research
that are also grappling with doing empirical research in complex situations. In particu-
lar, ecologists are faced with conducting research towards increasing our understand-
ing of complex adaptive systems. Considering the defining factors of complex adap-
tive systems may help to shed some light on the difficulties facing empirical research
in information visualization. These factors include non-linearity, holoarchy and inter-
nal causality [37, 49]. When a system is non-linear, the system behaviour comes only
from the whole system. That is, the system can not be understood by decomposing it
into its component parts which are then reunited in some definitive way. When a
system is holoarchical it is composed of holons which are both a whole and a part.
That is, the system is mutually inter-nested. While it is not yet common to discuss
information analysis processes in terms of mutual nesting, in practice many informa-
tion analysis processes are mutually nested. For instance, consider the processes of
search and verification: when in the midst of searching, one may well stop to verify a
find; and during verification of a set of results, one may well need to revert to search
again. Internal causality indicates that the system is self-organizing and can be charac-
terized by goals, positive and negative feedback, emergent properties and surprise.
Considering that it is likely that a team of information workers using a suite of visu-
alization and other software tools is some type of complex adaptive system suggests
that more holistic approaches to evaluation may be needed.
Already from this brief overview, one can see that useful research advice on the
evaluation of information visualization can be gathered from perceptual psychology,
cognitive reasoning research, as well as human computer interaction research. Many,
but not enough, information visualization researchers are already actively engaged in
this pursuit. The purpose of this paper is to applaud them, to encourage more such
research, and to suggest that the research community to be more welcoming of a
greater variety of these types of research results.
22 S. Carpendale

3 Choosing an Evaluation Approach

A recent call for papers from the information visualization workshop, Beyond Time
and Errors (BELIV06) held at Advanced Visual Interfaces 2006, stated that “Con-
trolled experiments remain the workhorse of evaluation but there is a growing sense
that information visualization systems need new methods of evaluation, from longitu-
dinal field studies, insight based evaluation and other metrics adapted to the percep-
tual aspects of visualization as well as the exploratory nature of discovery” [7]. The
purpose of this section is to encourage people to consider more broadly what might be
the most appropriate research methods for their purposes. To further this purpose a
variety of types of empirical research that can be usefully conducted are briefly out-
lined and these differing types are discussed in terms of their strengths and weak-
nesses. This discussion draws heavily from McGrath’s paper Methodology Matters
[50] that was initially written for social scientists. However, while social scientists
work towards understanding humans as individuals, groups, societies and cultures, in
information visualization – similarly to HCI – we are looking to learn about how
information visualizations do or do not support people in their information tasks
and/or how people conduct their information related tasks so that visualization can be
better designed to support them. To gain this understanding we sometimes study peo-
ple using information visualization software and sometimes it may be important to
study people independently of that software, to better understand the processes we are
trying to support.
There are some commonalities to all studies. They all must start with some ques-
tion or questions that will benefit from further study. Also, they all must relate their
research questions to the realm of existing ideas, theories and findings. These ideas,
theories, and concepts are needed to relate the new study to existing research. For
example, the results from a new study might be in contrast to existing ideas, in
agreement with existing ideas, or offer an extension of or variation to existing ideas.
A study must also have a method. This is what this section is about – possible types of
empirical methodologies.
All methods offer both advantages and disadvantages. One important part of em-
pirical research is choosing the most appropriate research methods for your content,
your ideas, and your situation. The fact that methods both provide and limit evidence
suggests that making use of a wide variety of methodologies will, in time, strengthen
our understandings. Thus, both conducting a greater variety of studies and encourag-
ing this by publishing research that employs a greater variety of methodologies will
help to develop a better understanding of the value of information visualization and its
potential in our communities.
When conducting a study there are three particularly desirable factors: generaliza-
bility, precision, and realism [50]. Ideally, one would like all of these factors in one’s
results. However, existing methodologies do not support the actualization of all three
simultaneously. Each methodology favours one or two of these factors, often at the
expense of the others; therefore the choice of a methodology for a particular goal is
important. To define these terms (as used in McGrath [50]):
• Generalizability: a result is generalizable to the extent to which it can apply to
other people (than those directly in the study) and perhaps even extend to other
situations.
 Evaluating Information Visualizations 23

• Precision: a result is precise to the degree to which one can be definite about the
measurements that were taken and about the control of the factors that were not
intended to be studied.
• Realism: a result is considered realistic to the extent to which the context in
which it was studied is like the context in which it will be used.
Figure 1 (adapted and simplified from McGrath [50]) shows the span of common
methodologies currently in practice in the social sciences. They are positioned around
the circle according to the labels: most precision, most generalizability and most real-
ism. The closer a methodology is placed to a particular label, the more that label ap-
plies to that methodology. Next, these methodologies are briefly described. For fuller
descriptions see McGrath 1995.

Fig. 1. Types of methodologies organized to show relationships to precision, generalizability

and realism. (adapted, simplified from McGrath 1995)

Field Study: A field study is typically conducted in the actual situation, and the ob-
server tries as much as possible to be unobtrusive. That is, the ideal is that the pres-
ence of the observer does not affect what is being observed. While one can put con-
siderable effort into minimizing the impact of the presence of an observer, this is not
completely possible [50]. Examples of this type of research include ethnographic
work in cultural anthropology, field studies in sociology, and case studies in industry.
In this type of study the realism is high but the results are not particularly precise and
likely not particularly generalizable. These studies typically generate a focused but
rich description of the situation being studied.
Field Experiment: A field experiment is usually also conducted in a realistic setting;
however, an experimenter trades some degree of unobtrusiveness in order to obtain
more precision in observations. For instance, the experimenter may ask the partici-
pants to perform a specific task while the experimenter is present. While realism is
24 S. Carpendale

still high, it has been reduced slightly by experimental manipulation. However, the
necessity of long observations may be shortened and results may be more readily
interpretable and specific questions are more likely to be answered.
Laboratory Experiment: In a laboratory experiment the experimenters fully design
the study. They establish what the setting will be, how the study will be conducted,
what tasks the participants will do, and thus plan the whole study procedure. Then the
experimenter gets people to participate as fully as possible following the rules of the
procedure within the set situation. Carefully done, this can provide for considerable
precision. In addition, non-realistic behaviour that provides the experimenter more
information can be requested such as a ‘think aloud’ protocol [43]. Behaviour can be
measured, partly because it is reasonably well known when and where the behaviour
of interest may happen. However, realism is largely lost and the degree to which the
experimenter introduces aspects of realism will likely reduce the possible precision.
Experimental Simulation: With an experimental simulation the experimenter tries to
keep as much of the precision as possible while introducing some realism via simula-
tion. There are examples where this approach is essential such as studying driving
while using a cell phone or under some substance’s influence by using a driving simu-
lator. Use of simulation can avoid risky or un-ethical situations. Similarly although
less dramatically, non-existent computer programs can be studied using the ‘Wizard
of Oz’ approach in which a hidden experimenter simulates a computer program. This
type of study can provide us with considerable information while reducing the dan-
gers and costs of a more realistic experiment.
Judgment Study: In a judgment study the purpose is to gather a person’s response to
a set of stimuli in a situation where the setting is made irrelevant. Much attention is
paid to creating ‘neutral conditions’. Ideally, the environment would not affect the
result. Perceptual studies often use this approach. Examples of this type of research
include the series of studies that examine what types of surface textures best support
the perception of 3D shape (e.g. [34, 38]), and the earlier related work about the per-
ception of shape from shading [39]. However, in assessing information visualizations
this idea of setting a study in neutral conditions must be considered carefully, as wit-
nessed by Reilly and Inkpen’s [62] study which showed that the necessity for an in-
teractive technique developed to support a person’s mental model during transition
from viewing one map to another (subway map to surface map) was dependent on the
distractions in the setting. This transition technique relates to ideas of morphing and
distortion in that aspects of the map remain visible while shifting. These studies in a
more neutral experiment setting showed little benefit, while the same tasks in a noisy,
distracting setting showed considerable benefit.
Sample Survey: In a sample survey the experimenter is interested in discovering
relationships between a set of variables in a given population. Examples of these
types of questions include: of those people who discover web information visualiza-
tion tools how many return frequently and are their activities social or work related?
Of those people who have information visualization software available at work what
is the frequency of use? Considering the increased examples of information visualiza-
tion results and software on the web, is the general population’s awareness of and/or
use of information visualization increasing? In these types of studies proper sampling
 Evaluating Information Visualizations 25

of the population can lead to considerable generalizability. However, while choosing

the population carefully is extremely important, often it is difficult to control. For
example in a web-based survey, all returned answers are from those types of people
who are willing to take the time, fill out the questionnaire, etc. This is a type of bias
and thus reduces generalizability. Also, responses are hard to calibrate. For instance, a
particular paper reviewer may never give high scores and the meta-reviewer may
know this and calibrate accordingly or may not know this. Despite these difficulties,
much useful information can be gathered this way. We as a community must simply
be aware of the caveats involved.
Formal Theory: Formal theory is not a separate experimental methodology but an
important aspect of all empirical research that can easily be overlooked. As such, it
does not involve the gathering of new empirical evidence and as a result is low in
both precision and realism. Here, existing empirical evidence is examined to con-
sider the theoretical implications. For example, the results of several studies can be
considered as a whole to provide a higher-level or meta-understanding or the results
can be considered in light of existing theories to extend, adjust or refute them. Cur-
rently this type of research is particularly difficult to publish in that there are no
new information visualizations and no new empirical results. Instead, the contribu-
tion moves towards the development of theories about the use of and practicality of
information visualizations.
Computer Simulation: It is also possible to develop a computer simulation that has
been designed as logically complete. This method is used in battle simulation, re-
search and rescue simulation, etc. This type of strategy can be used to assess some
visualizations. For instance, a visualization of landscape vegetation that includes
models of plant growth and models of fire starts and spread can be set to simulate
passage of several hundred years. If the resulting vegetation patterns are comparable
to existing satellite imagery this provides considerable support for the usefulness of
models [22]. Since this type of research strategy does not involve participants, discus-
sion of generalizability over populations is not applicable. Also, since the models are
by definition incomplete, notions of precision in measurement are often replaced with
stochastic results. On the other hand it does provide a method of validation and offers
a parallel with which we can study realistic situations, such as explosions, turbulence
in wind tunnels, etc.

4 Focus on Quantitative Evaluation

Quantitative evaluations, most well known as laboratory experiments or studies, are
those methodologies in which precision is relatively high and in which some declara-
tion can be made about the possible generalization to a larger population. These dec-
larations can include information about the characterization of this larger population
and how likely it is that the generalization will hold. These types of experiments or
studies are part of the traditional empirical scientific experimental approach and have
evolved and been refined through the centuries of scientific research. Science does
and has depended on these methods. Slowly, through careful and rigorous application
of the experimental process, knowledge has been built up, usually one piece at a time.
26 S. Carpendale

The experiments or studies involve a rigorous process of hypothesis development,

identification and control of the independent variables, observation and measurement
of the dependent variables, and application of statistics which enable the declaration
of the confidence with which the results can be taken. In these formal studies or con-
trolled evaluations, the experimenter controls the environment or setting, manipulates
chosen factor(s) or variable(s) – the independent variable(s) - in order to be able to
measure and observe the affect this manipulation has on one or more other factors –
the dependent variable(s). Ideally no other factors change during the experiment.
Once the changes to the dependent variables have been measured, statistical methods
can be applied to understand the relative importance of the results. Done with suffi-
cient thoroughness, this process can arrive at facts about which we can be relatively
certain. The application of this scientific process will to try to reduce the overall com-
plexity by fine tuning particular questions or hypotheses, using these hypotheses to
allow one to cull some of the complexity by trying to eliminate as many of the extra-
neous variables as possible. Traditionally experiments of this type are used to shed
light on cause and effect relationships; that is, to discover whether changes in some
factor result in changes to another factor.
This idea that we can observe simpler, more manageable subsets of the full com-
plex process is appealing, and it is clear from centuries of experiments that much can
be learnt in this manner.

4.1 Quantitative Methodology

Since quantitative empirical evaluations have evolved over the centuries the method-
ology has become relatively established (Figure 2). This brief overview is included
for completeness; the interested reader should refer to the many good books on this
subject [15, 17, 33]. This methodology includes:
• Hypothesis Development: Much of the success of a study depends on asking an
interesting and relevant question. This question should ideally be of interest to
the broader research community, and hopefully answering it will lead to a deeper
or new understanding of open research questions. Commonly the importance of
the study findings results from a well thought through hypothesis, and formulat-
ing this question precisely will help the development of the study.
• Identification of the Independent Variables: The independent variables are
the factors to be studied which may (or may not) affect the hypothesis. Ideally
the number of independent variables is kept low to provide more clarity and pre-
cision in the results.
• Control of the Independent Variables: In designing the experiment the ex-
perimenter decides the manner in which the independent variables will be
changed.
• Elimination of Complexity: In order to be clear that it is actually the change in
the independent variable that caused the study’s result, it is often the case that
other factors in the environment need to be controlled.
• Measurement of the Dependent Variables: Observations and measurements
are focused on the dependent variables as they change or do not change in
 Evaluating Information Visualizations 27

Fig. 2. A simple schematic of the traditional experimental process.

response to the manipulation of the independent variable. The aspects to be

measured are often called metrics. Common metrics include: speed, accuracy,
error rate, satisfaction, etc.
• Application of Statistics: The results collected can then be analysed through the
application of appropriate statistics. It is important to remember that statistics
tell us how sure we can be that these results could (or could not) have happened
by chance. This gives a result with a relative degree of certainty. There are many
good references such as Huck [33].
These steps sound deceptively simple but doing them well requires careful and rigor-
ous work. For instance, it is important that the study participants are valued, that they
are not over-stressed, and that they are given appropriate breaks, etc. Also, exactly
what they are being asked to do must be clear and consistent across all participants in
your study. Since small inconsistencies such as changes in the order of the instruc-
tions can affect the results, the common recommendation is that one scripts the expla-
nations. Perhaps most importantly, to eliminate surprises and work out the details, it is
best to pilot – run through the experiment in full – repeatedly.

4.2 Quantitative Challenges

Even though these types of experiments have been long and effectively used across all
branches of science, there remain many challenges to conducting a useful study. We
mention different types of commonly-discussed errors and validity concerns and re-
late these to the McGrath’s discussion as outlined in Section 3. In this discussion we
will use a simple, abstract example of an experiment that looks at the effect of two
visualization techniques, VisA and VisB, on performance in search. There are several
widely discussed issues that can interfere with the validity of a study.
Conclusion Validity: Is there a relationship? This concept asks whether within the
study there is a relationship between the independent and the dependent variables.
Important factors in conclusion validity are finding a relationship when one does not
exist (type I error) and not finding a relationship when one does exist (type II error).
28 S. Carpendale

Table 1. Type I and Type II Errors

Reality
H0 H0
TRUE FALSE
H0
ok Type II
Experimental TRUE
decision H0
Type I ok
FALSE

Type I and Type II Errors: If one is interested in which visualization technique

VisA or VisB helps people conduct a particular task faster one might formulate a null
hypothesis (H0 ) – there is no difference in speed of search between VisA and VisB.
The possible type I, false negative, and type II, false positive, errors are specified in
Table 1. The columns represent whether the null hypothesis is true or false in reality
and the rows show the decision made based on the results of the experiment. Ideally
the results of the experiment reflect reality and that if the hypothesis is false (or true)
in reality it will show as false (or true) in the experiment. However, it is possible that
the hypothesis is true in reality – VisA does support faster search than VisB – but that
this fails to be revealed by the experiment. This is a type II error. A type I error occurs
if the null hypothesis is true in reality (there is no difference) and one concludes that
there is a difference. Type I errors are considered more serious. That is, it is consid-
ered worse to claim that VisA improves search when it does not, than to say there was
no measurable difference.
Internal Validity: Is the relationship causal? This concept is important when an
experiment is intended to reveal something about causal relationships. Thus, internal
validity will be important in our simple example because the study is looking at what
effect VisA and VisB have on search. The key issue here is whether the results of
one’s study can properly be attributed to what happened within the experiment. That
is, that no other factors influenced or contributed to the results seen in the study. An-
other way of asking this question is: are there possible alternate causes for the results
seen in the study?
Construct Validity: Can we generalize to the constructs (ideas) the study is based
on? This concept considers whether the experiment has been designed and run in a
manner that answers the intended questions. This is an issue about whether the right
factors are being measured or whether the factors the experimenter intends to measure
are actually those being measured. For instance, does the experiment measure the
difference due to the techniques VisA and VisB or the difference in participant’s
familiarity with VisA and VisB. For instance, if the construct is that a person will
have higher satisfaction when using VisB, does measuring error rates and completion
times provide answers for this construct? An important part of this concept of con-
struct validity is measurement validity. Measurement validity is concerned with
questions such as is one measuring what one intends to measure and is the method of
measurement reliable and consistent. That is, will the same measurement process
provide the same results when repeated?
 Evaluating Information Visualizations 29

External Validity: Can we generalize the study results to other people/places/

times? External validity is concerned with the extent to which the result of a study
can be generalized. If a study has good internal and construct validity the results apply
to the setting, time, and participants being studied. The extent to which the results
apply beyond the immediate setting, time and participants depends, for participants,
on the participant sample and the population from which it was drawn. For instance,
in practice it is common to draw participants from the geographic region in which the
study is run. Does this mean that the results only apply to people from that region? If
culture has a possible impact on the results, they may not generalize. If one addresses
the need to include cultural variation by recruiting participants from different cultures
from a university’s foreign students, one might have at least partially addressed the
need to run the study across cultural variations but now have limited the demographic
to university students which may introduce its own skew. Understanding the popula-
tion to which one would like to be able to generalize the study results and successfully
obtaining an appropriate participant sample is a difficult issue. This does not mean we
can not learn from more specific participant samples. It does mean that reporting the
demographics of the sample and being cautious about generalizations is important.
Participant sample choice is just one factor influencing external validity. Setting in-
cludes other factors such as noise, interruption, and distractions. Possible temporal
factors include events that occurred before or are anticipated after the experiment.
Ecological Validity: Ecological validity discussions focus on the degree to which the
experimental situation reflects the type of environment in which the results will be
applied. This concept relates strongly to McGrath’s concept of realism. It is distinct
from the idea of external validity, in that external validity is concerned with whether
the experimental results generalize to other situations, while ecological validity is
concerned with how closely the experimental settings matches the real setting in
which the results might be applied. Thus it is possible to have good ecological valid-
ity; the study is conducted on site, but that the results are applicable only to that site.
This would indicate poor external validity in that the results do not generalize beyond
the specific setting.

4.3 Quantitative Studies Summary Remarks

The number of quantitative studies in information visualization is increasing. Early

examples include the series of studies done by Purchase and her collaborators that
examine the impact of graph drawing aesthetics on comprehension and usability [58,
59, 60, 61]. Dumais et al. [16] explored use of context techniques in web search. For-
lines et al. [23] looked at the effect of display configuration on relationship between
visual search and information visualization tasks. Recently, Willet et al. [81] studied
embedding information visualizations in widgets.
Quantitative experiments have formed the backbone of experimental science and it
is to be expected that they will continue to do so. However, it is relatively easy to find
fault in any given experiment because all factors can not usually be completely con-
trolled. If they are completely controlled, external and ecological validity can be im-
pacted. This is particularly true for studies involving humans. Designing and working
with experiments is often a matter of making choices about what factors are important
and understanding the strengths and limitations of any given study and its results. As
30 S. Carpendale

a community it is important that we recognise that we are working towards a larger

understanding and that any given study will not present the bigger answer. It instead
will contribute to a gradual building of a bigger understanding. For this bigger under-
standing we need to encourage authors to openly discuss the limitations of their stud-
ies, because both the results and the limitations are important. This is also true for
negative results. It can be just as important to understand when there are no differ-
ences among techniques and when these differences exist.

5 Focus on Qualitative Evaluation

Qualitative inquiry works toward achieving a richer understanding by using a more
holistic approach that considers the interplay among factors that influence visualiza-
tions, their development, and their use [56]. Qualitative techniques lend themselves to
being more grounded in more realistic settings and can also be incorporated into all
types of studies. This includes qualitative studies conducted as part of the design
process [64, 73], in situ interviews [83], field studies [72], and use of observational
studies to create design and evaluative criteria that are derived from observed data
[71]. These types of studies offer potential for improved understanding of existing
practices, analysis environments, and cognitive task constraints as they occur in real
or realistic settings. In providing a brief overview of a variety of qualitative methods,
we hope to spark further research and application of qualitative methods in informa-
tion visualization; to expand our empirical approaches to include the application of
qualitative methods to design and evaluation; and to encourage a wider acceptance of
these types of research methodologies in our field.

5.1 Qualitative Methods

At the heart of qualitative methods is the skill and sensitivity with which data is gath-
ered. Whether the records of the data gathered are collected as field notes, artefacts,
video tapes, audio tapes, computer records and logs, or all of these, in qualitative
empirical approaches there are really only two primary methods for gathering data:
observations and interviews. Observation and interview records are usually kept con-
tinually as they occur, as field notes, as regular journal entries as well as often being
recorded as video or audio tapes. Artefacts are collected when appropriate. These can
be documents, drawings, sketches, diagrams, and other objects of use in the process
being observed. These artefacts are sometimes annotated as part of use practices or in
explanation. Also, since the communities we are observing are often technology us-
ers, technology-based records can also include logs, traces, screen captures, etc. Both
observation and interviewing are skills and as such develop with practice and can, at
least to some extent, be learnt. For full discussions on these skills there are many
useful books such as Seidman [65] and Lofland and Lofland [45].

5.1.1 Observation Techniques

The following basic factors have been phrased in terms of developing observational
records but implicitly also offer advice on what to observe:
 Evaluating Information Visualizations 31

• Try to keep jotting down notes unobtrusively. Ideally, notes are taken as obser-
vations occur; however, if one becomes aware that one’s note taking is having
an impact on the observations, consider writing notes during breaks, when
shielded, or at the end of the day.
• Minimize the time gap from observations to note taking. Memory can be quite
good for a few hours but does tend to drop off rapidly.
• Include in observations the setting, a description of the physical setup, the time,
who is present, etc. Drawing maps of layouts and activities can be very useful.
• Remember to include both the overt and covert in activities and communica-
tions. For example, that which is communicated in body language and gestures,
especially if it gets understood and acted upon, is just as important as spoken
communications. But be careful of that grey area where one is not sure to what
extent a communication occurred.
• Remember to include both the positive and negative. Observed frustrations and
difficulties can be extremely important in developing a fuller understanding.
• Do not write notes on both sides of a paper. This may seem trivial but experi-
enced observers say this is a must [6]. You can search for hours, passing over
many times that important note that is on the back of another note.
• Be concrete whenever possible.
• Distinguish between word-for-word or verbatim accounts and those you have
paraphrased and/or remembered.

5.1.2 Interview Techniques

These are a few brief points of advice about interviewing. Do remember that while
sorting out the right questions to ask is important, actively listening to what the par-
ticipant says is the most important of all interviewing skills.
• Make sure that you understand what they are telling you and that the descrip-
tions, explanations they are giving you are complete enough. However, when
asking for clarification, try to avoid implying that their explanations are poor be-
cause one does not want to make one’s participants defensive. Ask instead for
what they meant by particular word usage or if they would explain again. The
use of the word again implies that the interviewer did not catch it all rather than
the explanation was incomplete.
• Limit your inclination to talk. Allow for pauses in the conversation, sometimes
note taking can be useful here. The participant will expect you to be taking
notes. In this situation note taking can actually express respect for what the par-
ticipant has said.
• Remember that the default is that the participant will regard the interview to
some extent as public and thus will tell you the public version. Do listen for and
encourage the less formal, less guarded expression of their thoughts. One exam-
ple, from Seidman [65], is the use of the word ‘challenge’. Challenge is an ex-
cepted term for a problem. The details of the problem might be explained more
fully if one asks what is meant in the given situation by the word challenge.
32 S. Carpendale

• Follow up on what the participant says. Do allow the interview to be shaped by

the information your participant is sharing.
• Avoid leading questions. An important part of minimizing experimenter bias is
wording questions carefully so as to avoid implying any types of answers. For
example, asking a participant what a given experience was like for them, leaves
space for their personal explanations.
• Ask open ended questions. This can involve asking for a temporal reconstruction
of an event or perhaps a working a day or asking for a subjective interpretation
of an event.
• Ask for concrete details. These can help trigger memories.
• With all the above do remember that one of the most important pluses of an in-
terview process is the humanity of interviewer. Being present, aware and sensi-
tive to the process is your biggest asset. These guidelines are just that; guidelines
to be used when useful and ignored when not.

5.2 Types of Qualitative Methodologies

This section is not intended to be a complete collection of all types of qualitative

inquiry. Rather it is meant to give an overview of some of the variations possible, set
in a discussion about when and where they have proven useful. This overview is di-
vided into three sections. First, the type of qualitative methodologies often used in
conjunction with or as part of more quantitative methodologies is discussed. Then, we
mention the approaches taken in the area of heuristic, or, as they are sometimes re-
ferred to ‘discount’, inspection methodologies. The last section will cover some study
methodologies that are intentionally primarily qualitative.

5.2.1 Nested Qualitative Methods

While qualitative methodologies can be at the core of some types of studies, some
aspects of qualitative inquiry are used in most studies. For instance, data gathered by
asking participants for their opinions or preferences is qualitative. Gorard [26] argues
that quantitative methods can not ignore the qualitative factors of the social context of
the study and that these factors are, of necessity, involved in developing an interpreta-
tion of the study results. There are many methods used as part of studies such as labo-
ratory experiments that provide us with qualitative data. The following are simply a
few examples to illustrate how common this mixed approach is.
Experimenter Observations: An important part of most studies is that the experi-
menter keeps notes of what they observe as it is happening. The observations them-
selves can help add some degree of realism to the data and the practice of logging
these observations as they happen during the study helps make them more reliable
than mere memory. However, they are experimenter observations and as such are
naturally subjective. They do record occurrences that were not expected or are not
measurable so that they will also form part of the experimental record. These observa-
tions can be helpful during interpretation of the results in that they may offer explana-
tions for outliers, point towards important experimental re-design, and suggest future
directions for study. Here, experimenter observations augment and enrich the primar-
 Evaluating Information Visualizations 33

ily quantitative results of a laboratory experiment and in this they play an important
but secondary role.
Think-Aloud Protocol: This technique, which involves encouraging participants to
speak their thoughts as they progress through the experiment, was introduced to the
human-computer-interaction community by [43]. Discussions about this protocol in
psychology date back to 1980 [19, 20, 21]. Like most methodologies, this one also
involves tradeoffs. While it gives the experimenter/observer the possibility of being
aware of the participants’ thoughts, it is not natural for most people and can make a
participant feel awkward; thus, think aloud provides additional insight while also
reducing the realism of the study. However, the advantage for hearing about a partici-
pant’s thoughts, plans, and frustrations frequently out-weigh the disadvantages and
this is a commonly used technique. Several variations have been introduced such as
‘talk aloud’ which asks a participant to more simply announce their actions rather
than their thoughts [21].
Collecting Participant Opinions: Most laboratory experiments include some method
by which participant opinions and preferences are collected. This may take the form
of a simple questionnaire or perhaps semi-structured interviews. Most largely quanti-
tative studies such as laboratory experiments do ask these types of questions, often
partially quantifying the participant’s response by such methods as using a Likert
scale [44]. A Likert scale asks a participant to rate their attitude according to degree.
For instance, instead of simply asking a participant, ‘did you like it?’ A Likert scale
might ask the participant to choose one of a range of answers ‘strongly disliked,’
‘disliked,’ ‘neutral,’ ‘liked,’ or ‘strongly liked.’
Summary of Nested Qualitative Methods: The nested qualitative methods men-
tioned in this section may be commonplace to many readers. The point to be made
here is that in the small, that is as part of a laboratory experiment, inclusion of some
qualitative methods is not only commonplace, its value is well recognized. This type
of inclusion of qualitative approaches adds insight, explanations and new questions. It
also can help confirm results. For instance, if participants’ opinions are in line with
quantitative measures – such as the fastest techniques being the most liked – this
confirms the interpretation of the fastest technique being the right one to chose. How-
ever, if they contradict – such as the fastest techniques not being preferred – interest-
ing questions are raised including questioning the notion that fastest is always best.

5.2.2 Inspection Evaluation Methods

We include a discussion of inspection methods because, while they are not studies per
se, they are useful, readily available, and relatively inexpensive evaluation approaches.
The common approach is to use a set of heuristics as a method of focusing attention
on important aspects of the software – interface or visualization – which need to be
considered [54]. These heuristics or guidelines can be developed by experts or from
the writings of experts. Ideally, such an inspection would be conducted by individual
experts or even a group of experts. However, it has been shown that in practice, that a
good set of heuristics can still be effective in application if a few, such as three or
four, different people apply them [54]. For information visualization it is important to
consider exactly what visualization aspects a given set of heuristics will shed light on.
34 S. Carpendale

Usability Heuristics: These heuristics, as introduced and developed by Nielson and

Mack [1994], focus on the usability of the interface and are designed to be applied to
any application, thus are obviously of use to information visualizations. They will
help make sure that general usability issues are considered. These heuristics are dis-
tilled down to ten items – visibility of system status, match between system and real
world, personal control and freedom, consistency and standards, error prevention,
recognition rather than recall, flexibility and efficiency, aesthetic and minimalist de-
sign, errors handling, and help and documentation.
Collaboration Heuristics: When interfaces are designed for collaboration, two addi-
tional major categories arise in importance: communication and coordination. Baker
et al. [4] developed a set of heuristics that explore these issues based on the Mechan-
ics of Collaboration [29]. As information visualizations start to be designed for col-
laborative purposes, both distributed [31, 78] and co-located [35], these heuristics will
also be important.
Information Visualization Heuristics: While the usability heuristics apply to all
infovis software and the collaboration heuristics apply to the growing body of col-
laborative information visualizations, there are areas of an information visualization
that these at best gloss over. In response, the Information Visualization research com-
munity has proposed a variety of specific heuristics. Some pertain to given data do-
mains such as ambient displays [46] and multiple view visualizations [5]. Others
focus on a specific cognitive level, for instance knowledge and task [1], or task and
usability [66]. Tory and Möller [74] propose the use of heuristics based on both visu-
alization guidelines and usability. As explored by Zuk and Carpendale [84], we can
also consider developing heuristics based on the advice from respected experts such
as design advice collected from Tufte’s writings [75, 76, 77], semiotic considerations
as expressed by Bertin [8] and/or research in cognitive and perceptual science as col-
lected by Ware [79]. Alternatively, we can start from information visualization basics
such as presentation, representation and interaction [68]. However, a concept such as
presentation cuts across design and perception, while representation advice, such as
what types of visuals might best represent what types of data, might be distilled from
the guidelines put forth by Bertin [8] and from an increasing body of cognitive sci-
ence as gathered in Ware [79]. Sorting out how to best condense these is a task in
itself [52, 85]. “At this stage of development of heuristics for information visualiza-
tion we have reached a similar problem as described by Nielson and Mack [54]. It is a
difficult problem to assess which list(s) are better for what reasons and under what
conditions. This leads to the challenges of developing an optimal list that comprises
the most important or common Information Visualization problems” (page 55, [85]).
Summary of Inspection Evaluation Methods: While experience in the human com-
puter interaction communities and the growing body of information visualization
specific research indicates that heuristics may prove a valuable tool for improving the
quality of information visualizations, there is considerable research yet to be con-
ducted in the development of appropriate taxonomies and application processes for
heuristics in information visualization.
The currently recommended application approach for usability heuristics is that
evaluators apply the heuristics in a two pass method. The first pass is done to gain an
overview and second is used to asses in more detail each interface component with
 Evaluating Information Visualizations 35

each heuristic [54]. The original use indicated that in most situations three evaluators
would be cost effective and find most usability problems [54]. However, subsequent
use of heuristics for web site analysis appears to sometimes need more evaluators [9,
69]. Further, this may depend on the product. While application of heuristics has not
yet been formally studied in terms of web sites, it does introduce the possibility that
information visualization heuristics may also need to be data, task or purpose specific.
Heuristics are akin to the design term guidelines in that both provide a list of ad-
vice. Design guidelines are often usefully applied in a relatively ad hoc manner as
factors to keep in mind during the design process and heuristic lists can definitely be
similarly used. While there are definitely benefits that accrue in the use of guidelines
and heuristics, it is important to bear in mind that they are based on what is known to
be successful and thus tend not to favour the unusual and the inventive. In the design
world, common advice is that while working without knowledge of guidelines is fool-
ish, following them completely is even worse.

5.2.3 Qualitative Methods as Primary

A common reason for using qualitative inquiry is to develop a richer understanding of
a situation by using a more holistic approach. Commonly, the qualitative research
method’s goal is to collect data that enables full, rich descriptions rather than to make
statistical inferences [3, 14]. There are a wealth of qualitative research methods that
can help us to gain a better understanding of the factors that influence information
visualization use and design. Just as we have pointed out how qualitative methods can
be effectively used within quantitative research, qualitative research can also include
some quantitative results. For instance, there may be factors that can be numerically
recorded. These factors can then be presented in combination with qualitative data.
For example, if a questionnaire includes both fixed-choice questions and open ended
questions, quantitative measurement and qualitative inquiry are being combined [56].
Qualitative methods can be used at any time in the development life cycle. A fin-
ished or near to finished product can be assessed via case studies or field studies.
Also, there is a growing use of these methods as a preliminary step in the design proc-
ess. The HCI and particularly the computer-supported cooperative work (CSCW)
research communities have successfully been using qualitative methods to gain in-
sight that can inform the initial design. CSCW researchers have learned a lot about
how to support people working together with technology through pre-design observa-
tion and qualitative analysis of how people work together without technology. The
basic idea is that through observations of participants' interactions with physical arte-
facts, a richer understanding of basic activities can be gained and that this understand-
ing can be used to inform interface design. This approach generally relies on observa-
tion of people, inductive derivation of hypotheses via iterative data collection, analy-
sis, and provisional verification [14]. For example, Tang's study of group design ac-
tivities around shared workspaces revealed the importance of gestures and the work-
space itself in mediating and coordinating collaborative work [73]. Similarly, Scott et
al. [64] studied traditional tabletop gameplay and collaborative design, specifically
focusing on the use of tabletop space, and the sharing of items on the table. Both
studies are an example of how early evaluation can inform the design of digital sys-
tems. In both cases, the authors studied traditional, physical contexts first, to under-
stand participants' interactions with the workspace, the items in the workspace, and
36 S. Carpendale

within the group. The results of these experiments are regarded as providing impor-
tant information about what group processes to support and some indication about
how this might be done. This type of research can be particularly important in com-
plex or sensitive scenarios such as health care situations [72]. Brereton and McGarry
[11] observed groups of engineering students and professional designers using physi-
cal objects to prototype designs. They found that the interpretation and use of physical
objects depended greatly on the context of its placement, indicating that the context of
people's work is important and is difficult to capture quantitatively. Their goal was to
determine implications for the design of tangible interfaces. Other examples include
Saraiya et al. [63] who used domain expert assessments of insight to evaluate bioin-
formatics visualizations, while Mazza and Berre [48] used focus groups and semi-
structured interviews in their analysis of visualization approaches to support instruc-
tors in web-based distance education.
The following are simply examples of empirical methods in which gathering of
qualitative data is primary. There are many others; for instance, Moggridge [51] men-
tions that his group makes active use of fifty-one qualitative methods in their design
processes.
In Situ Observational Studies: These studies are at the heart of field studies. Here,
the experimenter gets permission to observe activities as they take place in situ. In
these studies the observer does their best to remain unobtrusive during the observa-
tions. The ideal in Moggridge’s terms is to become as a ‘fly on the wall’ that no one
notices [51]. This can be hard to achieve in an actual setting. However, over time a
good observer does usually fade into the background. Sometimes observations can be
collected via video and audio tapes to avoid the more obvious presence of a person as
observer but sometimes making such recordings is not appropriate as in medical situa-
tions. In these studies the intention is usually to gather a rich description of the situa-
tion being observed. However, there is both a difference and an overlap in the type of
observations to be gathered when the intention is (a) to better understand the particu-
lar activities in a given of setting, or (b) to use these observations to inform technol-
ogy design. Thus, because different details are of prime interest it is important that
our research community conducts these types of observational studies to better inform
initial design as well as to better understand the effectiveness of new technology in
use. These studies have high realism, result in rich context explicit data and are time
and labour intensive when it comes to both data collection and data analysis.
Participatory Observation: This practice is the opposite of participatory design.
Here an information visualization expert becomes part of the application expert’s
team to experience the work practices first hand rather than application experts be-
coming part of the information visualization design team. In participatory observa-
tion, additional insights can be gained through first-hand observer experience of the
tasks and processes of interest in the context of the real world situation. Here, rather
than endeavouring to be unobtrusive, the observer works towards becoming an ac-
cepted part of the community. Participatory observation is demonstrably an effective
approach since as trust and rapport develop, an increasingly in-depth understanding is
possible. Our research community is interested in being able to better understand the
work practices of many different types of knowledge workers. These workers are
usually highly trained, highly paid, and often under considerable time pressures. Not
 Evaluating Information Visualizations 37

surprisingly, they are seldom willing to accept an untrained observer as part of their
team. Since information visualization researchers are of necessity highly trained
themselves, it is rare that an information visualization researcher will have the neces-
sary additional training to become accepted as a participatory observer. However,
domain expertise is not always essential for successful participatory observation.
Expert study participants can train an observer on typical data analysis tasks – a proc-
ess which may take several hours, and then “put them to work” on data analysis using
their existing tools and techniques. The observer keeps a journal of the experience and
the outcomes of the analysis were reviewed with the domain experts for validity.
Even as a peripheral participant, valuable understandings of domain, tasks, and work
culture can be developed which help clarify values and assumptions about data, visu-
alizations, decision making and data insights important to the application domain.
These understandings and constructs can be important to the information visualization
community in the development of realistic tools.
Laboratory Observational Studies: These studies use observational methodologies
in a laboratory setting. A disadvantage of in situ observations is that they often require
lengthy observations. For instance, if the observer is interested in how an analyst uses
visual data, they will have to wait patiently until the analyst does this task. Since an
analyst may have many other tasks – meetings, conference calls, reports, etc. – this
may take hours or even days. One alternative to the lengthy in situation wait is to
design an observational experiment in which, similarly to a laboratory experiment, the
experimenter designs a setting, a procedure and perhaps even a set of tasks. Consider,
for example, developing information visualizations to support co-located collabora-
tion. Some design advice on co-located collaborative aspects is available in the com-
puter supported cooperative work literature [35]. However, while this advice is useful,
it does not inform us specifically about how teams engage in collaborative tasks when
using visual information. Details such as how and when visualizations will be shared
and what types of analysis processes need to be specifically supported in collaborative
information visualization systems were missing. Here, an observational approach is
appropriate because the purpose is to better understand the flow and nature of the
collaboration among participants, rather than answering quantifiable lower-level ques-
tions. In order to avoid temporal biases in existing software, pencil and paper based
visualizations were used. This allowed for the observation of free arrangement of
data, annotation practices, and collaborative processes unconstrained by any particular
visualization software [36].
Contextual Interviews: As noted in Section 5.1, interviewing in itself is core to
qualitative research. Conducting an interview about a task, setting, or application of
interest within the context in which this work usually takes place is just one method
that can enrich the interview process. Here the realism of the setting helps provide the
context that can bring to mind the day-to-day realities during the interview process
(for further discussion see Holtzblatt and Beyer 1998). For example, to study how
best to support the challenging problem of medical diagnosis, observing and inter-
viewing physicians in their current work environment might help to provide insights
into their thought processes that would be difficult to capture with other methodolo-
gies. A major benefit of qualitative study can be seeing the big picture – the context in
which a new visualization support may be used. The participants' motives, misgiv-
38 S. Carpendale

ings, and opinions shed light on how they relate to existing support, and can effec-
tively guide the development of new support. This type of knowledge can be very
important at the early stage of determining what types of information visualizations
may be of value.
Summary of qualitative methods as primary: These four methods are just exam-
ples of a huge variety of possibilities. Other methods include action research [42],
focus groups [48], and many more. All these types of qualitative methods have the
potential to lessen the task and data comprehension divide between ourselves as visu-
alization experts and the domain experts for whom we are creating visualizations.
That is, while we can not become analysts, doctors, or linguists, we can gain a deeper
understanding of how they work and think. These methods can open up the design
space, revealing new possibilities for information visualizations, as well as additional
criteria on which to measure success.

5.3 Challenges for Qualitative Methods

A considerable challenge to qualitative methods is that they are particularly labour

intensive. Gathering data is a slow process and rich note taking is an intensive under-
taking, as are transcribing and subsequent analysis.

5.3.1 Sample Sizes

Sample sizes for qualitative research are determined differently than for quantitative
research. Since qualitative research is not concerned with making statistically signifi-
cant statements about a phenomenon, the sample sizes are often lower than required
for quantitative research. Often, sample sizes are determined during the study. For
instance, a qualitative inquiry may be continued until one no longer appears to be
gaining new data through observation [3]. There is no guideline to say when this
‘saturation’ may occur [70]. Sample sizes may vary greatly depending on the scope of
the research problem but also the experience of the investigator. An experienced in-
vestigator may reach a theoretical saturation earlier than a novice investigator. Also,
because each interview and/or observation can result in a large amount of data, some-
times compromises in sample size have to be made due to considerations about the
amount of data that can be effectively processed.

5.3.2 Subjectivity
Experimenter subjectivity can be seen as an asset because of the sensitivity that can
be brought to the observation process. The quality of the data gathering and analysis
is dependent on the experience of the investigator [56]. However, the process of gath-
ering any data must be concerned with obtaining representative data. The questions
circle about whether the observer has heard or understood fully and whether these
observations are reported accurately. Considerations include:
• Is this a first person direct report? Otherwise normal common sense about 2nd,
3rd, and 4th hand reports needs to be considered.
• Does the spatial location of the observer provide an adequate vantage point from
which to observe, or might it have led to omissions?
 Evaluating Information Visualizations 39

• Are the social relationships of the observer free from associations that might in-
duce bias?
• Does the report appear to be self-serving? Does it benefit the experimenter to the
extent that it should be questioned?
• Is the report internally consistent? Do the facts within the report support each
other?
• Is the report externally consistent? Do the facts in the report agree with other in-
dependent reports?
As a result it is important to be explicit about data collection methods, the position of
the researcher with respect to the subject matter, analysis processes, and codes. These
details make it possible for other researchers to verify results.
In qualitative research it is acknowledged that the researcher's views, research con-
text, and interpretations are an essential part of the qualitative research method as
long as they are grounded in the collected data [3]. This does not, however, mean that
qualitative evaluations are less trustworthy compared to quantitative research. Auer-
bach suggests using the concept of ‘transferability’ rather than ‘generalizability’ when
thinking about the concepts of reliability and validity in qualitative research [3]. It is
more important that the theoretical understanding we have gained can also be found in
other research situations or systems and can be extended and developed further when
applied to other scenarios. This stands in contrast to the concept of generalizability in
quantitative research that wants to prove statistically that the results are universally
applicable within the population under study.
Sometimes the point has been raised that if results do not generalize how can they
be of use when designing software for general use. For example, qualitative methods
might be used to obtain a rich description of a particular situation perhaps only ob-
serving the processes of two or three people. The results of a study like this may or
may not generalize and the study itself provides no proof that they do. What we have
is existence proof: that such processes are in use in at least two or three instances.
Consider the worst case; that is that this rich description is an outlier that occurs only
rarely. For design purposes, outliers are also important and sensitive design for out-
liers has been often shown to create better designs for all. For example, motion sen-
sors to open doors may have been designed for wheelchairs but actually are useful
features for all.

5.3.3 Analyzing Qualitative Data

Qualitative data may be analyzed using qualitative, quantitative, or a combination of
both methods. Mixed methods research includes a qualitative phase and a quantitative
phase in the overall research study in order to triangulate results from different meth-
ods, to complement results from one method with another, or to increase the breadth
and range of inquiry by using different methods [28].
Many of the qualitative analysis methods can be grouped as types of thematic
analysis, in which analysis starts from observations, then themes are sensed through
review of the data, and finally coded [10]. Coding is the process of subdividing and
labeling raw data, then reintegrating collected codes to form a theory [70]. Moving
from the raw data into themes and a code set may proceed using one of three ap-
40 S. Carpendale

proaches: data-driven, motivated from previous research, or theory-driven, each with

respectively decreasing levels of sensitivity to the data [10]. In the first style, data-
driven, commonly called open coding [14]; themes and a code set are derived directly
from the data and nothing else. If the analysis is motivated by previous research, the
questions and perhaps codes from the earlier research can be applied to the new data
to verify, extend or contrast the previous results. With theory-driven coding one may
think using a given theory, such as grounded theory [13], or ethno-methodology [24],
as a lens through which to view the data.
In either case the coded data may then be interpreted in more generalized terms.
Qualitatively coded data may then be used with quantitative or statistical measures to
try and distinguish themes or sampling groups.

5.4 Qualitative Summary

Qualitative studies can be a powerful methodology by which one can capture salient
aspects of a problem that may provide useful design and evaluation criteria. Quantita-
tive evaluation is naturally precision-oriented, but a shift from high precision to high
fidelity may be made with the addition of qualitative evaluations. In particular, while
qualitative evaluations can be used throughout the entire development life cycle in
other research areas such as CSCW [41, 52, 64, 73], observational studies have been
found to be especially useful for informing design. Yet these techniques are under-
used and under-reported in the information visualization literature. Broader ap-
proaches to evaluation, different units of analysis and sensitivity to context are impor-
tant when complex issues such as insight, discovery, confidence and collaboration
need to be assessed. In more general terms, we would like to draw attention to qualita-
tive research approaches which may help to address difficult types of evaluation ques-
tions. As noted by Isenberg el al. [36], a sign in Albert Einstein's office which read,
‘Everything that can be counted does not necessarily count; everything that counts
cannot necessarily be counted’ is particularly salient to this discussion in reminding
us to include empirical research about important data that can not necessarily be
counted.

6 Conclusions
In this paper we have made a two-pronged call: one for more evaluations in general
and one for a broader appreciation of the variety of and importance of many different
types of empirical methodologies. To achieve this, we as a research community need
to both conduct more empirical research and to be more welcoming of this research in
our publication venues. As noted in Section 4, even empirical laboratory experiments,
as our most known type of empirical methodology, are often difficult to publish. One
factor in this is that no empirical method is perfect. That is, there is always a trade-off
between generalizability, precision, and realism. An inexperienced reviewer may rec-
ommend rejection based on the fact that one of these factors is not present, while realis-
tically at least one will always be compromised. Empirical research is a slow, labour-
intensive process in which understanding and insight can develop through time. That
said, there are several important factors to consider when publishing empirical re-
search. These include:
 Evaluating Information Visualizations 41

• That the empirical methodology was sensitively chosen. The methodology

should be a good fit to the research question, the situation and the research goals.
• That the study was conducted with appropriate rigor. All methodologies have
their own requirements for rigor and these should be followed. However, while
trying to fit the rigor from one methodology onto another is not appropriate, de-
veloping hybrid methodologies that better fit a given research situation and
benefit from two or more methodologies should be encouraged.
• That sufficient details are published so that the reader can fully understand the
processes and if appropriate, reproduce them.
• That the claims should be made appropriately according to the strengths of the
chosen methodology. For instance, if a given methodology does not generalize
well, then generalizations should not be drawn from the results.
While there is growing recognition in our research community that evaluation infor-
mation visualization is difficult [55, 57, 67], the recognition of this difficulty has not
in itself provided immediate answers of how to approach this problem. Two positive
recent trends of note are: one, that more evaluative papers in the form of usability
studies have been published [25, 40, 47, 63, 80, 82], and two, that there are several
papers that have made a call for more qualitative evaluations and complementary
qualitative and quantitative approaches [18, 36, 48, 74].
This paper is intended merely as a pointer to a greater variety of empirical method-
ologies and encouragement towards their appreciation and even better their active use.
There are many more such techniques and these types of techniques are being devel-
oped and improved continuously. There are good benefits to be had through active
borrowing from ethnographic and sociological research methods, and applying them
to our information visualization needs. In this paper we have argued for an increased
awareness of empirical research. We have discussed the relationship of empirical
research to information visualization and have made a call for a more sensitive appli-
cation of this type of research [27]. In particular, we encourage thoughtful application
of a greater variety of evaluative research methodologies in information visualization.

Acknowledgments. The ideas presented in this paper have evolved out of many dis-
cussions with many people. In particular this includes: Christopher Collins, Marian
Dörk, Saul Greenberg, Carl Gutwin, Mark S. Hancock, Uta Hinrichs, Petra Isenberg,
Stacey Scott, Amy Voida, and Torre Zuk.

References
1. Amar, R.A., Stasko, J.T.: Knowledge Precepts for Design and Evaluation of Information
Visualizations. IEEE Transactions on Visualization and Computer Graphics 11(4), 432–
442 (2005)
2. Andrews, K.: Evaluating Information Visualisations. In: Proceedings of the 2006 AVI Work-
shop on BEyond Time and Errors: Novel Evaluation Methods for Information Visualiza-
tion, pp. 1–5 (2006)
3. Auerbach, C.: Qualitative Data: An Introduction to Coding and Analysis. University Press,
New York (2003)
42 S. Carpendale

4. Baker, K., Greenberg, S., Gutwin, C.: Empirical Development of a Heuristic Evaluation
Methodology for Shared Workspace Groupware. In: Proceedings of the ACM Conference
on Computer Supported Cooperative Work, pp. 96–105. ACM Press, New York (2002)
5. Baldonado, M., Woodruff, A., Kuchinsky, A.: Guidelines for Using Multiple Views in
Information Visualization. In: Proeedings of the Conference on Advanced Visual Inter-
faces (AVI), pp. 110–119. ACM Press, New York (2000)
6. Barzun, J., Graff, H.: The Modern Researcher, 3rd edn. Harcourt Brace Jovanvich, New
York (1977)
7. BELIV 2006, accessed http://www.dis.uniroma1.it/~beliv06/ (February 4,
2008)
8. Bertin, J.: Semiology of Graphics (Translation: William J. Berg). University of Wisconsin
Press (1983)
9. Bevan, N., Barnum, C., Cockton, G., Nielsen, J., Spool, J., Wixon, W.: The “Magic Num-
ber 5”: Is It Enough for Web Testing? In: CHI Extended Abstracts, pp. 698–699. ACM
Press, New York (2003)
10. Boyatzis, R.: Transforming Qualitative Information: Thematic Analysis and Code Devel-
opment. Sage Publications, London (1998)
11. Brereton, M., McGarry, B.: An Observational Study of How Objects Support Engineering
Design Thinking and Communication: Implications for the Design of Tangible Media. In:
Proceedings of the ACM Conference on Human Factors in Computing Systems (CHI’00),
pp. 217–224. ACM Press, New York (2000)
12. Chen, C., Czerwinski, M.: Introduction to the Special Issue on Empirical Evaluation of
Information Visualizations. International Journal of Human-Computer Studies 53(5), 631–
635 (2000)
13. Corbin, J., Strauss, A.: Basics of Qualitative Research: Techniques and Procedures for
Developing Grounded Theory, 3rd edn. Sage Publications, Los Angeles (2008)
14. Creswell, J.: Qualitative Inquiry and Research Design: Choosing Among Five Traditions.
Sage Publications, London (1998)
15. Dix, A., Finlay, J., Abowd, G., Beale, R.: Human Computer Interaction, 2nd edn. Prentice-
Hall, Englewood Cliffs (1998)
16. Dumais, S., Cutrell, E., Chen, H.: Optimizing Search by Showing Results In Context. In:
Proc. CHI’01, pp. 277–284. ACM Press, New York (2001)
17. Eberts, R.E.: User Interface Design. Prentice-Hall, Englewood Cliffs (1994)
18. Ellis, E., Dix, A.: An Explorative Analysis of User Evaluation Studies in Information
Visualization. In: Proceedings of the Workshop on Beyond Time and Errors: Novel
Evaluation Methods for Information Visualization, BELIV (2006)
19. Ericsson, K., Simon, H.: Verbal Reports as Data. Psychological Review 87(3), 215–251
(1980)
20. Ericsson, K., Simon, H.: Verbal Reports on Thinking. In: Faerch, C., Kasper, G. (eds.)
Introspection in Second Language Research, pp. 24–54. Multilingual Matters, Clevedon,
Avon (1987)
21. Ericsson, K., Simon, H.: Protocol Analysis: Verbal Reports as Data, 2nd edn. MIT Press,
Boston (1993)
22. Fall, J., Fall, A.: SELES: A Spatially Explicit Landscape Event Simulator. In: Proceedings
of GIS and Environmental Modeling, pp. 104–112. National Center for Geographic Infor-
mation and Analysis (1996)
23. Forlines, C., Shen, C., Wigdor, D., Balakrishnan, R.: Exploring the effects of group size
and display configuration on visual search. In: Computer Supported Cooperative Work
2006 Conference Proceedings, pp. 11–20 (2006)
24. Garfinkel, H.: Studies in Ethnomethodology. Polity Press, Cambridge (1967)
 Evaluating Information Visualizations 43

25. Gonzalez, V., Kobsa, A.: A Workplace Study of the Adoption of Information Visualization
systems. In: Proceedings of the International Conference on Knowledge Management, pp.
92–102 (2003)
26. Gorard, S.: Combining Methods in Educational Research. McGraw-Hill, New York (2004)
27. Greenberg, S., Buxton, B.: Usability Evaluation Considered Harmful (Some of the Time).
In: Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (2008)
28. Greene, J., Caracelli, V., Graham, W.: Toward a Conceptual Framework for Mixed-Method
Evaluation Design. Educational Evaluation and Policy Analysis 11(3), 255–274 (1989)
29. Gutwin, C., Greenberg, S.: The Mechanics of Collaboration: Developing Low Cost Usabil-
ity Evaluation Methods for Shared Workspaces. In: Proceedings WETICE, pp. 98–103.
IEEE Computer Society Press, Los Alamitos (2000)
30. Healey, C.G.: On the Use of Perceptual Cues and Data Mining for Effective Visualization
of Scientific Datasets. In: Proceedings of Graphics Interface, pp. 177–184 (1998)
31. Heer, J., Viegas, F., Wattenberg, M.: Voyagers and Voyeurs: Supporting Asynchronous
Collaborative Information Visualization. In: Proceedings of the Conference on Human
Factors in Computing Systems (CHI’07), pp. 1029–1038. ACM Press, New York (2007)
32. Holtzblatt, K., Beyer, H.: Contextual Design: Defining Customer-Centered Systems. Mor-
gan Kaufmann, San Francisco (1998)
33. Huck, S.W.: Reading Statistics and Research, 4th edn. Pearson Education Inc., Boston (2004)
34. Interrante, V.: Illustrating Surface Shape in Volume Data via Principal Direction-Driven 3D
Line Integral Convolution. Computer Graphics, Annual Conference Series, pp. 109–116
(1997)
35. Isenberg, P., Carpendale, S.: Interactive Tree Comparison for Co-located Collaborative In-
formation Visualization. IEEE Transactions on Visualization and Computer Graphics 12(5)
(2007)
36. Isenberg, P., Tang, A., Carpendale, S.: An Exploratory Study of Visual Information Analy-
sis. In: Proceedings of the Conference on Human Factors in Computing Systems (CHI’08),
ACM Press, New York (to appear, 2008)
37. Kay, J., Reiger, H., Boyle, M., Francis, G.: An Ecosystem Approach for Sustainability:
Addressing the Challenge of Complexity. Futures 31(7), 721–742 (1999)
38. Kim, S., Hagh-Shenas, H., Interrante, V.: Conveying Shape with Texture: Experimental
Investigations of Texture’s Effects on Shape Categorization Judgments. IEEE Transactions
on Visualization and Computer Graphics 10(4), 471–483 (2004)
39. Kleffner, D.A., Ramachandran, V.S.: On the Perception of Shape from Shading. Percep-
tion and Psychophysics 52(1), 18–36 (1992)
40. Kobsa, A.: User Experiments with Tree Visualization Systems. In: Proceedings of the
IEEE Symposium on Information Visualization, pp. 9–26 (2004)
41. Kruger, R., Carpendale, S., Scott, S.D., Greenberg, S.: Roles of Orientation in Tabletop
Collaboration: Comprehension, Coordination and Communication. Journal of Computer
Supported Collaborative Work 13(5–6), 501–537 (2004)
42. Lewin, C. (ed.): Research Methods in the Social Sciences. Sage Publications, London (2004)
43. Lewis, C., Rieman, J.: Task-Centered User Interface Design: A Practical Introduction (1993)
44. Likert, R.: A Technique for the Measurement of Attitudes. Archives of Psychology 140, 1–55
(1932)
45. Lofland, J., Lofland, L.: Analyzing Social Settings: A Guide to Qualitative Observation
and Analysis. Wadsworth Publishing Company, CA, USA (1995)
46. Mankoff, J., Dey, A., Hsieh, G., Kientz, J., Lederer, S., Ames, A.: Heuristic Evaluation of
Ambient Displays. In: Proceedings of CHI ’03, pp. 169–176. ACM Press, New York (2003)
47. Mark, G., Kobsa, A., Gonzalez, V.: Do Four Eyes See Better Than Two? Collaborative
Versus Individual Discovery in Data Visualization Systems. In: Proceedings of the IEEE
Conference on Information Visualization (IV’02), July 2002, pp. 249–255. IEEE Press,
Los Alamitos (2002)
44 S. Carpendale

48. Mazza, R., Berre, A.: Focus Group Methodology for Evaluating Information Visualization
Techniques and Tools. In: Proceedings of the International Conference on Information
Visualization IV (2007)
49. McCarthy, D.: Normal Science and Post-Normal Inquiry: A Context for Methodology (2004)
50. McGrath, J.: Methodology Matters: Doing Research in the Social and Behavioural Sci-
ences. In: Readings in Human-Computer Interaction: Toward the Year 2000, Morgan
Kaufmann, San Francisco (1995)
51. Moggridge, B.: Design Interactions. MIT Press, Cambridge (2006)
52. Morris, M.R., Ryall, K., Shen, C., Forlines, C., Vernier, F.: Beyond “Social Protocols”:
Multi-User Coordination Policies for Co-located Groupware. In: Proceedings of the ACM
Conference on Computer-Supported Cooperative Work (CSCW, Chicago, IL, USA), CHI
Letters, November 6-10, 2004, pp. 262–265. ACM Press, New York (2004)
53. Morse, E., Lewis, M., Olsen, K.: Evaluating Visualizations: Using a Taxonomic Guide.
Int. J. Human-Computer Studies 53, 637–662 (2000)
54. Nielsen, J., Mack, R.: Usability Inspection Methods. John Wiley & Sons, Chichester (1994)
55. North, C.: Toward Measuring Visualization Insight. IEEE Computer Graphics and Appli-
cations 26(3), 6–9 (2006)
56. Patton, M.Q.: Qualitative Research and Evaluation Methods, 3rd edn. Sage Publications,
London (2001)
57. Plaisant, C.: The Challenge of Information Visualization Evaluation. In: Proceedings of the
Working Conference on Advanced Visual Interfaces, pp. 109–116 (2004)
58. Purchase, H.C., Hoggan, E., Görg, C.: How Important Is the “Mental Map”? – An Empiri-
cal Investigation of a Dynamic Graph Layout Algorithm. In: Kaufmann, M., Wagner, D.
(eds.) GD 2006. LNCS, vol. 4372, pp. 184–195. Springer, Heidelberg (2007)
59. Purchase, H.C.: Effective Information Visualisation: A Study of Graph Drawing Aesthet-
ics and Algorithms. Interacting with Computers 13(2), 477–506 (2000)
60. Purchase, H.C.: Performance of Layout Algorithms: Comprehension, Not Computation.
Journal of Visual Languages and Computing 9, 647–657 (1998)
61. Brandenburg, F.J. (ed.): GD 1995. LNCS, vol. 1027. Springer, Heidelberg (1996)
62. Reilly, D., Inkpen, K.: White Rooms and Morphing Don’t Mix: Setting and the Evaluation
of Visualization Techniques. In: Proceedings of the SIGCHI Conference on Human Fac-
tors in Computing Systems, pp. 111–120 (2007)
63. Saraiya, P., North, C., Duca, K.: An Insight-Based Methodology for Evaluating Bioinfor-
matics Visualizations. IEEE Transactions on Visualization and Computer Graphics 11(4),
443–456 (2005)
64. Scott, S.D., Carpendale, S., Inkpen, K.: Territoriality in Collaborative Tabletop Workspaces.
In: Proceedings of the ACM Conference on Computer-Supported Cooperative Work
(CSCW, Chicago, IL, USA), CHI Letters, November 6-10, 2004, pp. 294–303. ACM Press,
New York (2004)
65. Seidman, I.: Interviewing as Qualitative Research: A Guide for Researchers in Education
and the Social Sciences. Teachers’ College Press, New York (1998)
66. Shneiderman, B.: The Eyes Have It: A Task by Data Type Taxonomy for Information
Visualizations. In: Proceedings of the IEEE Symposium on Visual Languages, pp. 336–
343. IEEE Computer Society Press, Los Alamitos (1996)
67. Shneiderman, B., Plaisant, C.: Strategies for Evaluating Information Visualization Tools:
Multi-Dimensional In-Depth Long-Term Case Studies. In: Proceedings of the Workshop
on BEyond Time and Errors: Novel Evaluation Methods for Information Visualization,
BELIV (2006)
68. Spence, R.: Information Visualization, 2nd edn. Addison-Wesley, Reading (2007)
69. Spool, J., Schroeder, W.: Testing Web Sites: Five Users is Nowhere Near Enough. In: CHI
’01 Extended Abstracts, pp. 285–286. ACM Press, New York (2001)
 Evaluating Information Visualizations 45

70. Strauss, A.L., Corbin, J.: Basics of Qualitative Research: Techniques and Procedures for
Developing Grounded Theory. Sage Publications, London (1998)
71. Tang, A., Tory, M., Po, B., Neumann, P., Carpendale, S.: Collaborative Coupling over
Tabletop Displays. In: Proceedings of the Conference on Human Factors in Computing
Systems (CHI’06), pp. 1181–1290. ACM Press, New York (2006)
72. Tang, A., Carpendale, S.: An observational study on information flow during nurses’ shift
change. In: Proc. of the ACM Conf. on Human Factors in Computing Systems (CHI), pp.
219–228. ACM Press, New York (2007)
73. Tang, J.C.: Findings from observational studies of collaborative work. International Jour-
nal of Man-Machine Studies 34(2), 143–160 (1991)
74. Tory, M., Möller, T.: Evaluating Visualizations: Do Expert Reviews Work. IEEE Com-
puter Graphics and Applications 25(5), 8–11 (2005)
75. Tufte, E.: The Visual Display of Quantitative Information. Graphics Press, Cheshire (1986)
76. Tufte, E.: Envisioning Information. Graphics Press, Cheshire (1990)
77. Tufte, E.: Visual Explanations. Images and Quantities, Evidence and Narrative. Graphics
Press, Cheshire (1997)
78. Viegas, F.B., Wattenberg, M., van Ham, F., Kriss, J., McKeon, M.: Many Eyes: A Site for
Visualization at Internet Scale. IEEE Transactions on Visualization and Computer Graph-
ics (Proceedings Visualization / Information Visualization 2007) 12(5), 1121–1128 (2007)
79. Ware, C.: Information Visualization: Perception for Design, 2nd edn. Morgan Kaufmann,
San Francisco (2004)
80. Wigdor, D., Shen, C., Forlines, C., Balakrishnan, R.: Perception of Elementary Graphical
Elements in Tabletop and Multi-surface Environments. In: Proceedings of the Confer-
ence on Human Factors in Computing Systems (CHI’07), pp. 473–482. ACM Press,
New York (2007)
81. Willett, W., Heer, J., Agrawala, M.: Scented Widgets: Improving Navigation Cues with
Embedded Visualizations. In: INFOVIS 2007. IEEE Symposium on Information Visuali-
zation (2007)
82. Yost, B., North, C.: The Perceptual Scalability of Visualization. IEEE Transactions on
Visualization and Computer Graphics 12(5), 837–844 (2006)
83. Zuk, T.: Uncertainty Visualizations. PhD thesis. Department of Compute Science, Univer-
sity of Calgary (2007)
84. Zuk, T., Carpendale, S.: Theoretical Analysis of Uncertainty Visualizations. In: Proceedings
of SPIE Conference Electronic Imaging, Vol. 6060: Visualization and Data Analysis (2006)
85. Zuk, T., Schlesier, L., Neumann, P., Hancock, M.S., Carpendale, S.: Heuristics for Infor-
mation Visualization Evaluation. In: Proceedings of the Workshop BEyond Time and Er-
rors: Novel Evaluation Methods for Information Visualization (BELIV 2006), held in con-
junction with the Working Conference on Advanced Visual Interfaces (AVI 2006), ACM
Press, New York (2006)
 Theoretical Foundations of Information Visualization

Helen C. Purchase1, Natalia Andrienko2, T.J. Jankun-Kelly3, and Matthew Ward4

1
Department of Computing Science, University of Glasgow,
17 Lilybank Gardens, G12 8QQ, UK,
[email protected]
2
Fraunhofer Institut Intelligent Analysis & Information Systems (FhG IAIS),
Schloss Birlinghoven, D-53754 Sankt-Augustin, Germany,
[email protected]
3
Department of Computer Science and Engineering,
Bagley College of Engineering, Mississippi State University,
Mississippi State, MS 39762, USA,
[email protected]
4
Computer Science Department, Worcester Polytechnic Institute,
100 Institute Road, Worcester, MA 01609-2280, USA,
[email protected]

Abstract. The field of Information Visualization, being related to many other

diverse disciplines (for example, engineering, graphics, statistical modeling) suf-
fers from not being based on a clear underlying theory. The absence of a frame-
work for Information Visualization makes the significance of achievements in this
area difficult to describe, validate and defend. Drawing on theories within associ-
ated disciplines, three different approaches to theoretical foundations of Informa-
tion Visualization are presented here: data-centric predictive theory, information
theory, and scientific modeling. Definitions from linguistic theory are used to
provide an over-arching framework for these three approaches.

1 Introduction
Information Visualization suffers from not being based on a clearly defined underly-
ing theory, making the tools we produce difficult to validate and defend, and meaning
that the worth of a new visualization method cannot be predicted in advance of im-
plementation. There is much unease in the community as to the lack of theoretical
basis for the many impressive and useful tools that are designed, implemented and
evaluated by Information Visualization researchers.
The purpose of a theory is to provide a framework within which to explain phe-
nomena. This framework can then be used to both evaluate and predict events, in this
case, users’ insight or understanding of visualization, and their use of it. An Informa-
tion Visualization theory would enable us to evaluate visualizations with reference to
an established and agreed framework, and to predict the effect of a novel visualization
method.
This is not to say that a single theory would be able to encapsulate the whole of
the Information Visualization field; it may be that multiple theories at different levels
are needed. We already make use of many existing cognitive and perceptual theories,
as well as established statistical methods. It might be that the complexity of Informa-

A. Kerren et al. (Eds.): Information Visualization, LNCS 4950, pp. 46 – 64, 2008.
© Springer-Verlag Berlin Heidelberg 2008
 Theoretical Foundations of Information Visualization 47

tion Visualization as relating to engineering, cognition, design and science requires

the use of several theories each taking a different perspective.
As a starting point, we liken the understanding of visualization to the understand-
ing of ideas expressed in language. We draw on two perspectives in linguistic theory:
language as representation and language as process.
In considering the representation of language, lexical tokens are syntactically or-
dered to produce a semantic concept which the reader understands with reference to a
learned code. The concept is understood within a context and the reader responds
pragmatically.
We can take this approach a step further and consider the semiotic theory of Saus-
sure [1] wherein a sign is a relation between a perceptible token (signifier, referrer) and
a concept (signified, referent) – giving us another useful term: the referent of the token,
i.e. what it actually means in a given context. For example, a pair of numbers (the refer-
rer) may mean a geographical location in one context (one possible referent), yet may
mean a student’s examination and coursework marks in another (a different referent).
Similarly, we can extend our consideration of pragmatics to include stylistics: the
style in which language is written. The same pragmatic response may be stimulated
by a different set of tokens (or the same set arranged with a different syntax) – this
may produce a different emotional response.
This Saussurian view of language as a static representation of meaning can be
contrasted with the view of Bhaktin [2] who considers language to be a dynamic
process whereby a text is interacted with and manipulated, and its meaning con-
structed dynamically. This active and engaged understanding, Bhaktin says, creates
new meanings: “… establishes a series of complex interrelationships, consonances
and dissonances … [and] various different points of view, conceptual horizons …
come to interact with one another” [2].
While Bhaktin’s theory of the dynamic interpretation and negotiation of linguistic
texts was primarily based around the social context of language interpretation and the
construction of ideologies within cultures and institutions, it is a useful complement to
the “language as representation” perspective presented above. We can use this alter-
nate view of “language as a process” in our framework for Information Visualization:
a model of data embodied in a visualization must be explored, manipulated and
adapted within a investigatory process resulting in enhanced understanding of its
meaning. When there are two processing agents in a human-computer interaction
context (the human and the computer), either or both can perform this processing.
Thus, we can relate discussion of theoretical approaches to Information Visualiza-
tion to the concepts of
• Interpretation of a visualization through its external physical form (referents,
and lexical, syntactic, semantic, pragmatic and stylistic structures), an activ-
ity typically performed by a reader;
• Exploration and manipulation of the external representation by the reader so
as to discover more about the underlying model, typically done through in-
teraction facilities provided by a visualization tool; and
• Exploration and manipulation of the internal data model by the system in or-
der to discover interrelationships, trends and patterns, so as to enable them to
be represented appropriately.
48 H.C. Purchase et al.

The three sections that follow each take a different approach to suggesting a theory
for Information Visualization. While they were not originally developed with the
above linguistic model in mind, each can be related in some way to this framework.
Natalia Andrienko takes a data-centric view, focusing on the dataset itself, and the
tokens that describe it. She considers how the characteristics of the dataset and the
requirements of the visualization for a task may be matched to determine patterns,
thus predicting the most appropriate visualization tool for the given task. Thus, this
section describes the exploration of the data model so as to identify the best syntax to
use for given tokens (taking into account their referents and the desired semantics).
She highlights the usefulness of systems which can explore the data model, predict
the patterns in datasets, and facilitate the perception of these patterns.
Matthew Ward’s starting point is communication theory, and this section is clearly
focused on information content – the meaning of the visualization and maintaining the
flow of information through all stages of the visualization pipeline. He discusses how
we may assess our progress in designing and enhancing visualizations through con-
sidering measurements of information transfer, content or loss, thus providing a useful
theoretical means for validating visualizations. In this case, there is no internal explo-
ration of the data, but it is the validity of the data after transfer from internal model to
external representation that is considered important.
T.J. Jankun-Kelly introduces two useful models for a scientific approach to visu-
alization, both of which are in their infancy. The visual exploration model describes
and captures the dynamic process of user exploration and manipulation of visualiza-
tion in order to affect its redesign, thus using the pragmatic response of the user to
determine a new syntactical arrangement. The second model, visual transformation
design, uses transformation functions applied to the data model to provide design
guidance based on visualization parameters, thus performing an initial exploration of
the data model to suggest syntax to enhance the pragmatic response of the user.
The paper concludes with a summary, and suggestions for future research.

2 Predictive Data-Centered Theory

Among other theories, Information Visualization requires a theory that could serve as
a basis for instructing Information Visualization users how to select the right tools for
their data and do data exploration and analysis with the use of these tools. The same
theory could also help tool designers in finding right solutions. The following argu-
mentation is meant to clarify what kind of theory this could be.
Most Information Visualization researchers agree that the primary purpose for us-
ing Information Visualization tools is to explore data in order to gain understanding of
the data and the phenomena behind. Gaining understanding may be thought of as
constructing a concept, or mental model, of the data or phenomenon. A model, in turn,
can be considered as a parsimonious representation capturing essential features of the
data rather than listing all individual data items; this means that a model necessarily
involves abstraction. For example, from observing morning temperatures over several
days, a person may build a concept of the increase or decrease of the temperature.
Such an abstraction is based on a holistic grasp of characteristic features embrac-
ing multiple data items. We shall use the term “pattern” to refer to such features. In-
 Theoretical Foundations of Information Visualization 49

crease and decrease are examples of patterns. A model may be a synthesis of several
patterns each representing some part or aspect of the data. Thus, when the observation
of the morning temperatures is performed over a sufficiently long period, the model
will probably incorporate the patterns of both increase and decrease of the tempera-
ture. Furthermore, patterns may also be composed of sub-patterns. For instance, the
behavior of the temperature may be conceptualized as a repeated “wave” where in-
crease is followed by decrease. Here, increase and decrease are basic, or atomic, pat-
terns, the “wave” is a composite pattern including the increase and decrease patterns,
and the repetition of the “wave” is a pattern of a yet higher level, which incorporates
the “wave” pattern.
The main role of Information Visualization tools can be understood as helping the
user to perceive patterns that could be used for building an appropriate model. This
means, in particular, that a tool should facilitate the perception of (sub)sets of data
items as units. For an appropriate support of the detection of patterns, a tool designer
should know in advance what types of patterns need to be perceived (or otherwise
detected) with the use of the tool. Then, after the tool is ready, it will be easy to ex-
plain to the users the purpose of the tool and instruct them how to detect the types of
patterns the tool is oriented to.
The types of patterns that may be meaningful for the user depend on the structure
and properties of the data under analysis. Thus, in the analysis of a temporal series of
numeric measurements (such as temperatures) it makes sense to look for such basic
patterns as increase, decrease, stability, fluctuation, peak, and low point. However,
when numeric measurements refer to a discrete unordered set as, for example, melting
temperatures of various substances, the possible types of patterns may be groupings of
elements with close values of the measurements and frequency-related patterns:
prevalence of certain values or value intervals, frequent values or exceptional values
(outliers).
To support the designers and users of Information Visualization tools in the way
described above, there is a need for a theory that could enable the possibility to pre-
dict, for a given dataset or a given class of datasets, what types of patterns may be
found there. We specially emphasize the term types to exclude the possible impres-
sion of attempting to predict (and on this basis automatically detect) all specific pat-
terns hidden in specific data. Thus, a prediction that a dataset may contain groups
(clusters) of objects with similar characteristics does not define what specific clusters
are there. However, it orients tool designers, who will know that the tool must help
the users to detect clusters, and users, who will know that they need a tool facilitating
the detection of clusters. Then, if each Information Visualization tool and technique is
supplied with an appropriate signature (i.e. what kind of data it is suitable for and
what types of patterns it is oriented to), the user will be able to choose the right tool.
The theory we are advocating in this section can be called data-centered predictive
theory. The theory needs to include
1. an appropriate generic framework for the characterization of various data
types and structures;
2. a general typology of patterns;
3. a mechanism for deriving possible pattern types from data characterizations.
Here, we present some preliminary ideas concerning these components of the theory.
50 H.C. Purchase et al.

2.1 Data Characterization Framework

Data may be viewed abstractly as a set of records with a common structure, each
record being a sequence of elements (such as numbers or strings) which either reflect
the results of some observations or measurements or specify the context in which the
observations or measurements were obtained. The context may include, for example,
the place and the time of observation or measurement, and the object or group of
objects observed. The elements that a data record consists of are called values.
All records of a dataset are assumed to have a common structure, with each posi-
tion having its specific meaning, which is common to all values appearing in it. These
positions may be named to distinguish between them. The positions are usually called
components of the data.
Definition: Characteristic component, or attribute, is a data component correspond-
ing to a measured or observed property of the phenomenon reflected in the data.
Characteristic is a value of a single attribute or a combination of values of several
dataset attributes.
Definition: Referential component, or referrer, is a data component reflecting an
aspect of the context in which the observations or measurements were made. Refer-
ence is the value of a single referrer or the combination of values of several referrers
that fully specifies the context of some observation(s) or measurement(s).
Definition: Reference set of a dataset is the set of all references occurring in this
dataset.
Definition: Characteristic set of a dataset is the set of all possible characteristics, (i.e.
combinations of values of the dataset attributes).
Definition: Multidimensional dataset is a dataset having two or more referrers. De-
pending on the number of referrers, a dataset may be called one-dimensional, two-
dimensional, three-dimensional, and so on.
For example, the geographical location and the time are referrers for measurements of
properties of the climate such as air temperature or wind direction, which are attrib-
utes. Each combination of location and time is a reference, and the corresponding
combination of air temperature and wind direction is a characteristic. This is a two-
dimensional dataset as it has two referrers; the attributes are not counted as dimen-
sions. Referrers are independent components and attributes are dependent since the
values of attributes depend on the context in which they are observed. In data analy-
sis, it is possible to deal with selected attributes independently from the others; how-
ever, all referrers present in a dataset need to be handled simultaneously.
Data may be viewed formally as a function, in the mathematical sense, with the re-
ferrers being independent variables and the attributes being dependent variables. The
function defines the correspondence between the references and the characteristics
where for each combination of values of the referential components there is at most
one combination of values of the attributes.
The structure of a dataset is characterized by specifying which components it in-
cludes, which of them are referrers, and which ones are attributes. Additionally to
this, it is necessary to specify the properties of the components. The relevant proper-
ties are:
 Theoretical Foundations of Information Visualization 51

• whether distances exist between the elements. Any continuous set such as
time, space, and values of temperature has distances, but there may be dis-
tances also in discrete sets such as a set of integer values denoting numbers of
some items. The discrete set of substances has no distances.
• whether and how the elements are ordered. Thus, time moments are linearly
ordered and may also be cyclically ordered, depending on the time span of
observations.
It should be noted that a set consisting of combinations of values of several compo-
nents does not inherit the properties of the individual components. Thus, a set of
combinations of values of melting temperature and atomic weight is only partly or-
dered although the value sets of the original attributes are fully ordered. This data
characterization framework is presented in more detail in [3].

2.2 Patterns

Definition: Pattern is an artifact that represents some arrangement of characteristics

corresponding to a (sub)set of references in a holistic way, i.e. abstracted from the
individual references and characteristics.

This is a more generic definition than is given in data mining: “a pattern is an ex-
pression E in some language L describing facts in a subset FE of a set of facts F [i.e.
a dataset, in our terms] so that E is simpler than the enumeration of all facts in FE” [4].
In our definition, we mean any kind of representation, for example, graphical or
mental.
We posit that all existing and imaginable patterns may be considered as instan-
tiations of certain archetypes (or, simply, types). It is quite reasonable to assume
that such archetypes may exist in the mind of a data analyst and drive the process of
visual data analysis, which is commonly believed to be based on pattern recogni-
tion: the analyst looks for constructs that can be regarded as instances of the exist-
ing archetypes.
A pattern-instance may be characterized by referring to its type and specifying its
individual properties, in particular, the reference (sub)set on which the pattern is
based. Properties may be type-specific (for example, amount and rate of increase).

2.3 Pattern-by-Data Typology

The following table defines the basic types of patterns in relation to the characteristics
of data for which such pattern types are relevant. We cannot guarantee at the present
moment that this typology is complete; further work is obviously needed. Note that
neither the columns nor the rows of the table are mutually exclusive. Thus, when the
characteristic set is ordered and has distances, the pattern types from all columns are
relevant. Similarly, when the reference set is linearly and cyclically ordered, the pat-
terns from all rows are possible.
These basic pattern types may be included in composite patterns. The types of
composite patterns depend on the properties of the reference set:
52 H.C. Purchase et al.

Characteristic set Any Ordered Has distances

Reference set
Any Even frequencies Tendency toward Groups (clusters) of
of the values, high, low or me- references with
prevailing values, dium values close characteristics
rare values
Linearly ordered Constancy, Increase, de- Gradual change,
change, specific crease, peak, low sharp change
value order point
Cyclically ordered Frequency of Cyclical increase Gradual or sharp
value appearing in and decrease changes within the
certain positions cycle and between
of the cycle cycles
Has distances Homogeneity or Flatness, eleva- Smoothness (small
heterogeneity, tion, depression, differences between
large or small peak, depth, pla- characteristics of
regions of con- teau, valley neighboring refer-
gruency ences), abruptness
(big differences)

1. For any kind of reference set: repeated pattern, frequent pattern, infrequent
pattern, prevailing pattern;
2. For a linearly ordered reference set: specific sequence of patterns, alterna-
tion;
3. For a cyclically ordered reference set: cyclically repeated pattern;
4. For a reference set with distances: constant distance between repetitions of a
pattern, patterns occurring close to each other.
Any composite pattern may, in turn, be included in a bigger composite pattern, for
example, a frequently repeated pattern where increase is followed by decrease.

2.4 Directions of Further Work

What is presented in this section is only an initial sketch of the data-centered predic-
tive theory. Further work is required to ensure the comprehensiveness of the pattern
typology. Particular attention needs to be paid to multi-dimensional data. It is also
necessary to define pattern types used to represent relationships between attributes or
between phenomena (represented by several datasets differing in structure) such as
correlation (co-occurrence) or influence.
Then, it is necessary to evaluate Information Visualization techniques according to
the types of data they are suited for and the types of patterns they help to elicit. This
can form an appropriate basis for instructive books and courses for users of Informa-
tion Visualization tools.
 Theoretical Foundations of Information Visualization 53

3 Information Theory
3.1 Visual Communication

Information visualization can be viewed as a communication channel from a dataset

to the cognitive processing center of the human observer. This suggests that it might
be possible to employ concepts from the theories of data communication as a mecha-
nism for evaluating and improving the effectiveness of information visualization
techniques. While there are several early papers that tried to establish linkages be-
tween HCI in general to information theory [5], it might be time to revisit this concept
in light of all the progress that has been made in information visualization in the past
two decades.
We must start with defining information, as it is the core of information visuali-
zation.
Schneider defined information as “always a measure of the decrease of uncertainty
at a receiver” [6] while Cherry stated “Information can only be received where there
is doubt; and doubt implies the existence of alternatives where choice, selection, or
discrimination is called for” [7]. Measuring information is a topic found in many
fields, including science, engineering, mathematics, psychology, and linguistics. In-
formation theory, which primarily evolved out of the study of hardware communica-
tion channels, defines entropy as the loss of information during transmission; it is also
referred to as a measure of disorder or randomness. Another important term is band-
width, which is a measure of the amount of information that can be delivered over a
communication channel in a given period of time. We will attempt to analyze infor-
mation visualization using this terminology.
Information can be categorized in a number of ways. MacKay [8] identifies three
types of information content:
• Selective information-content: This is information that helps the receiver
make a selection from a set of possibilities, or narrows the range of possibili-
ties. An example might be a symptom that helps make a diagnosis.
• Descriptive information-content: This type of information helps the user
build a mental model. Two types of descriptive information content have
been identified:
– Metrical: this type of observation increases the reliability of a pattern,
e.g., a new member of an existing cluster (sometimes termed a metron).
– Structural: this type of observation adds new features or patterns to a
model/representation, e.g., a new cluster (termed a logon).
• Semantic information-content: This type of information is not covered in
classical information theory. It lies between the physical signals of commu-
nications (called the syntactic) and the users and how they respond to the
signals (called the pragmatics). The pragmatics are the domain of psychol-
ogy, both perceptual and cognitive.
While the first two classes of information content lend themselves well to measure-
ment, it is much harder to determine measures of semantic content, as this in general
is very specific to individuals.
54 H.C. Purchase et al.

3.2 Measuring the Amount of Information

There have been many efforts to date to quantify the amount of information in a
communication stream. If we think of plain text, there are numerous quantifiable
features, including:
– The total number of words per minute
– The occurrence of specific words
– The frequency of occurrence for each word
– The occurrence of word pairs, triples, phrases, and sentences.
There are problems, however, with such simplistic, syntax-only measurement. Words
can have variable significance; some are unnecessary or redundant. Many words can
encode the same concept. In fact, reading text or hearing speech may have no affect
on one’s uncertainty regarding the subject of the text, e.g., you may already have
known it, or you don’t understand the meaning of the words or their implied concepts.
This implies that the measurement of information content or volume can be specific to
the individual receiver and, as we’ll see later, the task that is being performed based
on the communication.
Can we perform similar analysis on a dataset? Consider a table of numeric values.
Features of potential interest in the dataset include:
– The count of number of entries or dimensions
– The values
– Clusters and their attributes (number, size, relations, …)
– Trends and their attributes (size, rate of change, …)
– Outliers and their attributes (number, degree of outlierness, relation to dense
regions, …)
– Associations, correlations and any features between records, dimensions, or
individual values.
In fact, we can observe that a featureless dataset is not differentiable from random
noise: all values are equally likely. Features and relations can also vary in their mag-
nitude, certainty, complexity, and importance. Clusters may differ in size; outliers
may vary in their distance to the main body of data; features may be comprised of
many sub-features; in many cases, a feature that is significant to one observer may be
considered noise by another. Recently, researchers have proposed measuring and
counting insights [9], which are new knowledge gained during visual analysis. These
insights are generally specific to a particular task, some of which include [10]:
– Identify data characteristics
– Locate boundaries, critical points, other features
– Distinguish regions of different characteristics
– Categorize or classify
– Rank based on some order
– Compare to find similarities and differences
– Associate into relations
– Correlate by classifying relations.
For each of these tasks, we might have different accuracy requirements as well, which
can influence the resolution at which feature extraction is accomplished during com-
 Theoretical Foundations of Information Visualization 55

munication. Thus, for example, the tasks of detecting, classifying, and measuring a
particular phenomenon each have their own accuracy demands. The tasks to be per-
formed also have an implication on the types of information that the visualization
must be able to convey; categorization and ranking imply that the visualization must
have high selective information content, while identifying characteristics and bounda-
ries are part of building a mental model and thus require good descriptive information
content.
Returning to our dataset and the simplistic features and relations that are contained
in it, we can try to quantify the volume of information and then measure how much of
this volume a visualization technique is capable of effectively conveying. If we as-
sume a table of scalar values (M records, N dimensions or variables), the number of
individual values to be communicated is M*N, and the maximum resolution required
is the number of significant digits. Often, however, the available visual resolution is
far less than that of the data. We can then count all the pairwise relations between
records, or dimensions, or even values. For records, this would be M*(M−1)/2, and
similar for dimensions and values. Then there are relations that are 3-way, 4-way, or
even among an arbitrary number of elements, e.g., in clustering tasks. Clearly, there
are too many possibilities to consider them all, so perhaps we need a different tactic.

3.3 Measuring Information Loss

Perhaps it is easier to measure loss of information (entropy) during the visualization

process than the total information content of a dataset. There are several techniques in
common use for data transformation for visualization that provide an implicit measure
of information loss. For example, multidimensional scaling, a process commonly used
for dimensionality reduction, provides a measure of stress, which is the difference
between the distances between points in the original dimensioned space and the corre-
sponding distances in the reduced dimension space. Similarly, when using principal
component analysis for performing this reduction, the loss can be measured from the
dropped components. Cui et al. [11] developed measures of representativeness when
using processes such as sampling and clustering to reduce the number of data records
in the visualization. These measures, based on nearest neighbor computations, histo-
gram comparisons, and statistical properties, give the analysts control over what was
termed abstraction quality, so they are aware of the trade-offs between speed of ren-
dering, display clutter, and information loss. They, however, did not consider the
perceptual issues, which are very dependent on the particular visual encoding used.
Distortion techniques such as lens effects and occlusion reduction also provide the
analyst with trade-offs between accuracy and visual clarity. Each results in a trans-
formation (typically of an object’s position on the screen) that is meant to improve the
local interpretability at the cost of accuracy of global relations. It would be interesting
to see measures of these competing processes to gauge the overall implications.
Another transformation that can impact on the information being communicated in
a visualization is the ordering of records and dimensions. Ordering can reveal trends,
associations, and other types of relations, and is useful for many tasks. There are many
possible orderings of a table of M records and N dimensions. The key is to determine
which are the most useful. An ordering can convey many pairwise relations. If there
are M records, an ordering can communicate M-1 of the M*(M−1)/2 possible pair-
56 H.C. Purchase et al.

wise relations. Many researchers have studied ways of selecting a useful ordering,
including Bertin’s reorderable matrix [12], Seo and Shneiderman’s rank-by-feature
techniques [13], and Peng et al.’s reordering for visual clutter reduction [14]. In all
cases, a user should be able to prune orderings to emphasize those that show trends,
groupings, or other discernable patterns. Thus far most research has focused on sim-
ple 1-D orderings, but higher level orderings and structures (e.g., hierarchies) have
also been studied.

3.4 Hardware and Perceptual Limitations

This discussion of information content would not be complete without also consider-
ing the limitations imposed by the visual communication channel (i.e., the display)
and receiver (the human visual perception system). Regarding the channel and its
capacity, modern displays are limited to somewhere on the order of one to nine mil-
lion pixels, although tiled displays can increase this substantially. The color palette
generally has a size of 224 possible values, although the limitations of human color
perception take a big chunk out of this. Finally, the refresh rate of the system, typi-
cally between 20 and 30 frames per second, limits how fast the values on the screen
change, though again the human limitations of change detection mean that much of
this capability is moot.
Regarding these human limitations, from the study of human physiology we know
that there are approximately 800k fibers in the optic nerve. We can perceive 8-9 levels
of intensity graduation, and require a 0.1 second accumulation period to register a
visual stimulus. In addition, we have a limited viewable area at any particular time,
and a variable density of receptors (much less dense in the peripheral vision). Studies
have shown we have a limited ability to distinguish and measure size, position, and
color, and the duration of exposure affects our capacity. Finally, it has been shown
that our abilities are also related to the task at hand; we are much better at relative
judgment tasks than absolute judgment ones.

3.5 Measuring Information Content on Visualizations

We now look at methods that have been used in the past for measuring the informa-
tion content in a data or information visualization. For completeness sake, some of
these are quite trivial. For example, simply counting the number of data values shown
is a valid measure. The issue in this case would be how to deal with partial occlusion.
In some cases this would be acceptable if sufficient information remains visible to
make identification or recognition possible. Tufte [15] suggested the data-ink ratio as
an indicator of information content, though tick marks, labels, and axes are often
essential for appropriate identification. Many researchers have used counts of the
number of features or patterns found in a particular amount of time. Ward and Ther-
oux [16] counted the number of clusters and outliers found by users in different visu-
alizations, while Suraiya et al. [9] counted insights discovered. In each case, a ground
truth is needed to verify that what was found was really present. Similar experiments
have been used to measure classification, measurement, and recall accuracy.
There are many other issues when attempting to measure information in a visuali-
zation. As mentioned earlier, distortion and other transformations can improve the
 Theoretical Foundations of Information Visualization 57

readability of a visualization, but introduce errors in the data themselves. Data may
have uncertainty attributes associated with them, which can interfere with the meas-
urement. On the other hand, there are numerous examples of improving information
content by using novel layout, shape, and color strategies or augmenting the visualiza-
tion with links, boundaries, and even white space. The amount of information con-
tained may also be enhanced using redundant mappings, which improves the chances
of successful reception by the viewer. Finally, the use of animation to communicate
information in an incremental fashion can be quite effective; it is lossy communica-
tion, as viewers quickly forget some of what they have seen, but the ability to replay
the animation can replace some of this lost information.

3.6 Case Study: Parallel Coordinates

As an example of this information measurement activity, let us consider parallel coor-

dinates, a popular multivariate visualization technique. The first question is how well
does this technique present the values of a dataset? For individual values, this method
has very high resolution, given most of the screen height can be used to convey the
value. This implies the technique possesses high selective information content, at least
on individual dimensions, as separation into sub-ranges is facilitated by the amount of
screen space allocated to convey values. However, the loss due to occlusion can be
high, especially for nominal variables. This loss is somewhat mitigated by the varying
slopes of lines ending/starting at axes, which allows some degree of differentiation.
Relationships among data along an axis can be emphasized via embedded histograms,
as found in some implementations of parallel coordinates. In terms of relationships
between dimensions, this method is limited to showing pairwise relations, with N-1
out of the N*(N-1)/2 possible relations shown. Automated dimension ordering can
help reveal interesting relationships, as seen in [14]. Relationships between records
are problematic due to the ambiguous continuity of records that intersect on one or
more axes. Coloring the lines based on a record ID can help with modest sized data-
sets. We can also use animation along an axis or based on an order to expose some
inter-record relations. Intersections and near-parallel edges can reveal partial struc-
tures (between dimension pairs), and techniques such as Hierarchical Parallel Coordi-
nates [17] can show grouping relations, though there is loss of individual data values.
Each of these methods enhances the descriptive information content of the visualiza-
tion, thus helping analysts form mental models of the data.
Through analyzing these augmentations of parallel coordinates we see that many
recent innovations to parallel coordinates target different types of information loss
resulting from this means of mapping data. For example, we see efforts to preserve
and emphasize outliers in a paper by Novotny and Hauser [18]. However, many other
issues still exist; there are still many data features that cannot be readily perceived and
tasks that are difficult to perform using parallel coordinates.

3.7 Conclusions

To summarize, we feel there are many aspects of information visualization research

that can find analogies in the concepts of information theory - it is all about commu-
nication. Perhaps finding such a formal structure on which to ground our efforts can
58 H.C. Purchase et al.

potentially reduce the amount of ad hocness in the field. The key is to define meas-
ures of information transfer, content, or loss at all stages of the pipeline as a means of
assessing our progress in the development of new visualization techniques and en-
hancement of existing ones.

4 Formal Models for a Science of Information Visualization

Information visualization utilizes computer graphics and interaction to assist humans
in solving problems. As such, it incorporates elements of a constructive, formal sci-
ence (the algorithmic and development aspects) with aspects of an empirical science
(for measuring effectiveness and validity). Surrounding both are engineering efforts to
improve the overall system. This section discusses the relationship between these
three parts of the visualization discipline, and suggests that a deeper exploration of
formal, scientific models is needed for a strengthening of the field.

4.1 The Need for Models

Traditionally, visualization has focused on the engineering aspects while importing

“scientific” elements as needed. However, even this borrowing has not been suffi-
ciently utilized. To illustrate this, consider archetypical topics from an information
visualization course syllabus:
• Exploration: The process of visual exploration in a larger context
• Perception: Fundamental mechanisms for human visual perception
• Visual Cognition: How perception translates to thought and action
• Color: Aspects of color for visualization
• Techniques: Specific visualization metaphors, including interaction
• Evaluation: Measuring the effectiveness of a visualization design
A survey of visualization education programs has found that most such programs
focus on visualization techniques (the engineering core) in detriment to the founda-
tional aspects (the scientific core) [19]. As a further example, consider that rainbow
colormaps are still entrenched in visualization research [20] while ample scientific
evidence demonstrates their muddling effect [20,21,22]. The ease of utility for provid-
ing rainbow colormaps does not outweigh the costs in terms of a user's time - the
primary currency of users [23].
Examining the previous list, it is apparent that only the Techniques topic deals ex-
tensively with the engineering aspects of visualization design. While these efforts are
vital to providing actual tools to users, the other elements are needed to provide a
solid foundation to guide those efforts. For example, perceptual literature is grounded
in empirical results with a strong scientific pedigree. A key aspect of these results is
the formal models which are generated to explain the results. Such models are both
descriptive - they encapsulate the factors of the empirical experiment and describe a
mechanism for their operation - and predictive - they generalize the description to a
larger context by predicting future behavior. The predictive nature of the models fa-
cilitates visualization design: it is the predictive nature of color perception models that
explains the limitations of rainbow colormaps [22].
 Theoretical Foundations of Information Visualization 59

Fig. 1. Topics in a Formalized Information Visualization Course. Dark grey topics are based
upon formal foundations in other disciplines; light grey topics are yet-to-be-developed visuali-
zation-specific formal foundations.

4.2 A Move to Visualization Formalisms: The Two Models

There have been several recent calls for an establishment of a “theory” and “science”
behind visualization [24,25]; this need can be partially addressed via formal scientific
models. If we accept that information visualization needs a formal foundation, the
question remains whether the existing models from perceptual psychology and cogni-
tive science are sufficient. The problem with these formalisms is they do not address
the specific problems of visualization. While they provide general guidelines, models
from non-visualization fields do not consider the context of the visualization envi-
ronment - the user and the computer. What is needed is a set of formal foundations
that bridges the gap between the general human experience and the visualization do-
main (Figure 1). We propose two models for this purpose: an exploration model that
incorporates the user's interaction with the visualization and the dynamic aspects of
their analysis, and a transform design model which encapsulates the depiction and
constructive aspects of the visualization. These models would abstract fundamental
principles of visualization science and design, and thus proscribe (via their predictive
power) empirically driven practices.

4.3 Visualization Exploration Model

Visualization exploration is a goal-driven task incorporating visual search and infor-

mation seeking. It is an iterative process - a user creates a visualization result, evalu-
ates its worth, and then manipulates the visual parameters (e.g. color maps, selection
regions) creating new results until satisfied. Thus, any formal model of computer-
mediated visual exploration must capture the cognitive operations and how those
realized actions manipulate the visualization. Cognitive operations are the domain of
cognitive science, and several methods exist to model the human analysis process
[26]. To bridge this work to visualization, two additional levels are needed: first, a
description of the visual information search process and how it affects human cogni-
tion; second, a model of how the visualization session evolves due to human interac-
tion. Visual sensemaking models such as the information foraging work of Pirolli and
Card [27] begin to address the first need. Formalisms that capture the range of human-
visualization interactions are targeted at the second [28,29,30].
60 H.C. Purchase et al.

Original Visualization Session Graph Refined Visualization

Fig. 2. Analysis and evolution of a network traffic visualization. The original interface (a) uses
colored lines connected to the edges of a square to depict changes. A formal model was used to
capture interaction with the tool, and these sessions were analyzed to improve the interface (b).
The redesigned interface (c) makes the exploration more efficient by displaying all event types
individually and combined.

There are several benefits to a complete visualization exploration model. An under-

standing of how humans process visual cues in order to make exploration decisions
can inform visualization design. For example, this “information scent” has been used
to understand the cost-benefit trade-offs of different focus+context visualizations and
to formally understand efficient web interfaces [31,32]. Similarly, a formalism for the
visualization process lends itself to analysis to measure the user's efficiency of explo-
ration [33,34]. Clusters of similar results (based upon metrics) during the session
suggest redundant exploration; analysis of sessions based upon these metrics illumi-
nate the path to more effective design [29,35] (Figure 2).

4.4 Visualization Transform Design Model

An exploration model describes and predicts a human's interaction with a visualiza-

tion system based upon its design. This model neglects to describe the components
that compose the design or provide initial design guidance. To provide this guidance,
visualization transform design models are needed. A visualization transform is the func-
tion that computes the depicted result from visualization parameters – elements such as
brushed graph nodes or opacity maps that dictate the rendered result. Significant work
has expressed different mechanisms for constructing such transforms [36,37,38,39,40]
(see Figure 3 for an extended example), but these categorizing efforts lack two things
to provide a formal foundation. First, they do not utilize perceptual and cognitive
literature to suggest and evaluate design decisions; second, most do not address the
evolution of transforms and the utilization of extra-visualization tools important to the
analysis (e.g., statistical packages). Efforts in providing design guidance for data
display has been investigated [41,42,43] and formal algebras for transform modifica-
tion have been recently presented [44,45]. However, these distinct contributions re-
quire effort to unify and to validate for a complete and cohesive scientific foundation.
A complete, predictive transform design model will yield several benefits. Toolkits
for visualization creation benefit by providing guidance on suitable, less suitable, and
unsuitable component choices. Visualization pedagogy will improve due to a vali-
dated foundation for techniques. Further, formal models will lead to objective metrics
for evaluating a transform's effectiveness. All of these enhancements feed back into a
visualization system, improving its potential utilization.
 Theoretical Foundations of Information Visualization 61

(a) Routing (b) Visualization Transform Components

Visualization

Fig. 3. A depiction of the revised network routing visualization transform. Nodes represent the
state of the data (e.g., a table of events) while edges represent operators or interactions (e.g.,
parsing the data). In this example, the network visualization is combined with a graph visualiza-
tion by embedding the results of the former within the latter. The graph itself is a composition,
merging a spanning tree and the original graph to layout the selected sub-graph. Back-
propagation of state due to interaction is included. The depiction is based upon an extended
Data State transform model.

4.5 The Value of Visualization Craft

A research program investigating scientific grounding for visualization is not meant to

diminish the importance of the engineering component of visualization. Visualization
is a tool for humans; engineering efforts form the basis of providing such tools. For-
mal exploration and design models can guide the creation of a visualization system;
however, as is the case now, multiple comparable techniques will often solve the same
problem. Thus, the craft of visualization - the confidence in design choices gained
through experience - will still be needed to decide between the choices a formal
model provides. Inspiration and creativity will not be eliminated by a more rigorous
foundation; the foundation will serve as a springboard for such endeavours.

4.6 What Is Left to Be Done

Formal foundations for a science of information visualization are still in a nascent

stage. Elements of complete exploration and transform models exist; however, they
have neither been reconciled with each other nor validated for their correctness. A
close collaboration between perceptual psychologists, cognitive scientists, visualiza-
tion researchers, and practitioners is needed to drive research into foundations: per-
ceptual and cognitive scientists provide the human-based foundations, practitioners
provide the case studies for observation and validation of models, and visualization
researchers will form the domain-specific bridge between them.
Humans and computers play an integrated role in the development and utilization
of visualization. A formal foundation would measure the efficiency of the former, and
guide the design of the latter in order to create a more effective whole.
62 H.C. Purchase et al.

5 Conclusion
The three discussions of Information Visualization presented here draw on existing
theories of data-centric prediction, information communication and scientific model-
ing, and relate in different ways to the linguistic framework defined in the introduc-
tion. A single uniting theory of Information Visualization may be impossible due to
its strong relationship to and use of several other diverse disciplines (e.g. psychology
(perception, cognition and learning), graphic design and aesthetics).
Investigating theoretical approaches used in other disciplines, and their relation to
Information Visualization, is an obvious way forward, and can provide a useful way
for researchers in the area to present, discuss and validate their ideas; it is hoped that
the over-arching linguistics-based framework of representation, user exploration and
manipulation, and system exploration and manipulation will prove useful in linking
the constituent theories together. The more solid theoretical analyses that Information
Visualization researchers or tool designers can call on in defending or validating their
work, the more secure the discipline will be.

References
1. de Saussure, F.: Writings in General Linguistics. In: Bouquet, S., Engler, R., Sanders, C.,
Pires, M. (eds.), Oxford University Press, Oxford (2006)
2. Bakhtin, M.: The Dialogic Imagination, University of Texas Press (1981), quoted in Ball,
A.F., Freedman, S.W.: Bhaktinian Persepectives on Language, Literacy, and Learning,
Cambridge University Press, Cambridge (2004)
3. Andrienko, N., Andrienko, G.: Exploratory Analysis of Spatial and Temporal Data: A Sys-
tematic Approach. Springer, Heidelberg (2006)
4. Fayyad, U., Piatetsky-Shapiro, G., Smyth, P.: From data mining to knowledge discovery in
databases. AI Magazine 17, 37–54 (1996)
5. Card, S.K., Moran, T.P., Newell, A.: The Psychology of Human-Computer Interaction.
Erlbaum Associates, Hillsdale (1983)
6. Schneider, T.D.: Information Theory Primer. http://www.lecb.ncifcrf.gov/
~toms/paper/primer (April 14, 2007)
7. Cherry, C.: On Human Communication, 2nd edn. MIT Press, Cambridge (1966)
8. MacKay, D.: Information, Mechanism and Meaning. MIT Press, Cambridge (1969)
9. Saraiya, P., North, C., Duka, K.: An evaluation of microarray visualization tools for bio-
logical insight. In: Proc. IEEE Symposium on Information Visualization, pp. 1–8 (2004)
10. Keller, P., Keller, M.: Visual cues: Practical Data Visualization. IEEE Computer Society
Press, Los Alamitos (1993)
11. Cui, Q., Ward, M., Rundensteiner, E., Yang, J.: Measuring data abstraction quality in
multiresolution visualization. In: Proc. IEEE Symposium on Information Visualization, pp.
709–716 (2006)
12. Bertin, J.: Matrix theory of graphics. Information Design 10(1), 5–19 (2001)
13. Seo, J., Shneiderman, B.: A rank-by-feature framework for interactive exploration of multi-
dimensional data. In: Proc. IEEE Symposium on Information Visualization, pp. 96–113
(2005)
14. Peng, W., Ward, M., Rundensteiner, E.: Clutter reduction in multi-dimensional data visu-
alization using dimension reordering. In: Proc. IEEE Symposium on Information Visuali-
zation, pp. 89–96 (2004)
 Theoretical Foundations of Information Visualization 63

15. Tufte, E.: The Visual Display of Quantitative Information. Computer Graphics Press, Chesh-
ire (1983)
16. Ward, M., Theroux, K.: Perceptual benchmarking for multivariate data visualization. In:
Proc. Dagstuhl Seminar on Scientific Visualization, pp. 314–328 (1997)
17. Fua, Y.-H., Ward, M., Rundensteiner, E.: Hierarchical parallel coordinates for exploration
of large datasets. In: Proc. IEEE Conference on Visualization, pp. 43–50 (1999)
18. Novotny, M., Hauser, H.: Outlier-preserving focus+context visualization in parallel coor-
dinates. IEEE Trans. Visualization and Computer Graphics 12, 893–900 (2006)
19. Dommik, G.: Do We Need Formal Education in Visualization? IEEE Computer Graphics
and Applications 20(4), 16–19 (2000)
20. Borland, D., Taylor, R.M.: Rainbow Color Map (Still) Considered Harmful. IEEE Com-
puter Graphics and Applications 27(2), 14–17 (2007)
21. Brewer, C.A.: Color use guidelines for data representation. In: Proceedings of the Section
on Statistical Graphics, American Statistical Association, pp. 50–60 (1999)
22. Ware, C.: Information Visualization: Perception for Design. Morgan Kaufmann, San Fran-
cisco (2004)
23. van Wijk, J.J.: The Value of Visualization. In: Proceedings of IEEE Visualization, pp. 79–
86. IEEE Computer Society Press, Los Alamitos (2005)
24. Johnson, C., Moorehead, R., Munzner, T., Pfister, H., Rheingans, P., Yoo, T.S.: NIH-NSF
Visualization Research Challenges Report, 1st edn. IEEE Computer Society Press, Los
Alamitos (2006)
25. Thomas, J.J., Cook, K.A.: Illuminating the Path: The Research and Development Agenda
for Visual Analytics. IEEE Computer Society Press, Los Alamitos (2005)
26. Anderson, J.R.: Cognitive Psychology and its Implications, 6th edn. Worth (2005)
27. Pirolli, P., Card, S.K.: Information Foraging. Psychological Review 4, 643–674 (1999)
28. Brodlie, K., Poon, A., Wright, H., Brankin, L., Banecki, G., Gray, A.: Problem Solving
Environment Integrating Computation And Visualization. In: Nielson, G.M., Bergeron,
R.D. (eds.) Proceedings of the 4th IEEE Conference on Visualization, pp. 102–109 (1993)
29. Jankun-Kelly, T.J., Ma, K.-L., Gertz, M.: A Model and Framework for Visualization Ex-
ploration. IEEE Transactions on Visualization and Computer Graphics 13, 357–369 (2007)
30. Lee, J.P., Grinstein, G.G.: An Architecture For Retaining And Analyzing Visual Explora-
tions Of Databases. In: Nielson, G.M., Silver, D. (eds.) Proceedings of the 6th IEEE Con-
ference on Visualization, pp. 101–108 (1995)
31. Pirolli, P., Card, S.K., Van Der Wege, M.M.: Visual information foraging in a focus +
context visualization. In: Proceedings of the SIGCHI conference on Human factors in
computing systems, pp. 506–513. ACM Press, New York (2001)
32. Pirolli, P.: Rational Analyses of Information Foraging on the Web. Cognitive Sci-
ence 29(3), 343–373 (2005)
33. Jankun-Kelly, T.J., Ma, K.-L., Gertz, M.: A Model for the Visualization Exploration Proc-
ess. In: Moorhead, R.J., Gross, M., Joy, K.I. (eds.) Proceedings of the the 13th IEEE Con-
ference on Visualization (Vis ’02), pp. 323–330 (2002)
34. Lee, P.J.: A Systems and Process Model for Data Exploration. PhD thesis. University of
Massachuesetts, Lowell (1998)
35. Teoh, S.T., Jankun-Kelly, T.J., Ma, K.-L., Wu, S.F.: Visual Data Analysis for Detecting
Flaws and Intruders in Computer Network Systems. In: IEEE Computer Graphics and Ap-
plications, p. 24. IEEE Computer Society Press, Los Alamitos (2004)
36. Abram, G., Treinish, L.: An Extended Data-Flow Architecture For Data Analysis And Visu-
alization. In: Nielson, G.M., Silver, D. (eds.) Proceedings of the IEEE Conference on Visu-
alization 1995 (Vis ’95), pp. 263–270. IEEE Computer Society Press, Los Alamitos (1995)
37. Chi, E.H., Riedl, J.T.: An Operator Interaction Framework For Visualization Systems. In:
Dill, J., Wills, G. (eds.) Proceedings of the IEEE Symposium on Information Visualiza-
tion, pp. 63–70 (1998)
64 H.C. Purchase et al.

38. Haber, R.B., McNabb, D.A.: Visualization Idioms: A Conceptual Model for Scientific
Visualization Systems. In: Nielson, G.M., Shriver, B., Rosenblum, L. (eds.) Visualization
in Scientific Computing, pp. 74–93. IEEE Computer Society Press, Los Alamitos (1990)
39. Hibbard, W.L., Dyer, C.R., Paul, B.E.: A Lattice Model for Data Display. In: Bergeron,
R.D., Kaufman, A.E. (eds.) Proceedings of the 5th IEEE Conference on Visualization (Vis
’94), pp. 310–317 (1994)
40. Schroeder, W.J., Martin, K.M., Lorensen, W.E.: The Design and Implementation of an
Object-Oriented Toolkit for 3D Graphics and Visualization. In: Yagel, R., Nielson, G.M.
(eds.) Proceedings of the 7th IEEE Conference on Visualization, pp. 93–100 (1996)
41. Casner, S.M.: Task-analytic approach to the automated design of graphic presentations.
ACM Transactions on Graphics 10(2), 111–151 (1991)
42. Mackinlay, M.: Automating the Design of Graphical Presentations of Relational Informa-
tion. ACM Transactions on Graphics 5(2), 110–141 (1986)
43. Roth, S.F., Mattis, J.: Data Characterization for Intelligent Graphics Presentation. In:
Proceedings on Human Factors in Computing Systems (CHI’90), pp. 193–200 (1990)
44. Bavoli, L., Callahan, S.P., Crossno, P.J., Freire, J., Scheidegger, C.E., Silva, C.T., Vo, H.T.:
VisTrails: Enabling Interactive Multiple-View Visualizations. In: Proceedings of the 16th
IEEE Conference on Visualization (2005)
45. Weaver, C.: Building Highly-Coordinated Visualizations in Improvise. In: Proceedings
2004 IEEE Symposium on Information Visualization, pp. 159–166. IEEE Computer Soci-
ety Press, Los Alamitos (2004)
 Teaching Information Visualization

Andreas Kerren1 , John T. Stasko2 , and Jason Dykes3

1
School of Mathematics and Systems Engineering, Växjö University,
Vejdes Plats 7, SE-351 95 Växjö, Sweden,
[email protected]
2
School of Interactive Computing & GVU Center, Georgia Institute of Technology,
85 5th St., NW, Atlanta, GA 30332-0760, USA,
[email protected]
3
Department of Information Science, City University,
Northampton Square, London EC1V 0HB, UK,
[email protected]

Abstract. Teaching InfoVis is a challenge because it is a new and grow-

ing field. This paper describes the results of a teaching survey based on
the information given by the attendees of Dagstuhl Seminar 07221. It
covers several aspects of offered InfoVis courses that range from differ-
ent kinds of study materials to practical exercises. We have reproduced
the discussion during the seminar and added our own experiences. We
hope that this paper can serve as an interesting and helpful source for
current and future InfoVis teachers.

1 Introduction
Education is an important aspect of any emerging and rapidly evolving disci-
pline and this is certainly the case in Information Visualization (InfoVis) with its
emphasis on the exploratory development of knowledge. Most of the researchers
participating in the Dagstuhl seminar and contributing to this volume are in-
volved in helping students graduate with competencies in visualization. The
growing number of courses in Information Visualization is matched by the vari-
ety of styles of courses offered, in terms of course content, materials used, and
evaluation methodologies. Attendees at Dagstuhl seminar were curious to learn
about the courses others offered and the approaches and resources that were be-
ing used, and so a session on Information Visualization teaching and education
was held.
To prepare for that session and benchmark current offerings, Keith Andrews
from Graz University, Austria, prepared a survey about InfoVis-related courses
and distributed it to the attendees. The survey was intended to gather a va-
riety of information, mostly demographic, including teaching styles, textbooks,
enrollments, teaching aids, examinations, etc. Nineteen participants completed
the survey and described their courses. This paper presents the survey results
and includes the perspectives of some of the participants in relation to their
own teaching experience in light of these and discussions amongst colleagues at
Dagstuhl.

A. Kerren et al. (Eds.): Information Visualization, LNCS 4950, pp. 65–91, 2008.


c Springer-Verlag Berlin Heidelberg 2008
66 A. Kerren, J.T. Stasko, and J. Dykes

The survey consisted of four different parts. The specific questions included
in each part are listed below.

1. General Information
(a) Instructor name
(b) Educational organization
(c) Title of course
(d) Course home page (URL)
(e) Last taught (date)
(f) Course level (graduate or undergraduate)
(g) Course hours per week
(h) Course number of weeks
(i) Enrollment (number of students)
2. Teaching Aids
(a) Do you use one or more textbooks (yes, no)?
i. If so, which ones?
(b) Do you assign papers for compulsory assigned reading (yes, no)?
i. If so, which ones?
(c) Do you have your own set of lecture notes (yes, no)?
i. URL (if available)?
(d) Do you have teaching assistants for the course (yes, no)?
i. If so, how many?
3. Practical Exercises (Projects)
(a) Do you use practical exercises or projects (yes, no)?
i. If so, please describe a typical exercise or project.
ii. If so, how do you grade the practical exercises or project?
4. Examination or Test
(a) Do you have an examination (yes, no)?
i. If so, written or oral exam?
ii. If so, please describe a typical exam question.

Firstly, we briefly review the results of the survey. Section 3 summarizes the
topics discussed during the interactive session on teaching at the seminar. Fi-
nally a selection of participants reflect upon how these issues relate to their own
experience of teaching Information Visualization in Section 4.

2 Results

We present the results in four sections, one for each of the sections of the survey.
 Teaching Information Visualization 67

2.1 General Information

The first part of the survey gives an overview of the courses offered at the dif-
ferent universities represented by the Dagstuhl participants and provides some
details about the courses themselves. Table 1 shows the responses obtained from
this part. A balance of European and North-American universities were repre-
sented by participants in the survey results. The majority of courses were focused
on the “core field” of information visualization (about 68%). Two courses were
about visualization/computer graphics in general, and the rest were about appli-
cation fields (e.g. geographic visualization) or broader topics, such as information
interfaces or visual communication. This scope reflects the broad and interdisci-
plinary nature of Information Visualization and provides some indications as to
why developing an agreed Information Visualization curriculum may be difficult.
Most of the courses (79%) had their own publicly accessible web page provid-
ing access to course related information. Nearly all the referenced courses were
given in 2006 and 2007. Since all the responding instructors are active researchers
in the field as well, we can assume that all these courses covered the current state
of the art in information visualization. Because the detailed curriculum for the
courses was not part of the survey, we do not have details about actual course
content. The web pages associated with each of the courses are a rich source of
information however and we used these to gather keywords associated with the
curricula of each. Figure 1 shows a tag cloud generated from these keywords that
gives a flavor of the variety and importance of different topics across the courses.
The dominant words reflect some of the tensions in Information Visualization
education, with a collective need to focus on data—its dimensionality and struc-
ture, techniques for layout and visual encoding and people and their responses
to these methods and the systems through which they are accessed. Perhaps
the tag cloud and the varied responses suggest a need for systematic research to
learn about the range of approaches that are used in teaching Information Vi-
sualization and related topics. The session discussion, summarized in Section 3,
led to more insight about this, but it was not a comprehensive examination.
Most courses were taught at the graduate level, with only two being un-
dergraduate courses. At the bottom of Table 1, descriptive statistics about the
results of questions Q1g-Q1i on course duration and size are provided. The av-
erage duration of a course and the number of hours of weekly meeting time are
relatively consistent across the group. Note that most class sizes are relatively
small, echoing the fact that Information Visualization is still a relatively new
and growing area. Here, the undergraduate course #16 seems to be an outlier
because of its large enrollment. However, this course is a compulsory course
on computer graphics and visualization, and the instructor plans to divide this
course into two parts in the future.

2.2 Study Materials and Teaching Aids

Textbooks (Q2a): About 72% of all the instructors (13 in total) used one or
more textbooks in their courses. The most popular books were those by Colin
 Table 1. Results of Part 1 of the Survey on General Information together with a brief descriptive Analysis.
# Q1a Q1b Q1c Q1d Q1e Q1f Q1g Q1h Q1i
1 Keith Andrews TU Graz, Austria Information Visualisation [4] Summer 2007 g 3 14 15
2 Jason Dykes City University London, UK GeoVisualization – Spring 2007 g 3 12 22
3 Achim Ebert TU Kaiserslautern, Germany Information Visualization – Winter 04/05 g 2 14 18
4 Helwig Hauser TU Vienna, Austria Information Visualization [23] Summer 2007 g 3 15 30
5 Jeffrey Heer UC Berkeley, USA Visualization [25] Spring 2006 g 3 16 20
6 T.J. Jankun-Kelly Mississippi State University, USA Information Visualization [28] Fall 2006 g 3 15 12
7 Daniel Keim University of Constance, Germany Information Visualization [29] Summer 2007 g 5 14 20
8 Andreas Kerren TU Kaiserslautern, Germany Information Visualization [34] Winter 06/07 g 2 14 9
9 Robert Kosara UNC Charlotte, USA Visual Communication in Computer [39] Spring 2007 g 3 18 15
Graphics and Art
A. Kerren, J.T. Stasko, and J. Dykes

10 Kwan-Liu Ma UC Davis, USA Information Visualization [43] Winter 2006 g 3 10 12

11 Kwan-Liu Ma UC Davis, USA Information Interfaces [44] Spring 2007 ug 3 10 17
12 Guy Melançon University of Bordeaux, France Models and Algorithms for Informa- – Fall 2006 g 3 11 15
tion Visualization and Bioinformatics
13 Silvia Miksch TU Vienna, Austria Information Visualization [48] Winter 2006 g 3 14 30
14 Tamara Munzner UBC Vancouver, Canada Information Visualization [51] Fall 2006 g 3 13 15
15 Chris North Virginia Tech, USA Information Visualization [53] Spring 2007 g 3 15 25
16 Jonathan Roberts University of Kent, UK Computer Graphics and Visualisation – Summer 2005 ug 2 12 60
17 John Stasko Georgia Tech, USA Information Visualization [64] Spring 2006 g 3 15 35
18 Matt Ward WPI, USA Data Visualization [70] Spring 2006 g 3 14 10
19 Jing Yang UNC Charlotte, USA Information Visualization [72] Spring 2007 g 3 11 9
Arithmetic Mean 3 14 20
Median 3 14 17
Standard Deviation 0.6 2.1 12.1
68
 Teaching Information Visualization 69

Fig. 1. Tag cloud of course topics. Includes courses with Web pages in the English
language for which a URL was provided.

Ware and Robert Spence. The following list shows all books used by instructors
of InfoVis courses in descending order of popularity.

1. Information Visualization: Perception for Design. Colin Ware [71].

2. Information Visualization: Design for Interaction. Robert Spence [62].
3. Readings in Information Visualization: Using Vision to Think. Stuart Card,
Jock Mackinlay, and Ben Shneiderman (Eds.) [8].
4. Envisioning Information. Edward Tufte [67].
5. Visualisierung – Grundlagen und allgemeine Methoden, Heidrun Schumann
and Wolfgang Müller [58] (in German).
6. – Human-Centered Visualization Environments. Andreas Kerren, Achim
Ebert, and Jörg Meyer (Eds.) [35].
– Information Visualization: Beyond the Horizon. Chaomei Chen [9].

Some respondents noted that they also used Tufte’s other books [69, 68] for
specific aspects of the course or as a focused topic, rather than as a general
textbook. Other courses cover more general fields, such as (Data) Visualization
(#5, #18, . . . ), with information visualization as part of them. In these courses,
other textbooks were used, for example the VTK Book [57], Designing Visual
Interfaces by Mullet and Sano [50], or [58, 30, 42, 16, 18]. Those teaching visual-
ization in a particular domain (e.g. GeoVisualization (#2)) used more specific
texts associated with the relevant discipline [60,15,49]. Figure 2 shows the usage
of books for all courses listed in Table 1.

Papers for Compulsory Assigned Reading (Q2b): 68% of the courses

used research papers for compulsory assigned reading. Many different papers
were used so an exhaustive listing here is not appropriate. Most papers were
70 A. Kerren, J.T. Stasko, and J. Dykes

Kerren et al. (2007) [35] 1

Spence (2007) [62] 10
Chen (2006) [9] 2
Schroeder et al. (2006) [57] 2
Dykes et al. (2005) [15] 1
Few (2004) [18] 1
Slocum et al. (2004) [60] 2
Publication Date

Ware (2004) [71] 10

Fayyad et al. (2001) [16] 1
Tufte (2001) [69] 3
Schumann et al. (2000) [58] 2
Card et al. (1999) [8] 4
Lichtenbelt et al. (1998) [42] 1
Tufte (1997) [68] 1
Monmonier (1996) [49] 1
Mullet et al. (1994) [50] 1
Keller et al. (1993) [30] 1
Tufte (1990) [67] 5

Absolute Usage (13 instructors in total)

Fig. 2. Usage of books for all courses ordered by publication date.

selected from the IEEE InfoVis and Vis Conference proceedings as well as from
the ACM CHI proceedings.
Students typically had to prepare a short presentation about a research paper
in these courses. This helps students gain skills in oral communication (partic-
ularly if presentations are critiqued in class) and helps the courses to explore a
variety of different visualization approaches and techniques in discussion. Such
a presentation also could be part of a larger practical exercise or project (see
Section 2.3).

Own Lecture Notes (Q2c): Interestingly, all the instructors used their own
course lecture notes (many as PowerPoint slides). 58% published their lecture
notes on the course web site without any restriction. We assume that the re-
maining instructors either offered their notes on the web with restricted access
or simply used the notes to lecture from.
 Teaching Information Visualization 71

Teaching Assistants (Q2d): More than the half of all instructors (58%) had
no teaching assistant (TA) to support their course. In these cases, we can assume
that instructors also supervised practical exercises which were offered in almost
all courses. Only two courses were supported by two TAs, and the rest had
one TA. These figures may be due to small classes that were reported in most
cases, but might be indicative of a lack of support for teaching and learning in
Information Visualization, which is a very practical activity. Any such trend may
be of concern to InfoVis educators.

2.3 Practical Exercises

Survey analysis showed that nearly all courses (95%) had practical exercises
or projects. However, a wide variety of different types of exercises were noted.
One of the most common practical exercises cited was to have students use
InfoVis tools and then critique them. For example, students may choose from
a set of provided interesting datasets (election results, finance data, etc.) and
then they examine the datasets using the different InfoVis systems to reason
about analytic questions or tasks provided by the instructor or generated by
the student. These exercises often conclude with a written report evaluating
the tools and their effectiveness for the analysis. Alternately, some instructors
give students freedom to choose/generate their own input data. From the survey
results, the tools used most in such exercises are as follows:
– Spotfire, TIBCO Software, Inc. [63],
– TableLens, Inxight Software, Inc. [66],
– ILOG Visualization Suite, ILOG, Inc. [26],
– Tableau Desktop, Tableau Software, Inc. [65],
– InfoZoom, humanIT Software GmbH [27], and
– KidPad, University of Maryland [38].
The KidPad tool provides a set of zooming user interface and visualization facil-
ities so it can be used to develop small stand-alone visualizations on individual
topics. Thus, students may be asked to build a visualization on the basis of self-
chosen or self-generated data sets together with some conditions, e.g., to think
about designing overviews or spatial layout.
Another common approach of practical exercises is based on the idea that
students should implement the fundamental idea of a research paper. As a first
step, students (typically groups of at most three students) choose a paper from
a list given by the instructor. Next, the students prepare a brief presentation to
explain the fundamental idea and to give an overview on the planned implemen-
tation. This presentation helps to minimize the danger that the students will
become too focused on low-level details. To conclude, the students present and
demonstrate their tools in the classroom.
A last example of a common practical exercise is based on a special type of
data in information visualization: graphs and networks. In order to help students
better appreciate the difficulty of graph layout, the instructor has students draw
a small graph of 10-15 nodes based on only the connectivity of the graph’s
72 A. Kerren, J.T. Stasko, and J. Dykes

vertices. Students are instructed to make their graph drawing as aesthetically

pleasing as possible. When all the drawings are submitted, students view all the
different efforts and vote to select the best ones. In class discussion, students
explain why they voted as they did, and this can lead to a discussion of graph
drawing metrics, e.g., that edge crossings and bends have a negative effect on
aesthetic quality, etc. Such metrics are the basis of graph drawing algorithms
that could be analyzed by the students in a second step.

Grading of Practical Exercises: One can classify the described approaches

for practical exercises into two main groups:
1. implementation of a specific technique or implementation of a visualization
to solve a specific problem, and
2. examination of the usefulness of a given tool or visualization approach, such
as a commercial visualization system or graph drawing metrics, etc.
Grading of assignments of the first type can based on the overall quality of the
implementation itself, i.e., the instructor and/or TA evaluate the needed time,
aesthetic aspects, level of effort (complexity), usability, and capability provided
by the tool to get more insight about the chosen data set. Both types of assign-
ments can be combined with an oral presentation in the classroom. In this case,
the quality of the presented slides, lecture style, originality, and the presentation
quality itself can be used to evaluate presentation skills if these are intended
learning outcomes—and it is increasingly acknowledge as desirable to incorpo-
rate the teaching and learning of skills into the subject-area curriculum rather
than dealing with this independently. The survey yielded one interesting case in
which the students themselves voted on the best presentation(s).

2.4 Examination
More than the half (63%) of the surveyed courses had examinations of some
kind. The most common form was a written exam (37%), particularly in the
United States, but a notable portion (21%) used oral exams and 5% used both.
Oral examinations were used in Europe only, where there is a tradition of oral
examination for advanced level courses.
The survey also gave some insight into the different kinds of typical exam
questions (Q4a[ii]). There are a lot of different variants; a selection of the most
asked questions includes:
– Explain technique X for the visualization of problem Y .
– Given is a concrete problem and a task to be fulfilled. Which technique would
you use?
– Compare technique X with technique Y .
– Explain the construction of a Treemap, Starplot, Circle Segments, ...
– What are the advantages/disadvantages of technique X?
– What is a preattentive feature?
– What are the principles of using color?
 Teaching Information Visualization 73

This list only gives a rough overview about the issues that are important for the
instructors. The course web pages do not provide additional detail—we found
no specific exam questions or model answers when investigating methods of ex-
amination further. In general, we could observe that examiners focus not only
on technical approaches or methods, but also students’ capabilities for critical
reflection and to demonstrate their working knowledge of human visual percep-
tion.

3 Seminar Discussions

In the seminar session about teaching, Keith Andrews first presented the initial
survey result data. Next, workshop attendees discussed a variety of issues related
to teaching including “best practices” and ways to improve all our courses.

Curriculum:: A common problem reported by the attendees was some uncer-

tainty about how to logically organize the set of topics in an InfoVis course.
For example, Robert Spence’s 2nd edition textbook on information visualization
follows the classical pipeline model of representation (data types, tree represen-
tations, ...), presentation (space and time limitations, including zooming, distor-
tion, ...), and interaction (navigation, browsing, mental models, ...) [62]. Many
alternative ways of organizing course content exist, however. One suggestion was
to consider four cross-cutting dimensions: data types, domains, techniques, and
methodologies. Cognitive and perceptual issues were presented as an alternative
dimension of importance. The group did not come to any decisions about what
the most suitable structure would be since this is clearly dependent on the ori-
entation of the course and its learning aims. Simply being aware of alternate
orderings is valuable as instructors consider alternatives however.

Study Materials: The discussion on the use of research papers for compulsory
reading identified many different strategies for doing so. Many attendees echoed
a frustration about the difficulty in getting students to actually read assigned
articles, so many of the strategies addressed this particular issue. Several partic-
ipants reported about their own experiences and ideas, a number of which are
listed below.

– Papers are assigned, and students must present them. This takes place in
parallel with the regular lectures. This procedure seems to be pedagogically
beneficial because students learn to read actual research work, to prepare
a short talk and to give a presentation in the classroom. A disadvantage is
that student presentations vary greatly in quality. Some colleagues reported
on students losing interest in and not learning from poor presentations—
they would prefer the instructor to do all the lecturing. It is unclear if this
is truly a disadvantage, however. Perhaps, the amortized learning benefit is
high enough and this would justify the approach.
74 A. Kerren, J.T. Stasko, and J. Dykes

– The instructor gives lectures on mandatory meetings. Students can pick spe-
cific topics and lecture about a topic that the other students have not read
about.
– Papers are assigned, and there are written/oral questions on readings.
– Papers are assigned, and students must write a structured critical review
(about half a page) of them, i.e., a paragraph on the paper’s content, an
evaluative paragraph and an indication as to whether and why other students
might read the paper.

Practical Exercises: Software projects or practical exercises were viewed as

being very important to most attendees. These projects allow students to demon-
strate/learn the difficulty of many practical problems, such as the drawing of a
graph or navigation in large information spaces. Using a visualization tool (Spot-
fire, TableLens, . . . ) allows the students to interact with different visual represen-
tations and to gain experience about the advantages/disadvantages of different
visualization techniques. Furthermore, students become acquainted with com-
mercial tools. One interesting experience of attendees following this approach
was that students’ negative impressions of InfoVis systems mostly involved user
interface or HCI issues, not the actual visualization technique(s).
One problem with such exercises is the potential difficulty in gaining com-
mercial software for use in the projects. More vendors are making their systems
freely available for educational use, however. Tableau Software is an example of
a company doing so. Another issue is the challenge of finding “good” data sets to
be used by the students. One suggestion was use data sets from previous InfoVis
Conference contests but most attendees felt that the contest datasets are too big
and complex for introductory courses. Still, the contest data sets may be suitable
for advanced level courses where students have a good background in the most
important techniques. Perhaps, the contest should include the production of a
data subset, specifically designed for educational purposes.

Using Other Media: Workshop attendees discussed that a large and com-
prehensive public collection of InfoVis-related images and videos would be very
helpful for instructors. Videos of interaction scenarios that show the usability
and interaction capabilities of the tools would be especially beneficial. Images
also could help to illuminate the history of InfoVis and illustrate different vi-
sualization techniques. Unfortunately, gathering a collection of images or videos
in this way could cause copyright problems. This may be why many instructors
have their own image/video archives with private access. The HCC Digital Li-
brary [24] of Georgia Tech is an example of an effort to gather a large collection
of educational resources, but it is focused broadly on HCI, not just InfoVis.
Another possibility to obtain video material is to examine conference DVDs,
such as the annual VIS/InfoVis/VAST DVD. Many contributions provide an
additional video to clarify the usage and interaction techniques of their work.
Again, it may be beneficial to encourage attendees to develop video summaries
of their work, specifically for teaching and learning.
 Teaching Information Visualization 75

4 Personal Perspectives

In this section, three of the workshop attendees provide their own unique per-
spective on teaching InfoVis and InfoVis-related topics.

4.1 John T. Stasko, Georgia Institute of Technology, USA

I have been teaching a graduate course on Information Visualization at Georgia

Tech since 1999. My university changed from a quarter system to a semester
system then and I decided to create a course for this area that was becoming my
research focus and growing in interest worldwide. Over the years I have thought
of this course not merely as a teaching assignment but as a fundamental part
of my academic portfolio and research mission. The course provides training for
students to learn about the area and do subsequent research with me, or simply
to apply their knowledge in business or government. Also, the course has directly
led to a number of the research contributions made by students in the course
and by my research group. Student projects from the course have won major
contests [20, 54] or led to research papers [12]. Additionally, my dissatisfaction
with the state of knowledge and background articles on particular course topics
led to projects undertaken by my research group in those areas (i.e., analytic
goals [3], user tasks [1], and interaction [73]). Below I will provide more details
about the course and the projects that have resulted from it.
The Information Visualization course (CS 7450) is usually offered in the
Spring semester each year. At Georgia Tech, semesters are 15 weeks long and and
my course meets for 1.5 hours twice a week. The web pages for the most recent
course offering can be found at http://www.cc.gatech.edu/~stasko/7450.
Human-computer interaction (HCI) is an important area of research in my
home School—we have both a Masters degree in HCI and a doctorate in Human-
Centered Computing (HCC), in addition to our undergraduate and graduate
degrees in computer science. Students are drawn to the HCI and HCC degree
programs from a variety of undergraduate majors, so many do not have formal
computer science training. I made a conscious decision to designate the graduate
HCI course as the only prerequisite for Information Visualization in order to
encourage students from a wide variety of disciplines to enroll. Consequently,
many students who do not have a strong background in programming typically
take Information Visualization.
This fact has implications on the way that I teach the course. I do not use
homework assignments in which the students must implement a visualization
technique or system. Instead, assignments are more oriented toward design, cri-
tiquing and evaluation. I employ a group project where I do require that some
type of software system be built, but the course demographics allow there to be
at least one or two team members who are experienced programmers.
My high-level goal for the course is to have students learn the different infor-
mation visualization techniques that have been created including the strengths
and weaknesses of each, and to have the students become better critics of infor-
76 A. Kerren, J.T. Stasko, and J. Dykes

mation graphics and visual systems. More specifically, the learning outcomes for
the course include
– Students should gain an in-depth understanding of the field of Information
Visualization including key concepts and techniques.
– Students should be able to critique visualization designs and make sugges-
tions to improve them.
– Students should be able to design effective visualization solutions given new
problems and data domains.
– Students should learn about the spectrum of commercial system solutions
available in this area and how to choose one for a particular task or problem.
Perhaps the main challenge that I have faced in this course over the years is to
construct a coherent syllabus and flow of topics throughout the term. Information
Visualization is still a new area that is growing and maturing. Consequently,
it does not exhibit a well-understood and agreed-upon set of topics that flow
smoothly from one to the next. In my experience teaching the course, a number
of key ideas have risen to the surface and I make these the important components
of the course:
– Data foundations - A description and model of the different types of data
that are encountered and how this data is transformed and stored for easier
subsequent manipulation.
– Cognitive issues - A discussion of the user’s goals and tasks in using an
information visualization system. What cognitive benefits can visualization
provide?
– Visualization techniques - A description of the different visual represen-
tations and interaction techniques that have been invented.
– Interaction - A discussion of the different types and the many issues sur-
rounding interaction.
– Data types/structures - An introduction to specific types of data (e.g.,
time series, hierarchical, textual) and the visualization techniques that are
well-suited at representing those data types.
– Data domains - An examination of different domains (e.g., software engi-
neering, social computing, finance and business) and the visualization tech-
niques that are helpful to people working in those areas.
– Evaluation - A dialog about the challenges of evaluation in information
visualization and a review of different evaluation techniques that have been
used in the area.
Some of these topics are fundamentally interwoven so the flow of concepts is
not clearly self-contained and independent. For instance, certain visualization
techniques are best used for specific data types (e.g., treemaps for hierarchical
data). In organizing the course content, I feel this tension and often struggle
with which topics to teach first. Nonetheless, my course uses this progression of
topics as its organizational framework.
The course is lecture-based but I try to engage the students in discussions
about the different concepts being studied. I have used Bob Spence’s textbook
 Teaching Information Visualization 77

some terms augmented by selected papers, and other terms I have used only
research papers. I have settled on having students read one or possibly two
research papers for each class. Typically, the paper is an important one for that
topic or it is a good overview of the issues involved. When I have assigned more
papers than this, I find that the students often do not adequately prepare and
read all the papers. To cover more recent research, I typically select two or
three recent articles on the topic of the day and I assign two or three students
who must recap and describe their particular paper’s key ideas to the class in
less than five minutes. I believe that experience giving presentations like this is
important and valuable to the students. All of my lecture slides can be found at
the course website and in 2007 I created in-studio video versions of each lecture.
These videos can be found at the website http://vadl.cc.gatech.edu.
I use a number of relatively small homework assignments in the course,
along with one larger homework and a group project. I will frequently employ
a midterm or final exam as well. The small homeworks often involve a visual-
ization design exercise (on paper) given a data set. Of course, such assignments
do not engage the interactive component of information visualization that is so
important, so they are fundamentally limited.
The larger homework assignment is a commercial tools critique. Students
are given five example datasets and asked to choose the two that they find
most interesting. Before using any systems, the students examine the datasets
and generate questions about them. Next, the students use a few information
visualization systems to explore the data and try to answer those questions.
I also alert the students to note any serendipitous findings that occur during
exploration. Finally, the students must write a report in which they critique the
different systems used, the visualization techniques each employs, and whether
the systems led to insights and discoveries. I have used systems such as Spotfire,
SeeIt, Advizor, Eureka (Table Lens), InfoZoom, InfoScope, and Grokker over
the years. I find this assignment to be extremely valuable to the students as it
allows them to gain hands-on experience with sophisticated systems and shows
them how visualizations can (or cannot) be helpful in analysis and exploration.
This particular assignment even led to an interesting research contribution
by my group. We studied the analytic queries generated by students over many
years of the course and clustered these inquiries into different low-level analytic
tasks that visualizations may assist. Our taxonomy of these tasks was presented
at the 2005 Symposium on InfoVis [1].
I also employ a group project in the course in which students design and build
a visualization system for a particular problem and data set. Teams of three or
four students work together for most of the term and find a client with a data
analysis problem or they simply choose a data set and envision the kinds of ana-
lytic queries that one would expect on it. The students explore different visualiza-
tion designs, then they choose one to implement. In the past, student teams have
often chosen to work on the contest datasets from the IEEE InfoVis or VAST
Conferences. In fact, student teams from my course have won these contests on
multiple occasions or have had competitive entries in the contests [20,54]. Group
projects have even led to full papers at the InfoVis Symposium as well [12].
78 A. Kerren, J.T. Stasko, and J. Dykes

One ongoing tension with the group project is simply when to begin the
assignment. By initiating the project early in the term, students have more
time to work on it and make better progress. However, at that early point,
students have engaged very little course material and so their understanding of
information visualization concepts and ideas is not as rich. I have found that the
simple topic chosen for the project can have a profound impact on the results,
and better knowledge of the information visualization area leads students to
make better choices in project topics. This has led me to wait until the midterm
point to distribute the project in some semesters, but then the students have
much less time to work on it.
Overall, the Information Visualization course has been valuable to me in
many different ways. Perhaps most importantly, the process of preparing lec-
tures and course material has made me reflect on the topics that I would be
discussing and question “accepted” knowledge in the domain. I believe that this
has made me a better researcher and it has generated ideas for new projects and
investigations.

4.2 Andreas Kerren, Växjö University, Sweden

My experiences in teaching Information Visualization go back to the year 2003.
At this time, I was a temporary assistant professor at the Institute of Com-
puter Graphics and Algorithms of the Vienna University of Technology, Austria.
The institute offered one course on InfoVis and several other related courses.
In this environment, I had the opportunity to give several lectures on Soft-
ware Visualization and domain-specific visualization, such as visualization in
Bioinformatics. In 2005, I moved to the University of Kaiserslautern in Ger-
many. There, I was responsible for the annual InfoVis courses. Based on my
experiences from Vienna and current flows in research, I designed a completely
new syllabus for this course. Originally, this syllabus provided 15 lectures (one
semester at TU Kaiserslautern) plus practical exercises for Masters level stu-
dents; each lecture took 1.5 hours once a week. As I have been appointed for a
faculty position at Växjö University (VXU), Sweden, in 2007, some modifications
were needed to address the different course and teaching system at this univer-
sity. Web pages (in English) for the most recent InfoVis courses at VXU can be
found at http://cs.msi.vxu.se/isovis/courses/. In this section however, I
will focus to my experiences with my courses given at TU Kaiserslautern from
2005-2007, because the details of my last course there (in the winter semester
2006/2007 (WS06/07)) reflect my particulars within the Dagstuhl survey.
My general learning aims for the course are more or less identical to John
Stasko’s four learning outcomes at Page 76. Therefore, I don’t want to repeat
them at this place. It was very important for me to give students the opportunity
for critical reflections and to show the newest directions in research. Furthermore,
my course covers basic principles from cognitive psychology that have influence
on InfoVis, such as human visual perception or Gestalt laws. As a result of this
position, each technical approach and tool was discussed with respect to its value
(please compare [17] in this book), its usefulness, and—if existent—its usage in
 Teaching Information Visualization 79

Challenges
Challenges Perception
Perception

Basics
Basics
Evaluation
Evaluation

Introduction
Introduction Dynamic
Dynamic
GeoVis
GeoVis Queries
Queries
Zoom
Zoom
Interaction
Interaction &
WebVis Applications
WebVis Applications &
Techniques
Techniques Pan
Pan
Time-
Time-
dependent
dependent Trees
Trees
data
data Focus
Focus
BioVis
BioVis &
&
Visual
Visual Context
Context
Special
Special Structures
Structures
Data
Data
1D,
1D, 2D,
2D,
Text
Text &
& 3D,
3D, 4D,…
4D,…
SoftVis
SoftVis Graphs
Graphs &
&
Docs
Docs
Networks
Networks

Fig. 3. Course structure of the WS06/07 InfoVis course given by Andreas Kerren at
TU Kaiserslautern.

commercial products. Of course, this was a challenge and sometimes a little bit
subjective because of missing quality metrics or missing evaluations of tools and
techniques.
The design of the syllabus was partly influenced by courses given at Georgia
Tech and TU Vienna in 2005 as well as by the textbooks of Spence [61], Ware [71],
a pre-version of the textbook of Kerren et al. [35], and many research papers.
The course is divided into three parts which are illuminated by Figure 3:
1. In the first part, I discuss basic knowledge that is important for the design
or analysis of InfoVis concepts. As a first step, I introduce the field itself,
give motivations for the need, and present several traditional and modern
examples, mainly from Tufte’s [67, 69] and Spence’ books. Important is the
differentiation between InfoVis and SciVis, also, if that is not so easy in
some cases. After this introduction, a larger discussion on perception and
cognitive issues is given. Here, I provide information about the perception
of colors, textures, etc., preattentive features, and Gestalt laws. This course
component is mainly based on the book of Ware, but also on a lot of examples
and animations that can be found in the WWW. The last lecture of this
first part describes basics, such as the InfoVis Reference Model (data tables,
visual mapping, interaction, etc.) or data types and dimensionality. These
issues are mostly based on the book of Card et al. [8].
80 A. Kerren, J.T. Stasko, and J. Dykes

2. The second part is the largest one of my course. Here, I discuss the most
important interaction techniques at first, for example Dynamic Queries,
Zoom&Pan, and several Focus&Context related techniques. This component
is more or less geared to the InfoVis Reference Model [8], i.e., I distinguish
interaction by means of data transformations, visual mapping, and view
transformations. I use actual research papers to exemplify the different ap-
proaches. From a didactic point of view, this is a little bit tricky, because I
presume some knowledge in visual representations or structures to explain
my examples. I decided to discuss interaction at first and the visual struc-
tures for different data types after this. One advantage is the possibility to
refer later to discussed interaction techniques directly with less additional
explanations. My experiences with students show that they accept this order,
and that they have no problem in understanding the differences/correlations.
But this should be communicated previously.
As described before, a discussion of the most important visual structures for
more basic data types follows the interaction component. Here, I introduce
visualization techniques for multivariate data, hierarchies and graphs mostly
on the basis of research papers as well as the books of Spence and Kerren et
al. Individual solutions for special kind of data types, e.g., time-series data,
text, or software, follow this component directly. The second part finishes
with visualization techniques for different data domains, such as BioVis,
WebVis, GeoVis, etc. During the past years, I vary this part of the lecture a
little bit depending on my current research interests or hot-topics. Resources
for these lectures are current papers and articles, but also the second part
of the textbook [35].
Each course component of this second part is accompanied by short video or
tool demonstrations. From my perspective, this is absolutely needed, espe-
cially for the different interaction techniques and their interplay with visual
structures. It is fun to keep an eye on the students during such demon-
strations, and as a result they are motivated to ask deeper questions. One
interesting and traditional example is the claim to preserve the mental map
in dynamic graph drawing. Only with the help of a video or demo it is pos-
sible to illustrate the difference between morphing or other techniques, such
as foresighted layout [14]. However, the usage of video or tools is not always
possible because of unavailability.
3. My course concludes with 1-2 lectures on possible evaluation techniques and
the most important InfoVis challenges for the next years. Because of the miss-
ing time at the end of the semester, I focus on specific aspects of these issues.
For example, the intended learning aim is to impart students an overview
knowledge of basic evaluation techniques and—perhaps more important—an
impression of the difficulties to perform such an evaluation. The final discus-
sion of the most important challenges gives an idea about the current state
of the field and leads to take part of it, for example, by working on a thesis
in my research group. A good source for these issues are the corresponding
chapters in Kerren et al. [35].
 Teaching Information Visualization 81

Another important part of the course are the assignments. They are composed
of a brief presentation, of a software implementation, and of a short software
demonstration at the end of the semester. Each student or student group (con-
sisting of maximal two students) chooses a specific research paper from a list
given on the course web page. I take care that the papers’ topics and the pre-
sented approaches are not too complex. The final aim of the assignment is that
the main idea of a paper should be implemented in any programming language.
It is not needed that all features or interaction possibilities are implemented. But
a GUI is mandatory in order to give me and the other students the chance to
load another input file etc. Data sets depend on the paper topic, e.g., if the paper
presents a new treemap layout then the students can choose their own input,
such as the hierarchical file system on their own personal computer. All these
topics should be discussed in the first presentation in the middle of the course.
In a first step, each student or group prepares a presentation (10 minutes plus
3-5 minutes of discussion) about the chosen paper followed by a working plan.
In this way, I can steer the processes, give hints, and prevent nasty surprises. At
the end of the course, all implementations are presented and discussed in class.
I have found that this division into two presentations and demos respectively
helps students to think about important concepts. Furthermore, they have al-
ready learned the most important theoretical concepts during the course before
they start to program. My overall impression of this practical project is very
positive. At the beginning, the students often had doubts because it appears
time-consuming and complex, but they had a lot of fun in the progress of the
semester. The results were mostly really great; for me it is important that they
learn to see the difficulties and to carefully reflect about the paper, not so much
the result itself. Often, however, the resulting programs were amazingly good.
My pedagogical concept, especially for the assignments, clearly follows moder-
ate constructivistic learning approaches, as described in the following written by
Jason Dykes or in some of my papers on learning concepts in context of using
Software Visualization techniques [33, 32, 59].
The course evaluation by my students led to very good results for this course.
They liked the way I structured the course, the motivating examples and videos,
and they had the subjective feeling that they have learned a lot of interesting
things. At large, it was not difficult to motivate students for InfoVis. It is a very
interesting field also suitable for the solution of practical problems. Therefore,
it was sometimes not so easy to explain why people cannot find more InfoVis in
standard software products. This leads to a problem that is discussed in paper
The Value of Information Visualization [17] of this book.
I would like to discuss one further issue that is important to me: Finding a
good balance between giving a good overview of the field as compared to explain-
ing the details of specific visualization techniques is pretty difficult, especially in
the frame of 15 course lectures. Some students liked to get more overview knowl-
edge of InfoVis, but they also disliked that some topics were only briefly covered.
For instance, I used 1-2 lectures for the visualization of graphs. It is enough time
to explain the most important things, but not enough time to explain the dif-
ferent graph drawing techniques in detail. Thus, I abstracted in many cases,
82 A. Kerren, J.T. Stasko, and J. Dykes

but some students would like to learn more. The level of detail/abstraction in
teaching InfoVis is not obvious. In general, my solution for this problem is to
offer a seminar back-to-back after the InfoVis course, where interested students
can choose a specific topic and prepare a presentation on it. This allows for
more deeper discussions. Additionally, such a seminar is a good starting point
for subsequent thesis work.
Typically, my courses terminate with an oral examination. Regarding our
survey results, this is consistent with the examination practice of many colleagues
coming from Europe, cp. Section 2.4.

4.3 Jason Dykes, City University London, UK

Introduction and Context: My background is in geography and the geo-
sciences, where there is a strong tradition in the use of maps and graphics. Tra-
ditionally undergraduate courses in geography have taught and assessed manual
skills in cartography and mapping activities and projects remain at the core of
many geology degrees. Increasingly these activities are being augmented or re-
placed by learning that involves Geographic Information Systems (GISystems)
and other digital techniques, but a focus on cartographic principles for design-
ing effective maps and communicating geographic information effectively is still
regarded as important.
My ‘Visualization’ Module. I have been teaching a Masters level module in ‘Geo-
Visualization’ for eight years. The postgraduate module was originally developed
to provide Geographic Information Science (GIScience) students with skills in
creating, evaluating and using maps and graphics in their analysis and communi-
cation of geographic information. It replaced a cartography module and updated
this with recent advances in the use of dynamic and interactive graphics for ex-
ploratory analysis.
During this time the GIScience community and GIScience students have be-
come more aware of Information Visualization. Equally students from non GI
programs have wanted to learn about cartography and visualization. It is chal-
lenging to develop a coherent module that is relevant to this range of students
whilst embracing Information Visualization and a traditional cartography syl-
labus.
My module consists of 12 sessions with 3 hours contact time and 7 hours
guided individual study. Students are expected to spend another 30 hours par-
ticipating in assessed activities. The module is thus designed such that an average
student spends 150 hours studying for 15 credits—-it uses a shared credit frame-
work that does not conform to the requirements of the Bologna process [5]. The
module is available to distance learners through a managed learning environment
and digital resources ensure distance learners have an equivalent experience those
who attend campus.
Many students go on to do research projects in visualization involving data
sets and applied problems that feed my research and are subsequently used in
teaching. This effective feedback loop is important in perpetuating bi-directional
links between research and teaching.
 Teaching Information Visualization 83

Teaching Issues in Information Visualization at Dagstuhl: I don’t know

the content or culture of InfoVis or Computer Science education. This is one of
the difficulties associated with having interests that lie between rapidly devel-
oping disciplines. But the Dagstuhl discussions and survey suggest that many
of those involved are facing similar issues with which I am familiar. The sur-
vey draws attention to a range of approaches, topics, academic disciplines, text
books involved.
There is great variation in the approaches and syllabuses of the courses re-
ported in the survey. ‘Information Visualization’ is the most popular title, but
course names that begin: ‘Models and Algorithms for . . . ’; ‘Data Visualization’;
‘Computer Graphics and . . . ’; ‘GeoVisualization’ are evidently situated in dif-
ferent ways and will require different emphases. This is evident in Figure 1—a
tag cloud showing relative occurrences of topics listed in courses included in the
Dagstuhl survey.
Despite this texture, some consistent themes emerge from the survey and
particularly the discussion at Dagstuhl. Three particular themes resonated with
me and related to my experience. They are also evident in a report produced
following an open workshop at IEEE Visualization 2006 in which I graphically
depict the relationships between related courses with an Information Visualiza-
tion emphasis [56]:
1. a trend/desire for evaluative and critique-based learning that can draw from
a range of related disciplines;
2. an emphasis on ‘learning through doing’ as opposed to a transmissive ap-
proach to learning;
3. some concern about developing appropriate exercises and assessment for vi-
sualization classes.
I’ll briefly consider these key issues and reflect upon their relationship with
my teaching. Doing so may provide some synergies in the cross disciplinary
educational mash-up that is developing in Information Visualization.

Critiquing: Critiquing involves students applying and developing their knowl-

edge through evaluation and review. Robert Kosara argued convincingly for a
critique-based approach to Information Visualization at Dagstuhl. He describes
this as a ‘highly interactive and human-centered way of designing things’ [40].
An approach that evaluates graph drawing algorithms and evaluation criteria is
documented in the Dagstuhl survey and described in Section 2.3 above.
As a learning device, critiquing may involve evaluating existing work, soft-
ware or graphics, but importantly should involve the practical application of
theory. I ask students to critique existing graphics, those that they have devel-
oped and software systems using a number of criteria, such as:
– Graphical integrity, graphical excellence and Tufte’s ‘theory of data graph-
ics’ [69]
– Using appropriate symbolism [6, 10]
– Map symbolism [45, 60]
84 A. Kerren, J.T. Stasko, and J. Dykes

– Map design [7]

– Use of colour [22]
– Interactivity and functionality [11, 41]
– Animation [21, 55]

Each of these sources provides useful criteria to structure critiques and against
which judgments can be made. Amar and Stasko’s framework [2] also offers
opportunities and Dagstuhl has drawn attention to the scope for using more
knowledge from InfoVis in developing critiquing criteria.

Paper Summaries. Presenting summaries of research papers is clearly a core

activity in InfoVis teaching and learning. This fits the critiquing trend if the
work is evaluative and assessed against existing theory and criteria. Tamara
Munzner’s generic ‘what makes a good InfoVis paper’ criteria as presented and
discussed at the Dagstuhl seminar [52] could prove useful in this kind of learning
activity.
The validity of critiquing depends upon the level of learning being supported.
In the UK some of the more advanced levels of knowledge, such as those devel-
oped through critiquing, relate to the more advanced levels of learning. Masters
level and level-3 courses require students to evaluate and so the approach is
particularly appropriate. It can be used to provide feedback and for assessment.
I’ve found that extended abstracts work well as a focus for quick and informal
‘in class’ critiquing that breaks up a learning session.

Learning by Doing:

“I hear and I forget, I see and I remember, I do and I understand”

(Chinese proverb).

Participants in an Information Visualization seminar are likely to be persuaded

that seeing is a powerful learning device. But ‘learning by doing’ may be even
more effective. It is a form of active learning that is popular in education and
there is particular opportunity for using this approach in visualization education.
Doing can involve summarizing information, developing graphics or software or
analyzing data sets and the Dagstuhl participants provide plenty of examples of
these activities in their courses as we have seen in this paper.
Fieldwork is used to provide opportunities for learning by doing in the geo-
sciences and has an important role in education [31]. There are parallels between
the kind of observation, interaction and exploration that occur in fieldwork and
those associated with the kind of exploratory analysis that visualization sup-
ports. I have explored some of these parallels and opportunities when developing
learner-focused activities that emphasize learning by doing. The constructivist
approach emphasizes learning through an interpretive, non-linear and recursive
process in which active learners interact with their surroundings and the re-
sources that are provided to help [19]. Such methods are regularly used in geo-
science fieldwork and may result in ‘deep understanding’ of the kind experienced
 Teaching Information Visualization 85

when researchers use visualization to interact with complex structured datasets.

They seem particularly suited to visualization education.
Multiple perspectives are important in constructing knowledge and so group
work is often employed in such activities. In this context, symbolic and graphi-
cal representations of knowledge and ideas and can be important in negotiating,
mediating and constructing shared meanings [19]. This is particularly so if the
graphics can be manipulated in an exploratory context as knowledge is derived.
Consequently strong arguments exist for using software that is interactive and
exploratory to develop opportunities for learner-centered constructivist activi-
ties [74].
I have developed exploratory visualization software for use in fieldwork in
this context. It encourages ‘learning by doing’ through a series of linked carto-
graphic, statistical and photographic views of a study area and a highly interac-
tive interface through which these can be manipulated. It forms part of a learner
constructed activity - the software is one of a series of resources made available
to students who are expected to develop an approach to a problem relating to
land cover and land use. The contrasts with passive or transmissive education
are emphasized by the exploratory, student-led visualization and the ability to
add data recorded in the field into the software for analysis [46]. Our evaluations
show that the software and the constructivist activity that it supports are an
effective learning device [47].
The Dagstuhl survey suggests that a number of similar activities may be
taking place. It certainly seems that there is scope for using exploratory graphical
software to support constructivist learning methods in visualization that give
students the opportunity to learn actively.

A Portfolio Approach to Assessment: The issue of exercises and assessment

was discussed at Dagstuhl primarily because some teachers were finding it dif-
ficult to set and assess meaningful exercises. Critiquing and ‘learning by doing’
provide scope for effective objective assessment that encourage active learning.
My experience with portfolio-based assessment has been very positive in this
context and compliments these methods well.
Portfolios involve students developing and collating annotated evidence of
their capabilities throughout a course. They receive formative feedback on their
work and use selected work to demonstrate that they have achieved a set of
learning outcomes. Portfolio-based assessment can spread workload for staff and
students, offers the opportunity for developmental feedback and review, and
provides a synoptic view of what has been learned. It also results in a tangible
end-product that students can show to colleagues and prospective employers—
potentially—bringing learning to life and engendering pride.
I use portfolio-based assessment in the GeoVisualization module. Students
participate in a practical exercise associated with each learning session that
requires them to ‘learn by doing’. Their work is discussed and improved. Selected
work is submitted for formal formative feedback. At the end of term students
are asked to submit a portfolio of three exercises and a reflective essay that uses
these as items of evidence to demonstrate that module learning outcomes have
86 A. Kerren, J.T. Stasko, and J. Dykes

been achieved. Amongst other competencies, outcomes require that students are
able to . . .
– explain the complex issues associated with GeoVisualization with clarity and
from an informed perspective by drawing upon recent academic research;
– design maps and data graphics that are effective, informative and consistent
and that exhibit graphical excellence and graphical integrity;
– use data graphics, maps and visualization tools to present and explore mul-
tifaceted data sets in a manner that is professional, informed and ethically
sound;
– evaluate data graphics, maps and visualization tools by drawing upon prin-
ciples and theories of design.
The approach seems to work nicely and addresses some of the issues discussed at
the Dagstuhl meeting. It may be useful for those wishing to help students learn
to critique and assess their developing skills. I have received favorable feedback
from students and internal and external evaluators.
It seems particularly appropriate for developing skills in graphicacy where
Tufte’s concept of ‘redesign’ is a key element. Portfolios or long-term developing
group projects that provide opportunities for feedback and critique can be very
beneficial here. It should be noted that portfolios are frequently used in the arts
where critiquing and redesign are key learning activities.

Conclusion: These arguments are personal perspectives, developed through

reflections on discussions at Vis 2006 and the Dagstuhl seminar and the Dagstuhl
survey, in the light of my teaching experience. I suggest a focus in Information
Visualization education not on curriculum (which may vary to suit particular
disciplines and student groups) but on general qualities and competencies that
can be applied across a range of curricula. I’ve identified what some of these
might be and presented some ideas on how they might be supported and assessed
through the example of my GeoVisualization module.
The critiquing and ‘doing’ themes support the notion that underlying skills
in the use and evaluation of graphics are broadly (I hesitate to say ‘universally’)
valuable. They may be a way of helping Information Visualization educators
bridge the multiple multi-disciplinary divides and certainly help justify the ap-
proach taken on my GeoVisualization module.
They may also help us deal with change—we are in a discipline where the
specifics of curricula change very rapidly. Perhaps we’re moving away from a
text-book based model of teaching as disciplines change so quickly and student’s
expectations and abilities to access information increase. This is good news from
the perspectives of information sharing and efforts to link research and teaching,
which many are evidently doing very effectively, cf. Section 4.1. The bad news is
that we need more flexibility in terms of course content and perhaps structure.
This is no bad thing in itself, but is difficult to achieve when the levels of docu-
mentation that are expected by students and (in the UK at any rate) required
for quality assurance and to gain approval for changes in provision are consid-
ered. Guided approaches in which links to a selected set of papers and examples
 Teaching Information Visualization 87

are used to support learning against broad aims and outcomes provide a way
forward. It is well worth using the URLs listed in the Dagstuhl survey to learn
from colleagues who use this approach (see Stasko, Munzner and Heer’s courses
for example). The kinds of repositories of examples and teaching materials dis-
cussed and suggested in Section 4.1 will help, as will a focus on generic methods
of teaching such as those that involve critique and active learning rather than
developing monolithic curricula that will age rapidly in response to new devel-
opments. Portfolio-based assessment that involves the focused combination of a
series of activities supports this flexible approach.
Adoption of the ideas discussed here would continue to move visualization
education away from core Computer Science. Doing so will continue the trend of
enabling more students to participate in visualization education and help address
the difficulties associated with multi-disciplinary domains - how do we focus si-
multaneously on the concepts listed in our tag cloud of the scope of Information
Visualization education (Figure 1)—computer science and algorithms, the sci-
ence of perception and cognitive studies and concepts derived from the arts such
as composition and design? The Dagstuhl survey has certainly helped inform
my approach to visualization education. Perhaps this discussion will help the
community when considering the nature of Information Visualization education
and how to best it might be supported and developed. I’d certainly be delighted
to debate the ideas and their relevance further.

5 Conclusion
This paper describes the results of our teaching survey based on the information
given by the Dagstuhl attendees. It covers several aspects of offered InfoVis
courses that range from different kinds of study materials to practical exercises.
We have reproduced the discussion during the Dagstuhl Seminar and added
our own experiences. In this regard, we have found that teaching InfoVis is
challenging because it is a new and growing field. There exist a lot of open
questions regarding the syllabus, a consistent theory, or the abstraction level of
single topics. In consequence, it is also a great subject for teachers, not only for
students: we are convinced that teaching InfoVis also leads to a better reflection
on the topics and to new ideas which can induce new projects. Finally, we hope
that this paper can serve as an interesting and helpful source for current and
future InfoVis teachers.

Acknowledgments. We would like to thank all participants at the seminar [13,

36] for filling out the teaching survey and for the lively discussions. The survey
was developed by Keith Andrews, and a first analysis of the results was also
made by him. We thank him for his ideas and efforts.

References
1. Amar, R., Eagan, J., Stasko, J.: Low-level components of analytic activity in
information visualization. In: Proceedings of the 2005 IEEE Symposium on In-
formation Visualization - INFOVIS ’05, October 2005, pp. 111–117 (2005)
88 A. Kerren, J.T. Stasko, and J. Dykes

2. Amar, R., Stasko, J.: A knowledge task-based framework for design and evaluation
of information visualizations. In: IEEE Symposium on Information Visualization
(InfoVis), Austin, TX, pp. 143–150. IEEE Computer Society Press, Los Alamitos
(2004)
3. Amar, R.A., Stasko, J.T.: Knowledge precepts for design and evaluation of infor-
mation visualizations. IEEE Transactions on Visualization and Computer Graph-
ics 11(4), 432–442 (2005)
4. Andrews, K.: Course: Information Visualisation (2007),
http://courses.iicm.tugraz.at/ivis
5. BENELUX Bologna Secretariat: About the Bologna Process (2008),
http://www.ond.vlaanderen.be/hogeronderwijs/bologna/about/
6. Bertin, J.: Semiology of Graphics: Diagrams, Networks, Maps. In: Translation of
Semilologie Graphique, University of Wisconsin Press, Madison (1983)
7. Brewer, C.A.: Designing Better Maps: A Guide for GIS Users. ESRI Press, Red-
lands (2005)
8. Card, S., Mackinlay, J., Shneiderman, B. (eds.): Readings in Information Visual-
ization: Using Vision to Think. Morgan Kaufmann, San Francisco (1999)
9. Chen, C.: Information Visualization: Beyond the Horizon. Springer, Heidelberg
(2006)
10. Cleveland, W.S., McGill, R.: Graphical perception: Theory, experimentation and
application to the development of graphical methods. Journal of the American
Statistical Association 79, 531–554 (1984)
11. Crampton, J.: Interactivity types in geographic visualization. Cartography and
Geographic Information Science 29(2), 85–98 (2002)
12. Csallner, C., Handte, M., Lehmann, O., Stasko, J.: FundExplorer: Supporting the
diversification of mutual fund portfolios using Context Treemaps. In: Proceedings
of the 2003 IEEE Symposium on Information Visualization - INFOVIS 2003,
October 2003, pp. 203–208 (2003)
13. Dagstuhl: Seminar 07221 Information Visualization – Human-Centered Issues and
Perspectives (2007),
http://www.dagstuhl.de/07221
14. Diehl, S., Görg, C., Kerren, A.: Preserving the Mental Map using Foresighted
Layout. In: Proceedings of Joint Eurographics – IEEE TCVG Symposium on Vi-
sualization (VisSym ’01), Eurographics, pp. 175–184. Springer, Heidelberg (2001)
15. Dykes, J., MacEachren, A.M., Kraak, M.-J.: Exploring Geovisualization. Perga-
mon Press, Oxford (2005)
16. Fayyad, U., Grinstein, G., Wierse, A.: Information Visualization in Data Mining
and Knowledge Discovery. Morgan Kaufmann, San Francisco (2001)
17. Fekete, J.-D., van Wijk, J.J., Stasko, J.T., North, C.: The Value of Information
Visualization. In: Kerren, A., Stasko, J.T., Fekete, J.-D., North, C.J. (eds.) Infor-
mation Visualization. LNCS, vol. 4950, Springer, Heidelberg (2008)
18. Few, S.: Show Me The Numbers. Analytics Press (2004)
19. Fosnot, C.T., Perry, R.S.: Constructivism: A psychological theory of learning. In:
Fosnot, C.T. (ed.) Constructivism: Theory, Perspective and Practice, 2nd edn.,
pp. 8–38. Teacher’s College Press, New York (2005)
20. Grinstein, G., O’Connell, T., Laskowski, S., Plaisant, C., Scholta, J., Whiting,
M.: Vast 2006 contest - a tale of alderwood. In: Proceedings of the 2006 IEEE
Symposium on Visual Analytics, Science and Technology - VAST 2006, October
2006, pp. 215–216 (2006)
21. Harrower, M.A.: Tips for designing effective animated maps. Cartographic Per-
spectives 44, 63–65 (2003)
 Teaching Information Visualization 89

22. Harrower, M.A., Brewer, C.A.: Colorbrewer.org: An online tool for selecting color
schemes for maps. The Cartographic Journal 40(1), 27–37 (2003)
23. Hauser, H.: Course: Information Visualization (2007),
http://www.cg.tuwien.ac.at/courses/InfoVis/
24. HCC: HCC Education Digital Library (2007), http://hcc.cc.gatech.edu/
25. Heer, J.: Course: Visualization (2006), http://vis.berkeley.edu/courses/
cs294-10-sp06/
26. ILOG Visualization Suite. ILOG, Inc.
http://www.ilog.com/products/visualization/index.cfm, 2007.
27. InfoZoom: humanIT Software GmbH (2007), http://www.infozoom.com/enu/
28. Jankun-Kelly, T.: Course: Information Visualization (2006),
http://www.cse.msstate.edu/~ tjk/teaching/cse8990/
29. Keim, D.: Course: Information Visualization (2007), http://infovis.uni-
konstanz.de/index.php?region=teach&event=ss07&course=infovis
30. Keller, P.R., Keller, M.M.: Visual Cues – Practical Data Visualization. IEEE
Computer Society Press, Los Alamitos (1993)
31. Kent, M., Gilbertson, D.D., Hunt, C.O.: Fieldwork in geography teaching: a crit-
ical review of the literature and approaches. Journal of Geography in Higher
Education 21(3), 313–332 (1997)
32. Kerren, A.: Generation as Method for Explorative Learning in Computer Sci-
ence Education. In: Proceedings of the 9th Annual Conference on Innovation and
Technology in Computer Science Education (ITiCSE ’04), Leeds, UK, pp. 77–81.
ACM Press, New York (2004)
33. Kerren, A.: Learning by Generation in Computer Science Education. Journal of
Computer Science and Technology (JCS&T) 4(2), 84–90 (2004)
34. Kerren, A.: Course: Information Visualization (2007),
http://w3.msi.vxu.se/~ kerren/courses/lecture/ws06/infovis/
35. Kerren, A., Ebert, A., Meyer, J. (eds.): Human-Centered Visualization Environ-
ments. LNCS, vol. 4417. Springer, Heidelberg (2007)
36. Kerren, A., Stasko, J.T., Fekete, J.-D., North, C.: Workshop Report: Information
Visualization – Human-centered Issues in Visual Representation, Interaction, and
Evaluation. Information Visualization 6(3), 189–196 (2007)
37. Kerren, A., Stasko, J.T., Fekete, J.-D., North, C.J. (eds.): Information Visualiza-
tion. LNCS, vol. 4950. Springer, Heidelberg (2008)
38. KidPad: University of Maryland (2007), http://www.kidpad.org
39. Kosara, R.: Course: Visual Communication in Computer Graphics and Art (2007),
http://eagereyes.org/VisComm
40. Kosara, R.: Visualization Criticism: One Building Block for a Theory of Visualiza-
tion. In: Kerren, A., Stasko, J.T., Fekete, J.-D., North, C. (eds.) Abstracts Collec-
tion – Information Visualization - Human-Centered Issues in Visual Representa-
tion, Interaction, and Evaluation. Dagstuhl Seminar Proceedings, Dagstuhl, Ger-
many, vol. 07221, Internationales Begegnungs- und Forschungszentrum für Infor-
matik (IBFI) (2007), http://drops.dagstuhl.de/opus/volltexte/2007/1136
41. Krygier, J.B., Reeves, C., DiBiase, D.W., Cupp, J.: Design, implementation and
evaluation of multimedia resources for geography and earth science education.
Journal of Geography in Higher Education 21(1), 17–39 (1997)
42. Lichtenbelt, B., Crane, R., Naqvi, S.: Introduction to Volume Rendering. Prentice-
Hall, Englewood Cliffs (1998)
43. Ma, K.-L.: Course: Information Visualization (2006),
http://www.cs.ucdavis.edu/~ ma/ECS272/
90 A. Kerren, J.T. Stasko, and J. Dykes

44. Ma, K.-L.: Course: Information Interfaces (2007),

http://www.cs.ucdavis.edu/~ ma/ECS163/
45. MacEachren, A., Taylor, D.: Visualization in Modern Cartography. Pergamon,
Oxford (1994)
46. Marsh, S., Dykes, J.: Using usability to evaluate geovisualization for learning
and teaching. In: Proceedings of the GIS Research UK 13th Annual Conference
GISRUK 2005, Glasgow (2005)
47. Marsh, S., Dykes, J.: Visualization software works as a focus for fieldwork! evalu-
ating geovisualization software in a student-centred exercise with usability engi-
neering techniques. Journal of Geography in Higher Education (submitted)
48. Miksch, S.: Course: Information Visualization (2006),
http://www.ifs.tuwien.ac.at/~ silvia/wien/vu-infovis/
49. Monmonier, M.: How to Lie with Maps, 2nd edn. University of Chicago Press,
Chicago (1996)
50. Mullet, K., Sano, D.: Designing Visual Interfaces: Communication Oriented Tech-
niques. Prentice-Hall PTR, Englewood Cliffs (1994)
51. Munzner, T.: Course: Information Visualization (2006),
http://www.cs.ubc.ca/~ tmm/courses/infovis/
52. Munzner, T.: Process and Pitfalls in Writing Information Visualization Research
Papers. In: Kerren, A., Stasko, J.T., Fekete, J.-D., North, C.J. (eds.) Information
Visualization. LNCS, vol. 4950, Springer, Heidelberg (2008)
53. North, C.: Course: Information Visualization (2007),
http://infovis.cs.vt.edu/cs5764/
54. Plaisant, C., Fekete, J.-D., Grinstein, G.: Promoting insight-based evaluation of
visualizations: From contest to benchmark repository. IEEE Transactions on Vi-
sualization and Computer Graphics 14(1), 120–134 (2008)
55. Rana, S., Dykes, J.: Framework for augmenting the visualization of dynamic raster
surfaces. In: EuroSDR Commission 5 Workshop on Rendering and Visualisation
(Electronic Proceedings), Enschede, The Netherlands (2003)
56. Rushmeier, H., Dykes, J., Dill, J., Yoon, P.: Revisiting the need for formal edu-
cation in visualization. IEEE Computer Graphics and Applications 27(6), 12–16
(2007)
57. Schroeder, W., Martin, K., Lorensen, B.: The Visualization Toolkit – An Object-
Oriented Approach To 3D Graphics, 4th edn. Kitware, Inc (2006)
58. Schumann, H., Müller, W.: Visualisierung – Grundlagen und allgemeine Metho-
den. Springer, Heidelberg (2000)
59. Shakshuki, E., Kerren, A., Müldner, T.: Web-based Structured Hypermedia Al-
gorithm Explanation System. International Journal of Web Information Sys-
tems 3(3), 179–197 (2007)
60. Slocum, T.A., McMaster, R.B., Kessler, F.C.: Thematic Cartography and Geo-
graphic Visualization, 2nd edn. Prentice-Hall, Englewood Cliffs (2004)
61. Spence, R.: Information Visualization, 1st edn. ACM Press, New York (2001)
62. Spence, R.: Information Visualization: Design for Interaction, 2nd edn. Prentice-
Hall, Englewood Cliffs (2007)
63. Spotfire: TIBCO Software, Inc. (2007), http://spotfire.tibco.com
64. Stasko, J.T.: Course: Information Visualization (2006),
http://www.cc.gatech.edu/~ stasko/7450/
65. Tableau Desktop: Tableau Software, Inc. (2007),
http://www.tableausoftware.com
66. TableLens: Inxight Software, Inc. (2007),
http://www.inxight.com/products/sdks/tl/
 Teaching Information Visualization 91

67. Tufte, E.: Envisioning Information. Graphics Press, Cheshire (1990)

68. Tufte, E.: Visual Explanations: Images and Quantities, Evidence and Narrative.
Graphics Press, Cheshire (1997)
69. Tufte, E.: The Visual Display of Quantitative Information, 2nd edn. Graphics
Press, Cheshire (2001)
70. Ward, M.: Course: Data Visualization (2006),
http://web.cs.wpi.edu/~ matt/courses/cs525D/
71. Ware, C.: Information Visualization: Perception for Design, 2nd edn. Morgan
Kaufmann, San Francisco (2004)
72. Yang, K.: Course: Information Visualization (2007),
http://www.cs.uncc.edu/~ jyang13/infovis2007.html
73. Yi, J.S., Kang, Y.a., Stasko, J., Jacko, J.: Toward a deeper understanding of the
role of interaction in information visualization (Paper presented at InfoVis ’07).
IEEE Transactions on Visualization and Computer Graphics 13(6), 1224–1231
(2007)
74. Zerger, A., Bishop, I.D., Escober, F., Hunter, G.: A self-learning multimedia
approach for enriching gis education. Journal of Geography in Higher Educa-
tion 26(1), 67–80 (2002)
 Creation and Collaboration: Engaging New
Audiences for Information Visualization

Jeffrey Heer1 , Frank van Ham2 , Sheelagh Carpendale3 , Chris Weaver4, and
Petra Isenberg3
1
Electrical Engineering and Computer Sciences,
University of California, Berkeley,
360 Hearst Memorial Mining Building, Berkeley, CA 94720-1776, USA,
[email protected]
2
IBM Research, Visual Communications Lab,
1 Rogers Street, Cambridge, MA 02142, USA,
[email protected]
3
Department of Computer Science, University of Calgary,
2500 University Dr. NW, Calgary, AB, Canada T2N 1N4,
{sheelagh, petra.isenberg}@ucalgary.ca
4
GeoVISTA Center and the North-East Visualization and Analytics Center,
Department of Geography, Penn State University,
302 Walker Building, University Park, PA 16802, USA,
[email protected]

1 Introduction

In recent years we have seen information visualization technology move from

an advanced research topic to mainstream adoption in both commercial and
personal use. This move is in part due to many businesses recognizing the need
for more effective tools for extracting knowledge from the data warehouses they
are gathering. Increased mainstream interest is also a result of more exposure to
advanced interfaces in contemporary online media. The adoption of information
visualization technologies by lay users – as opposed to the traditional information
visualization audience of scientists and analysts – has important implications for
visualization research, design and development. Since we cannot expect each of
these lay users to design their own visualizations, we have to provide them tools
that make it easy to create and deploy visualizations of their datasets.
Concurrent with this trend, collaborative technologies are garnering increased
attention. The wide adoption of the Internet allows people to communicate across
space and time, and social software has attained a prominent position in con-
temporary thinking about the Web. For example, one can think of software
teams distributed over different time zones or multiple people collaborating to
build an online encyclopedia. Furthermore, collaborative issues are not limited
to the web: novel display and interaction technologies, including wall-sized and
tabletop interfaces, introduce new possibilities and challenges for co-located col-
laborators. An increased need for specialization means that we can no longer
rely on a single person to perform deep analyses of complex phenomena. These

A. Kerren et al. (Eds.): Information Visualization, LNCS 4950, pp. 92–133, 2008.


c Springer-Verlag Berlin Heidelberg 2008
 Creation and Collaboration 93

developments signify an increased desire for collaboration around complex data,

yet, information visualization tools are still primarily designed according to a
single user model. To meet the demands of an increasingly diverse audience, the
design of information visualization technologies will have to incorporate features
for sharing and collaboration.
In this paper, we discuss creation and collaboration tools for interactive visu-
alization. Our goal is to begin to characterize the increasingly diverse audience
for visualization technology and map out the design space for new creative and
collaborative tools to support these users. In section 2 we classify the expand-
ing user base for visualization technologies by looking at their skills, goals and
the data they are trying to analyze. We then take a look at existing informa-
tion visualization tools and classify them along these dimensions. In sections 3
and 4 we examine the new collaborative trends. Section 3 discusses co-located
collaboration, while section 4 explores the area of distributed, asynchronous col-
laboration on the Web. Finally, we conclude by considering the ways the research
community should respond to these developments.

2 End-User Creation of Visualizations

The term “end-user visualization” encompasses a broad range of visualization

users and use-cases. For example, a marketing executive might create an overview
of the sales in different product segments to show to his manager, a scientist may
create a coordinated visualization application to study a biomedical dataset,
or a Facebook user may present her social network in a visualization on the
site. All these use cases involve different types of users employing information
visualization to tackle different types of problems. If we wish to provide end-
users with the ability to construct and deploy custom information visualizations
of their own data, we need an understanding of these users, their goals, and their
data. In the following sections, we will broadly classify each of these dimensions.
Note that we do not intend to construct a formal taxonomy of users. Instead,
our goal is to broaden the discussion on who our users are and how visualization
can help them.

2.1 Data

Scientific, geographic, economic, demographic, and other domains of human

knowledge produce vast amounts of wildly different forms of information, varied
in terms of both individual interest and broad social importance. Visualization
seeks to provide perceptually and cognitively effective tools to display and inter-
act with these different kinds of data. Data is commonly categorized by inherent
complexity (e. g., data homogeneity, number of dimensions) or size. In this sec-
tion, however, we consider data from the perspective of users by categorizing
three different kinds of data in terms of potential audience.
94 J. Heer et al.

Personal Data: Personal data encompass all types of organized information

collections that are of personal interest to a particular user, but less interesting
to a broader community. This may involve data on user-owned media (such as
DVD collections or playlists), data on life organization (financial data or address
books) or data related to hobbies and general interests (photo collections, fitness
schedules or coin collections). Visualizing personal data might not always lead to
deep new insights about the data itself. In such cases, the visualization instead
may serve more as a compact visual artifact that can be used to remember
certain events in ones life and serve as a visual representation of self [95]. These
visual representations of self may then be used as online avatars, or simply as
catalysts for storytelling, much like photo albums.

Community Data: By community data we mean data that might be relevant

to a broad community of users due to similar interests or general appeal. Exam-
ples of community data include the content of political speeches, the number of
users online in a World of Warcraft realm, or voting results per county. Often
this type of data has a social component associated with it: data might be related
to a social application such as Facebook or MySpace [45], contain statistics on
a large population as with census data [43], or may be related to current events
[104]. Precisely because community data has a lot of general appeal it will often
generate a lot of discussion.

Scientific Data: Scientific data is data that is of interest to a (relatively) small

number of specialists. Traditionally, information visualization has focused on the
sciences, because they generate a wealth of structured and often numerical data
in ready need of analysis. This makes them very suitable to mathematical analy-
sis techniques and visual mapping. In the humanities, however, most information
comes in unstructured raw text format. If we want visualization to be applied in
domains such as literature and political science, we will need to define suitable
pre-processing techniques that can extract meaningful information from a body
of text. This will often require some amount of natural language processing or
expert input. While there are a few applications of information visualization to
data from the humanities (e. g., [101]), the area remains largely untapped despite
substantial promise to yield many useful techniques with applicability to many
different areas of everyday life.

Interplay of Data Types: Note that the distinction between types of data
is not always clear cut and many data sets could fall into different categories
depending on their use. For example, a community data set on World of Warcraft
users and their interactions might be considered a scientific data set by social
scientists, while the personal data of celebrities might have a broad general
appeal. Visualizations of all these types of data can be shared, albeit for different
purposes. Personal data might be shared with other users as a means of personal
expression. Community data is often shared to spark broad discussion, while
scientific data often needs to be shared because it is too complex for one person to
analyze on their own or because it requires multiple specialized skills to analyze.
 Creation and Collaboration 95

The recent trend toward visual analytics [91] is driven by the increasing need
to support open-ended management and exploration of large, loosely-connected,
and often unstructured information sources as well as the smaller, isolated, struc-
tured data sets typical of information visualization applications. Information col-
lection often involves assembling “shoeboxes” of loosely related nuggets and data
sets [107]. Visual analysis of information occurs by following chains of evidence,
evaluating formal hypotheses [27], testing competing explanations [86], or telling
stories [37] using visual metaphors to convey relationships and dynamics. These
activities are particularly challenging in intelligence analysis, emergency man-
agement, epidemiology, and other critical areas that involve high-dimensional
abstract information [83] and large geospatial datastores [36]. However, the het-
erogeneous and idiosyncratic nature of the data sets and analysis activities in
these endeavors are similar to those in everyday domains, making it likely that
the outcomes of visual analytics research will translate readily into visualization
approaches that will help to engage broad audiences.

2.2 Skills

Novice Users: By novice users we mean users who have experience operating
a computer, but no experience with programming in general, let alone program-
ming visualization techniques. The vast majority of novice visualization users
act as consumers: they will interact with the visualization within the possibil-
ities offered but will rarely extend existing functionality to suit their analysis
needs. If we want these users to be able to produce visualizations, we have to
take care to make this process as easy as possible. Some points of consideration
when designing visualizations for novice users are:

Data Input: We cannot expect a novice user to write their own data parser,
write database queries that export data to a particular format or understand
the file formats for more complex data types. Most novice users seem to take to
using spreadsheet programs such as Microsoft Excel to store and analyze their
data. One useful input format then, is a simple tab delimited input file, as this
format is both human readable and can be directly copied from the spreadsheet
editor.

Automatic Selection of Visualization Type: Novice users have no experience

designing visual mappings and may even choose mappings that produce non-
sensical visualizations. Recurring examples include the use of line charts over
categorical data dimensions, for which a bar chart would be a better choice, and
using a pie chart for data that do not form part of a whole. For this reason, visu-
alization techniques geared towards novice users should at least partly automate
the selection of visual metaphors. This may involve analyzing the data dimen-
sions to see if there are any ordinal attributes, check for aggregated variables
and totals, and examine values in dimensions for possible hierarchical structure
[59,60].
96 J. Heer et al.

Useful Defaults: Novice users likely will not spend time tuning an ugly looking
visualization to fit their needs. It is therefore important to provide a set of
sensible defaults for data and view parameters (such as scales, colors, item sizes
and viewpoints) to help constrain the parameter space that users have to explore.
Multiple combinations of these parameters can be offered by providing a preset
list. As an added bonus, a good set of presets can show users what is possible
and educate them on what is sensible.

Contextual Information: With contextual information we mean visual items that

explain to the user what data is being mapped to the screen and what encod-
ings are being applied. This involves legends, scales, labels, pop-ups, titles and
explanations of visual mappings. Although visual graphics in print media take
great care to provide contextual information, interactive visualizations are often
lacking in this respect because most of the design attention is focused on the
visual mapping itself.

Savvy Users: By savvy users we mean people who have experience performing
relatively sophisticated data organization and manipulation, using a combination
of manual processing and limited amounts of programming or scripting. Because
savvy users are a small but non-trivial part of the population of visualization
consumers, they are a critical bridge between experts and novices. As such, savvy
visualization users may act variously as:

– experts who train or guide novice users in the use of particular visualizations
by clarifying exploratory and analytic functionality in terms of interface
appearance and behavior,
– designers who plan, construct, debug, test, and deploy new visualizations
for ongoing evaluation and routine operation by novice users,
– end-users who can bring more extensive experience to bear when using ex-
isting visualizations to analyze data from their own knowledge domains, to
browse data with which they are less familiar, and to share their results with
others, and
– explorers (or user-designers) who combine the roles of designer and end-user
by extending and redesigning visualizations on the fly during open-ended
exploration of their data.

Expert Users: By expert users we mean people who have extensive experience
with interactive graphical software development and the theory and applica-
tion of data modeling, data processing, and visual data representation. As such,
visualization experts may act both as:

– researchers who invent, specify, and evaluate methods for accessing, query-
ing, rendering, and interacting with data, often with an eye toward extending
and enhancing the functionality of existing visualization systems and tools,
and
 Creation and Collaboration 97

– developers who design and implement visualization modules, toolkits, sys-

tems, and tools of various sizes and scopes, often adapting and integrating
existing functionality from other visualization toolkits and systems.
In particular, visualization research frequently involves the development of pro-
totypes for evaluating the correctness, flexibility, and performance of new data
processing algorithms and the usability and utility of new interaction techniques.
Facilitating the interdependent needs of novice, savvy, and expert users is
a key part of supporting broader audiences for information visualization. The
number of people who can act as visualization designers or visualization develop-
ers – let alone the core visualization researchers who by necessity often fill these
roles – is rapidly becoming overwhelmed by demand for visual tools brought on
by blossoming public awareness of the power and accessibility of information
visualization techniques. It will become increasingly necessary to provide users
of all skill levels, including novices, with the capability to explore and analyze
data sets of personal and professional interest without direct assistance from
traditional visualization practitioners. However, understanding how to design
accessible yet flexible software artifacts for individual visual exploration and
analysis is only half of this equation. Social organization of visualization roles
through collaboration and other means, as described later in this paper, is the
critical second half.

2.3 Goals
One of the traditional rationales for information visualization is that the human
visual system has high input bandwidth and has evolved as an excellent tool
for spotting patterns and outliers in our surroundings. If we then map large
amounts of data into visual form, we can use these innate human abilities to
explore the data to find patterns that would have been exceedingly difficult to
identify through purely automated techniques. A current prominent example is
bioinformatics research that visually explores gigabytes of gene experiments to
investigate the mechanisms that drive a particular disease. Such “explorative”
use-cases have dominated most of the research in visualization over the past
two decades. Explorative use can either be open-ended, where the user wants
to browse their data without having a predefined question in mind, or analyti-
cally driven, in which the user has a particular question in mind and uses the
visualization to answer it. Often times these two types of exploration will be
intertwined: a user will explore a previously unknown data set without a par-
ticular question in mind, stumble on an interesting data point and then use the
analytic features in the visualization to either answer the question or redirect
their open-ended exploration.

Exploration and Analysis: Recent visualization environments have begun

to offer users various degrees of interactive control over different parts of the
entire information interface design process, thereby opening up possibilities for
much deeper exploration of data. Such environments allow computer-savvy user-
designers to interactively access data, create, layout, and coordinate views, and
98 J. Heer et al.

connect data to views. Design typically occurs directly within the interface that
contains data views, and often take effect immediately without the need for
a separate compilation or build stage. This live, amodal approach to interface
design allows users to switch rapidly between building and browsing tasks during
exploration and analysis. The result is a form of exploration that is free form and
open-ended, particularly during initial inspection of newly encountered data sets.
IVEE [2], DEVise [58], DataSplash [71], Snap-Together Visualization [69],
GeoVISTA Studio [88], Improvise [102], and Tableau/Show Me [60] are a few
of many well-known visualization environments that support open-ended data
exploration to various degrees. Such environments typically consist of a graphic
user interface on top of a library of visualization components which may or
may not be exposed as a visualization programming toolkit in its own right.
This combination of user interface and underlying library can enable open-ended
exploration in a very broad sense if it bridges the activities of visualization users
performing various roles with different levels of expertise, whether as individuals
or in collaborative groups.
To connect developers and designers, a key advantage for open-ended explo-
ration is an extensible library that provides an application programming inter-
face (API) for adding new software modules for various visualization components
(including data access, queries and other data transformation algorithms, views,
and visual data encodings). In particular, the most useful APIs support the def-
inition of new data transformation operators—including appropriate input and
output data object types—that give designers the ability to express rich relation-
ships between data, queries, and views. This requirement is essential for applying
newly discovered visualization techniques to emerging sources and forms of in-
formation, without needing to constantly architect and implement new toolkits
(and retrain visualization designers in their use).
To connect designers with users, the user interface must support the abil-
ity to access data sets (and metadata) from local or remote sources in various
formats, create and position views on the screen, specify how navigation and
selection affects views, specify queries on data, parameterize queries in terms of
interaction, and attach data sets and queries to views. In particular, designers
should be able to specify the appearance and behavior of their visualizations
directly within the user interface, without resorting to programming or other
workarounds for interface limitations. To do otherwise would effectively require
that designers be trained as developers.
User interfaces that truly support open-ended exploration would exceed the
requirements of basic visualization design and operation by: supporting live
building of complete browser interfaces, including immediate designing, debug-
ging, and testing of intended functionality; facilitating collaboration between
end-users and designers to turn analytical questions into structural changes
(through remote, nearby, or side-by-side efforts to communicate and effect rapid
visualization prototyping and polishing); and enabling rapid switching between
building and browsing to perform more extensive exploratory visualization by
modifying visualization views and queries on the fly. In particular, it is highly de-
sirable for explorers to be able to see all raw data quickly to make decisions about
 Creation and Collaboration 99

how to visualize it, rapidly create and lay out views, rapidly attach data and
queries to views, rapidly modify queries, store, copy, and reuse views, copy-and-
paste/drag-and-drop visualization components, and use macros to build com-
mon multiple view constructions. Many of these capabilities are also desirable
for non-exploring designers who prepare visualizations for domain analysts.
In all of this, availability of common and familiar interface functionality is
essential to broad adoption. The user interface should run in the user’s normal
working environment, require no programming or design activities, and provide
a way to disseminate analytical results. For communication and collaboration,
it is highly desirable for the user interface to run easily on any platform, allow
visualizations to be opened and saved as normal documents for sharing between
users, and provide the ability to bookmark or screen capture visualizations in
different graphical states.

Communication: At the other end of the spectrum of information visualiza-

tion goals is the “communicative” use-case, where the main user goal is simply to
convey a message to others. This use-case is already present in many traditional
media: think of diagrams explaining the numbers behind a news story or a bar
chart that has been included in a slideshow presentation. Although these partic-
ular representations are fairly static because of the affordances of the media used,
this does not mean that communicative information visualization is limited to
static visualizations. Interactivity is a very useful means of engaging users and
may make them more receptive to a particular message. However, interactiv-
ity also poses some problems when communicating visualizations, because it’s
hard to reproduce interactive features in a static medium. Many information
visualizations use tooltips and mouse-overs to provide contextual information,
offer the user different viewpoints on the data, and allow for dynamic analysis of
data. Videos alleviate this problem only in part, because it is often hard to follow
what is going on and much of the context in the exploration process is missing.
Simply sharing findings using static representations of interactive visualization
is therefore not the optimal solution, and we would do better to consider these
issues beforehand when designing visualizations for communicative use.
Apart from traditional mass communication, the communicative use-case also
plays a pivotal role in collaborative applications, especially ones that are non-
collocated and/or asynchronous. If the analysts do not share the same time or
space it is important for them to be able to communicate findings and bring
each other up to speed on the current state of the process quickly. Furthermore,
each of these types of collaboration (collocated or distributed) has its own type
of requirements, which we discuss in-depth in sections 3 and 4. In general, com-
municative use of information visualization usually involves a small investment
of time on the user end, with a small but guaranteed payoff. On the other hand,
explorative use involves a large amount of investment in tools, training and time,
while the (potentially high) payoff is not always guaranteed. (See also [93] for a
discussion of these tradeoffs.)
In the next section, we consider a number of representative tools that help
us meet these differing goals of communication and exploration.
100 J. Heer et al.

Fig. 1. A few of the many available information visualization tools, roughly mapped
according to targeted end-user and targeted goal. Light lines connect toolkits and
development environments to examples of visualizations created in them. Dark lines
roughly capture similar ranges of user/goal targets for relevant tools.

2.4 Tools
Note that most real-world uses of information visualization will form a combina-
tion of the use-cases and roles described in the preceding sections. A researcher
might program a new visualization technique to explore his complex data and
then present findings to a manager by sending a screenshot. In this case the
researcher takes on the roles of both consumer and developer and performs both
exploration and communication. Most current information visualization tools
and toolkits are geared towards one particular user skill and goal, although a
recent trend towards more flexible tools can be observed. To illustrate the rough
classification outlined in the previous subsections, in this section we give an
indicative sample of an end-user visualization tool for each user skill and goal
combination. Figure Fig. 1 illustrates a number of available visualization tools
categorized according to the skill level of the target user base and the degree to
which the tools support analytic and communicative tasks. Systems that span
a range of tasks or skills are presented as line segments indicating the range of
users and usage.

Expert Communication: In the bottom right corner of the matrix we find

information visualization tools and toolkits that are geared towards commu-
nicative use, but assume a significant amount of knowledge on suitable visual
techniques and their implementation. One such toolkit is Adobe’s Flash develop-
 Creation and Collaboration 101

ment environment. Flash is an browser based interactive graphical design tool.

Because of its ease of online deployment, it is particularly suited to communi-
cating messages in a graphical way. In fact, many of the interactive graphics on
the Internet today are Flash-based. However, Flash does not offer the developer
tools that would support structured data exploration. (In fact, it offers only the
most basic of data structures.) Moreover, its timeline-based design environment
is not particularly well-suited to interactive visualization development.
This situation has improved with the release of Actionscript 3 and Flex,
offering a more advanced programming model and a full-fledged user interface
development package. The Flare toolkit [41] implements basic visualization ca-
pabilities within Flash, making it easier to develop interactive information vi-
sualizations, while still retaining the benefits of Flash, such as its relatively
lightweight means of online deployment.
Savvy Communication: Information visualization tools in this category (mid-
dle right) allow users to create and share complex information visualizations, but
require a base level understanding of computer programming and information
visualization. A concrete example of such a tool is AT&T’s GraphViz [33] li-
brary, which allows users to generate static images of graphs but requires some
programming effort to integrate it with existing applications because it uses a
proprietary data format.
Use-cases for GraphViz often involve reporting engines that need to be able
to display networked data of some sort. Many of the features in GraphViz are
geared towards presentation instead of exploration. For example, it is possible
to heavily customize node rendering. Special care has been taken to avoid label
overlapping, as this would make static images completely unreadable. Both of
these issues are less of a problem in interactive systems in which users can use
tooltips to get more information, or zoom into a dense cloud of labels to remove
overlap.
Novice Communication: Until recently, if novice users wanted to share infor-
mation visualizations with others they would be limited to taking screenshots of
information graphics for sending by e-mail, etc. This mode is often sufficient if
the goal is one-way information dissemination. For example, a pie chart may be
included in a presentation, or an advanced information graphic may be printed
in a newspaper. However, this mode of publication fails if users want to collab-
oratively analyze a complex data set.
Recent tools like Many Eyes [97], Sense.us [46], Swivel.com, and Spotfire
Decision Site Posters make this process much easier, allowing easy sharing of
interactive visualizations. As discussed in greater detail in section 4, users of
these systems can share a particular state of a visualization encoded as a URL
and add custom annotations and comments while still having access to the in-
teractive features of the visualization. This makes it possible to quickly switch
between analysis and communication, a necessity for successful collaboration.
Expert Exploration and Analysis: The bottom left corner of the matrix
contains visualization software that supports deep and broad exploration of the
102 J. Heer et al.

space of visualization techniques, as well as more focused exploration and anal-

ysis of particular data sets. Such flexibility in the overall process of visualization
almost always requires substantial expertise, typically requiring programming
skills. As a result, visualization software for exploration by experts often takes
the form of toolkits that are written in a popular programming language but
that encapsulate well-known visualization components and techniques.
One such toolkit, prefuse [44], provides a Java-based library of visualization
building blocks including views, visual encodings, processing algorithms, multi-
view coordinations, and a common data model that supports tables, trees, and
graphs. Graphs, hyperbolic trees, treemaps, and scatter plots support accessing,
filtering, rendering, and displaying data using a variety of layout and distortion
algorithms.
Similarly, the InfoVis Toolkit [34] is a set of Java visualization components
designed around OpenGL and a data model that represents tables, trees, graphs,
and metadata in column format for efficient selection, filtering, visual encod-
ing, and coordination. Views include scatter plots, parallel coordinate plots,
treemaps, and a variety of node-edge tree and graph displays that can incorpo-
rate fisheye lenses and dynamic labeling of items. Visualizations created in the
toolkit display textboxes, sliders, and other controls alongside views for dynamic
editing of visual encodings.
The extensible programming interfaces of both toolkits and those like them
provide a means to incorporate new components and techniques, in essence ex-
panding the scope of exploration, considered broadly, to include the results of
future visualization research.

Savvy Exploration and Analysis: As described in the previous section, visu-

alization in the expert exploration category revolves more around programming
rather than around interaction in integrated user interfaces intended for design-
ing and building tools. Research on integrated visualization environments focuses
on packaging the exploratory capabilities of toolkits in ways that are accessible
to users who are visualization savvy but not necessarily visualization experts.
For instance, Improvise [102] is a self-contained Java application that appears
and behaves like other office productivity applications based on the multiple doc-
ument desktop metaphor. Users build Improvise visualizations by interactively
constructing the data, queries, views, and coordinations of tools that can be
saved, opened, copied, and shared as self-contained Extensible Markup Language
(XML) documents. Users browse visualizations using the mouse and keyboard
to navigate and select data items in multiple coordinated views.
Similarly, GeoVISTA Studio [88] is an integrated visualization development
environment for building geovisualizations interactively using a graph-based vi-
sual coordination editor. Any component that conforms to the JavaBeans specifi-
cation can be a view. Development of new views by the community of GeoVISTA
Studio users has resulting in a large library of views utilized in numerous visual-
izations. A particular strength of GeoVISTA Studio is its extensive functionality
for representing and displaying geospatial information (based on the GeoTools
[57] open source Java GIS toolkit).
 Creation and Collaboration 103

The combination of browsing with rapid, iterative building in a single applica-

tion (much like in spreadsheet programs) enables improvisational visualization,
in which it is possible to design and evaluate different ways of analyzing partic-
ular data sets in the form of rapid prototypes having more concrete and stable
collections of analytic functionality.

Novice Exploration and Analysis: As far as we know, there are no tools

that truly allow novice users to interact with their data in the broadest sense
of exploration. This may result from an apparent fundamental tradeoff between
flexibility and accessibility in visual analysis, in that increased expressiveness ne-
cessitates greater expertise when it comes to data manipulation and visual repre-
sentation. Even in savvy exploration-analysis tools like Improvise that strive for
a balance between these factors, reproducing many common visual components
and techniques currently requires a high degree of visual language expressiveness
that necessitates a corresponding high level of expertise beyond that of most
novice users. Conversely, novice analysis-communication tools like Many Eyes
seek to increase visualization flexibility for broad audiences keenly interested in
modest analytic expressiveness as a means to better communicate ideas about
information. In between, analysis tools like Tableau/Show Me successfully oc-
cupy analysis niches that provide bounded but particularly useful forms of data
interaction to relatively broad audiences who are sufficiently motivated to devote
time and effort to modest training. It may well be that open-ended exploration
tools for novices will evolve from future research into ways of combining these
three seemingly complementary directions.

2.5 Directions
Current end user visualization tools are becoming more and more flexible in
the types of scenarios and goals they can handle. Tools like Many Eyes allow
novice users to create advanced visualizations with very little effort and also
support communicative use-cases by allowing flexible sharing of visualization
states. Tools like Improvise allow tight integration of many different types of
visualizations, but require some programming skills on the side of the end-user,
an expectation that is not always reasonable of domain experts dealing with the
visualization. Tableau allows end users to set up and pivot different types of ba-
sic visualizations in a fairly intuitive manner and the recent addition of Tableau
Server allows sharing of and commenting on these visualizations in an online envi-
ronment, making it also suitable for communicative purposes. Although flexible,
the only visualization types allowed are 2-dimensional small-multiple displays,
which limits the visualization and analysis types to basic business graphics.
In our opinion, the ultimate goal of letting novice users flexibly specify their
visualization needs and couple different types of views together has not been fully
realized yet. We expect that users’ visual literacy will increase as information
visualization becomes more mainstream, and will start demanding advanced
visualizations beyond the trusted bar chart. Integrating advanced visualizations
in an flexible, collaborative and easy to understand framework for open-ended
104 J. Heer et al.

exploration and analysis is an important and solvable problem. We expect this

solution will have important implications for many areas of human endeavor that
necessitate the handling of complex data.

3 Co-located Collaborative Visualization

Given the choice, it is common and natural for people to work together. This is
not a new phenomenon. Small groups of people gather for all kinds of reasons
including many that are work related; such as to get a job done faster, to share
expertise for a complex task, and to benefit from different insights from different
people. Also, when one considers the rapid growth in size and complexity of
datasets, it is not surprising that increasingly the practicality of an individual
analyzing an entire data-set is becoming unrealistic. Instead, the expertise to
analyze and make informed decisions about these information-rich datasets is
often best accomplished by a team [91]. For instance, imagine a team of medical
practitioners examining a patient’s medical record to plan an operation, a team
of biologists looking at test results to find causes for a disease, or a team of
businessmen planning next year’s budgets based on a large financial dataset. All
of these situations involve a group of people making use of visual information to
proceed with their work. Research towards supporting these team-based infor-
mation processes will expand the situations in which information visualization
can be used and is part of considering how to best support people in their normal
everyday information work practices.
This section draws from a wide variety of literature to shed light on questions
and issues that need to be considered during the development of co-located
collaborative information visualizations. We do not consider this discussion to be
exhaustive; rather it is our intention that the discussion will form the beginning of
design guidelines and considerations that will be modified and extended through
future research in collaborative information visualization.
Research in information visualization draws from the intellectual history of
several traditions, including computer graphics, human-computer interaction,
cognitive psychology, semiotics, graphic design, statistical graphics, cartogra-
phy, and art [64]. The synthesis of relevant ideas from these fields is critical
for the design and evaluation of information visualization in general and it is
only sensible to think that fields concerned with collaborative work also add
valuable information to our understand-ing of requirements for collaborative in-
formation visualization systems. Our sources include work in co-located collab-
oration in computer supported cooperative work [39,53,75,73,76,77,80,81,82,90],
information visualization [85,105,109,110,111], and empirical work investigating
collaborative visualization use [61,68,72].
The organization of this section is as follows. A brief overview of existing
research that relates to co-located collaborative information analysis is given in
section 3.1. Next, section 3.2 discusses the impact of recent advances in hardware
configurations and section 3.3 focuses on more general human computer inter-
action issues important for the support of the co-located collaborative process,
 Creation and Collaboration 105

primarily drawing upon computer supported co-located collaborative literature.

Then section 3.4 presents information visualization specific issues that may need
re-consideration in light of co-located collaborative applications.

3.1 Related Research

Co-located collaborative information visualization is a relatively new and still
under explored research area. Only a few tools designed specifically to support
synchronous collaboration between co-located people using visualizations to ex-
plore information have emerged thus far. These are discussed below. However,
as noted below, existing visualization tools designed from a single-user perspec-
tive have been studied with co-located collaborative tasks [61]. There has been
considerable research in the area of scientific visualization in distributed systems
(see [38] for an overview). Recently, there has been new primarily web-based re-
search on asynchronous distributed collaborative information visualization sys-
tems. This new direction is the focus of section 4.

Co-located Collaborative Visualization: The Responsive Workbench was

one of the first visualization systems for developed co-located collaboration
around a large horizontal surface [103]. The responsive workbench is a virtual
reality environment in which the displayed 3D scene is seen through shuttered
glasses and interaction is achieved with a glove which has an attached Polhemus
sensor on the back. Agrawala et al. [1] extended this workbench to support two
simultaneous users. Several scientific visualization applications were developed
for this platform including fluid dynamics and situational awareness applications.
On tabletop displays information visualization interaction techniques have
been used to support co-located people in information sharing and exploration
tasks. DTLens [35] provides a local non-linear magnification technique enabling
multiple lenses for up to four people with for two-handed interaction. Personal
Digital Historian uses radial layouts to display photos, video and text documents
to supports conversation and story telling for small groups of people [84].

Studying Collaborative Use of Information Visualizations: While re-

search on collaborative data analysis using information visualizations is relatively
scarce, collaborative use of existing single user systems has been studied. Mark
and Kobsa [61] conducted a user study in which they observed pairs working
in co-located and distributed settings with two different visualization systems
designed for single users. Their findings suggest that the benefit of collaborative
vs. individual problem solving was heavily dependent on the visualization sys-
tem used and also that, in general, groups were better than individuals working
alone at locating errors. From this study, they derive a model for the collabora-
tive problem-solving process. Their model consists of an iterative sequence of five
stages: parsing a question, mapping variables to the program, finding the cor-
rect visualization, and two validation stages. From studying collaborative work
on scientific visualizations in virtual environments using CAVEs, Park et al. [72]
report a five-step activity model that was common for the observed collaboration
106 J. Heer et al.

sessions. Their study also noted that participants showed a strong tendency for
independent work, if the option was available. Isenberg et al. studied co-located
collaborative data analysis scenarios and posit an eight-process framework that
relates to previous work on the Sensemaking Cycle [17] and the two studies by
Mark and Kobsa [61] and Park et al. [72]. However, a common temporal order
of analysis processes as posited by some previous work did not emerge.

3.2 Choosing Hardware to Support Co-located Collaboration

We start with a discussion of hardware because some of the recent interest in
co-located collaboration is at least in part due to new hardware innovations.

Display Size: In information visualization, the size of the available display

space has always been problematic for the representation of large datasets (e. g.,
[65]). In a common desktop environment, typically a single user will use all avail-
able screen space to display their visualization and, most commonly, this space
will not be sufficient. Frequently, visualization software will include interactive
features to help the user cope with limited display space. It seems sensible to
think that, if we are going to adequately support collaborative or team explo-
ration of visualizations, available display space will be an important issue. In
collaborative systems, screen space not only has to be large enough for the re-
quired information display, it might also have to be viewed and shared by several
users. As the number of people using a shared information display grows, the
size of the display and workspace needs to be increased in order to provide a
viewing and interaction area that gives adequate access to all group members.

Display Configuration: Several configuration possibilities exist that could

increase the amount of available display space, all of which will affect the type of
visualization systems possible and the type of collaboration work that would be
most readily supported. Many types of configurations are possible; for instance,
one could provide team members with interconnected individual displays, as in
the ConnecTable system [89], or one could make use of large, interactive, single-
display technology, like display walls or interactive tabletop displays (e. g., [90]).
An additional possibility is to link wall, table, and personal displays (e. g., [105]),
or to consider immersive displays (e. g., [72]). The type of setup most appropriate
for an information visualization system will depend on the specific task and group
setup. For example, individual interconnected displays allow for private views of
at least parts of the data which might be required if data access is restricted.
Tabletop displays have been found to encourage group members to work together
in more cohesive ways, whereas wall displays are beneficial if information has to
be discussed with a larger group of people [76].

Input: In the common desktop setup, input is provided for one person through
one keyboard and one mouse. To support collaboration, ideally, each person would
have at least one means of input. In addition, it would be helpful if this input was
identifiable, making it possible to personalize system responses. If a collaborative
 Creation and Collaboration 107

system supports multi-user input, the access to a shared visualization and data set
has to be coordinated. Also, synchronous interactions on a single representation
may require the design and implementation of new types of multi-focus visualiza-
tions. Ryall et al. [77] have examined the problem of personalization of parameter
changes for widget design, allowing widgets to be dynamically adapted for indi-
viduals within a group. Similar ideas could be implemented for personalization of
information visualizations during collaborative work.

Resolution: Resolution is an issue both for the output (the display) and for the
input. The display resolution has a great influence on the legibility of information
visualizations. Large display technology currently often suffers from relatively
low display resolution so that visualizations might have to be re-designed so that
readability of text, color, and size not affected by display resolution. Also, large
interactive displays are often operated using fingers or pens which have a rather
low input resolution. Since information visualizations often display large data
sets with many relatively small items, the question of how to select these small
items using low input resolution techniques becomes an additional challenge that
needs special attention [48].

3.3 Creating a Collaborative Environment

The key characteristics of co-located synchronous interactions as described by
Olson and Olson [70] will apply to information visualization scenarios designed
to support co-located collaboration. These characteristics include: a shared local
context in which participants can interact with work objects, rapid feedback,
and multiple channel information exchange (voice, gesture, etc.), and visibility
of others’ actions. These characteristics are further specified in the mechanics of
collaboration [73], which describe basic operations of teamwork, or small scale
actions and interactions that help people solve a task as a team. These mechan-
ics apply to a variety of group and task settings. This section is discussed under
the two major groupings of the mechanics of collaboration—communication and
coordination—and under those issues relating to supporting varying collabora-
tion styles.

Communication: Communication is an important part of successful collab-

orations. People need to be able to trigger conversations, communicate their
intentions, indicate a need to share a visualization, and to be generally aware
of their team members’ actions. Group members need to be informed that some
parameter of a shared display might have changed while they were busy working
with an information visualization in a different part of the workspace. There has
been considerable discourse on the importance of all team members being aware
of all other team members’ actions. Pinelle et al. [73] make a distinction between
explicit and implicit communications. The design of support techniques for both
types of communication needs to respect common social and work protocols [70].
For example, the interface should not require a group member to reach into or
across another person’s workspace in order to acquire or share visualizations or
controls.
108 J. Heer et al.

Explicit Communication: Enabling direct exchange of information through many

channels such as voice, gestures, and deictic references facilitates collaborative
work in general [73] as well as co-located collaboration [70]. It has been shown
that the ability to annotate data and share insights in a written way is an
essential part of the discovery process in distributed information visualization
settings [46]. This collaborative need for annotation exists in traditional use of
pen and paper based information as was observed in a study of teams working on
information analysis tasks in a shared setting [68]. However, in digital systems
messages of all types, written, voice, etc. might not always be as easily shared
and how best to support this will require further research.

Implicit Communication: In co-located non-digital collaboration people are ac-

customed to gathering implicit information about team members’ activities
through such things as body language, alouds, and other consequential commu-
nications. This is an active research area in distributed collaboration research
since the co-located evidence does not naturally become distributed. Co-located
collaboration benefits from many of the co-present advantages, however, issues
still arise. Some examples include: digital actions are not always readily visible
(cursors are hard to see on large screens), menu actions can affect a remote
part of the screen, as well as the general problems of change awareness [74].
Thus while implicit communications do support awareness in a co-located set-
ting already to some extent, some system changes made by a collaborator can
still remain unnoticed if the collaborative system does not provide appropriate
feedthrough (i. e. a reflection of one person’s actions on another person’s view).
In collaborative information visualization, for example, it might be important to
consider appropriate awareness for operations that make changes to the under-
lying dataset.
Imagine a co-located system in which each collaborator works in parallel on
a different view using a different file-system representation. If one collaborator
discovers an old version of a file and decides to delete it (a value operation
[23]), this change might go unnoticed if the other person is looking at a view
of the data that does not include the current file or it might be completely
surprising to the other person to see a file in their representation disappear.
Some research has proposed policies to restrict certain members from making
unsuspected global changes to a dataset [75]; however, while earlier research
on information visualization discussed the differences between view and value
operators (e. g. [23]), most recent research in multiple-view visualization tends
to favour view operations (filtering of unwanted data rather than deletion). This
seems likely to be most appropriate during collaboration.
It has also been shown the location and orientation of artifacts is used to
support implicit communication in non-digital settings providing information on
such things as who is working with which artifacts and when one person wants to
initiate communication about a particular artifact [53] and that this translates
to digital settings [54]. This consideration, providing for artifact mobility and
freedom orientation, will probably also be important in supporting information
visualization collaboration.
 Creation and Collaboration 109

Coordination: In group settings, collaborators have to coordinate their actions

with each other. Here, we describe several guidelines for how to support the
coordination of activities in collaborative information visualization applications.

Workspace Organization: Typical single-user information visualization systems

impose a fixed layout of windows and controls in the workspace. Previous re-
search has shown that, on shared workspaces, collaborators tend to divide their
work areas into personal, group, and storage territories [81]. This finding im-
plies that a group interaction and viewing space is needed for collaborative data
analysis where the group works on a shared representation of the data or in
which they can share tools and representations. Also, the possibility of explor-
ing the data separately from others, in a personal space, is necessary. Flexible
workspace organization can offer the benefit of easy sharing, gathering, and pass-
ing of representations to other collaborators. By sharing data in the workspace,
representations will be viewed by team members with possibly different skill
sets and experiences and, therefore, subjected to different interpretations. Also,
by being able to move and rotate representations in the workspace, an individ-
ual can gain a new view of the data and maybe discover previously overlooked
aspects of the data display.
Collaborative information visualization systems should allow for social in-
teraction around data displays [46]. If visualizations can be easily shared, team
members with different skill sets can share their opinions about data views, sug-
gest different interpretations, or show different venues for discovery. By offering
mechanisms to easily rotate and move objects, comprehension, communication,
and coordination can be further supported [54]. Rotation can support compre-
hension of a visualization by providing alternative perspectives that can ease
reading and task completion, coordination by establishing ownership and cat-
egorizations, and communication by signaling a request for a closer collabora-
tion [53]. By allowing free repositioning, re-orientation we can also make use of
humans’ spatial cognition and spatial memory and possibly better support in-
formation selection, extraction, and retrieval tasks [68]. Mechanisms for transfer
and access to information visualization in the workspace should be designed in
a way that they respect common social work protocols [53,81].

Changing Collaboration Styles: Tang et al. [90] describe how collaborators

tend to frequently switch between different types of loosely and closely coupled
work styles when working over a single, large, spatially-fixed information display
(e. g., maps or network graphs). A study by Park et al. [72] in distributed CAVE
environments discovered that, if the visualization system supports an individual
work style, users preferred to work individually on at least parts of the problem.
For information visualization systems, an individual work style can be supported
by providing access to several copies of one representation. The availability of
unlimited copies of one type of representation of data allows group members
to work in parallel. More closely coupled or joint work on a single view of the
data can be supported by implementing the possibility of concurrent access and
interaction with the parameters of an information visualization. Free arrange-
110 J. Heer et al.

ments of representations also support changing work styles. Representations can

be fluidly dragged into personal work areas for individual or parallel work and
into a group space for closer collaboration.

3.4 Designing Information Visualizations for Co-located

Collaboration
Many known information visualization guidelines still apply to the design of in-
formation visualizations for co-located collaborative use (e. g., [10,92,99]). In this
section, we discuss changes and additions to aspects that need to be considered
when designing information visualizations for co-located collaborative settings.
Thus, much of this discussion simply delineates research questions that may of
specific interest when designing information visualizations to support co-located
collaboration.

Representation Issues: Spence [87] defines representation as “the manner in

which data is encoded,” simplifying Marr’s [62] definition of representation as a
formal system or mapping by which data can be specified. The concept of rep-
resentation is core to information visualization since changes in representations
cause changes in which types of tasks are most readily supported. As in Marr’s
[62] example, the concept of thirty-four can be represented in many ways. To
look at three of them; Arabic numerals, 34, ease tasks related to powers of ten;
Roman numerals, XXXIV, simplify addition and subtraction; and a binary rep-
resentation, 100010, simplifies tasks related to powers of two. Not surprisingly,
Zhang and Norman [110] found that providing different representations of the
same information to individuals provides different task efficiencies, task com-
plexities, and changes decision-making strategies. Questions arise as to what are
the most effective representations during collaboration. Will certain representa-
tions be better suited to support small group discussions and decision making?
Will multiple representations be more important to support different people’s
interpretation processes? Will new encodings or representations be needed for
collaborative work scenarios? Appropriate representations might have to be cho-
sen and adapted depending on the display type chosen but whether completely
new designs are required is not yet clear.
For example, different representations may have to be accessible in an inter-
face because in a collaborative situation, group members might have different
preferences or conventions that favour different types of representations. Gutwin
and Greenberg [39] have discussed how different representations of the workspace
affect group work in a distributed setting. They point out that providing mul-
tiple representations can aid the individual but can restrict how the group can
communicate about the objects in the workspace. This extends to co-located set-
tings, in which several representations of a dataset can be personalized according
to taste or convention, making it harder to relate individual data items in one
representation to a specific data item in another. For example, relating one spe-
cific node in a treemap [50] to another node in a node-link diagram might require
a search to locate the respective node in the other representation. Implementing
 Creation and Collaboration 111

mechanisms to highlight individual data items across representations might aid

individuals when switching between group and parallel data exploration.
Findings suggest that the availability of multiple, interactively accessible
representations might be important for information visualization applications
since the availability of multiple data representation can change decision mak-
ing strategies [52]. Also differing representations have an influence on validation
processes in information analysis [79], and more easily support people working
in parallel on information tasks [72]. While this is probably applicable, empirical
evidence directly linking these finding to collaborative information visualization
has not yet been gathered.
It is also possible that the actual mappings used in representations may have
to be re-thought. For example, spatiality or the use of position/location is com-
monly an important aspect of representation semantics. However, spatiality as
manifested in territoriality is a significant factor for communication and coordi-
nation of small group collaboration. It is an open question as to whether there
is a trade-off between these two uses of spatiality.

Presentation Issues: Presentation has been defined as ’something set forth for
the attention of the mind’ [63] and as ’the way in which suitably encoded data
is laid out within available display space and time’ [87]. From these definition
is clear that changing display configurations, as is usually the case to support
co-located collaboration, will impact the types of presentations techniques that
are possible and/or appropriate. Common presentation techniques include pan
& zoom, focus & context, overview & detail, filtering, scrolling, clutter reduc-
tion, etc.
A common theme in information visualization is the development of pre-
sentation techniques that overcome the problem of limited display space (e. g.
[4,20,49]). In collaborative scenarios, information visualizations might have to
cover larger areas than in a single user scenario as group members might prefer
to work in a socially acceptable distance from each other. The display space
might also have to be big enough to display several copies of one representation
if team members want to work in parallel.
If groups are working over a shared presentation of data, presentations might
have to be adapted to allow collaborators to drill down and explore different parts
of the data in parallel. Collaborative information visualizations will likely have
to sup-port multiple simultaneous state changes. This poses additional problems
of information context. Team members might want to explore different parts of
a dataset and place different foci if the dataset is large and parts of the display
have to be filtered out. Information presentations might have to be changed to
allow for multi-focus exploration that does not interfere with the needs of more
than one collaborator. For example, DOI Trees [18] or hyperbolic trees [55] are
examples of tree visualizations in which only one focus on the visualization is
currently possible. ArcTrees [67] and TreeJuxtaposer [65], for example, allow
for multi foci over one tree display but these were not designed to take the
information needs of multiple collaborators into account and might still occlude
valuable information.
112 J. Heer et al.

An example for visualization presentation changes based on a collaborative

circular tabletop environment has been presented in [94]. The presentation of
the circular node-link tree layout was modified to rotate all nodes towards the
boundary and a “magnet” was implemented to rotate nodes towards just one
team member. Nodes were also changed in size; as leaf nodes were placed closer
towards team members, in their personal space [81], they were decreased in size
and the nodes towards the center of the table were enlarged to allow for easier
shared analysis of the node contents in the group space [81]. A possible extension
of this work is to think about placing and re-arranging nodes automatically
based on the placement and discovery interests of team members or based on
the individual or shared discoveries that have been made.
The presentation of visualizations might also have to take available input
devices on a shared large display into account. If fingers or pens are used as
an input device, the selection might not be accurate enough to select small
information items. A common task in information visualization is to re-arrange
data items (e. g. by placing points of interest), to request meta-information [85]
(e. g. by selecting an item), or to change display parameters by selecting an
item. If the displayed dataset is large, it often covers the full screen and reduces
individual items to a few pixels. Previous research has attempted to solve the
issue of precise input for multi-touch screens (e. g. [8]) but they might not be
applicable if the whole visual display is covered with items that can possibly
be selected. Alternatively, information presentations could be changed to allow
for easier re-arrangement and selection of items, for example, with lenses [20].
DTLens [35] presents an initial exploration of the use of lenses in co-located
collaboration.
The resolution of a large display has an influence on the legibility of data
items. It is known that the reading of certain visual variables is dependent on the
size and resolution in which they are displayed [99]. Information visualizations
also often rely on textual labels to identify data items which may be hard to
read on low-resolution displays. The presentation size of individual items and
labels may have to be adapted to compensate for display resolution.

View Issues: The term view is common in information visualization literature

and view operations (changing what one currently sees) have been defined as
distinct from value operations (changing the underlying data) [23], however, this
use of the term view also incorporated changes in visual aspects of representation,
presentation. Blurring the distinction between view and presentation changes has
not been problematic because with a single viewer and a single display these are
often concurrent. A change in view can be simply looking at exactly the same
presentation and representation of the same data merely from a different angle
or it can include changes in all three factors.
In a co-located collaborative setting, of necessity there are as many views
of a given presentation as there are people in the group. Also since collabora-
tion practices often include mobility, a given person’s view will change as they
move in the physical setup. This factor has recently begun to receive attention
in the CSCW community. Nacenta et al. [66] have shown that righting (orient-
 Creation and Collaboration 113

ing a piece of 2D information into the proper perspective, by means of motion

tracking, really aids comprehension. Hancock and Carpendale [40] consider the
same problem for horizontal displays looking for non-intrusive interactive solu-
tions. Since a study by Wigdor et al.[105] has indicated that angle of viewing
affects readability of certain visual variables; this issue will be an important one
for collaborative information visualizations. This research on how view-angle
distortion affects perception in a single and multi-display environment suggests
that certain types of representations may need to be modified in order to be used
on a digital tabletop display and that information visualizations should not be
compared across multiple display orientations. However, as visual variables were
tested in isolation (e. g. length, direction, only) further evaluations have to be
conducted to see whether participants will correct for possible distortion if the
variables are presented in conjunction with others or whether view correction
[66,40] might compensate.
Visualizations that can be read from multiple angles and orientations (e. g.
circular tree layouts vs. top-down layouts) might be more appropriate for display
on a horizontal surface. However, it is not clear whether participants would try to
read oriented visualizations upside down and make wrong conclusions based on
these readings or whether they would simply re-orient the visualization to correct
the lay-out. Observations of collaborative information visualization scenarios
point in the latter direction [68].
Gutwin and Greenberg [39] discuss issues about viewing a representation in
relation to distributed scenarios. However, parallels can be drawn to co-located
scenarios in which collaborators work with multiple linked copies of the same
representation of a dataset. These essentially represent multiple movable “split
viewports” [39]. The suggested solutions for distributed settings include radar
views, overview+detail solutions, and cursor eye-view. Whether the benefit of
these solutions in a co-located setting outweighs the possible distractions they
might create, however, would have to be evaluated. Further guidelines for using
multiple views in information visualization can be found in [5] and provide a
starting point for tailoring multiple views for collaborative visualization.
Issues of view may develop in collaborative information visualization settings
if collaborators want to switch from loosely-coupled to closely-coupled workstyles
[90] and share discoveries they have made with the other group members. If one
collaborator worked with different view of the dataset it might be difficult to
locate the information in the other persons’ view. Another important factor to
consider when developing a collaborative viewing strategy is the establishment
of territories [81] for personal, group and storage purposes that is suggested as
beneficial for group coordination (see “Coordination” above),
A study by Yost and North [109] compared the ability of visualizations to
display large amounts of data normalized across either a small or a large high-
resolution vertical display. Their study showed that the visualizations used were
perceptually scalable but that people preferred different visualizations on the
large vs. small display, as some were found to be easier to read than others de-
pending on the screen size. How these preferences would change for collaborative
work would have to be evaluated.
114 J. Heer et al.

Interaction Issues: Most interaction issues deal with interaction with repre-
sentations, presentations and views, thus discussing them here would overlap
with points raised under these headings. However, there are some more general
interaction issues. When people are co-located, they are in the situation in which
people naturally collaborate, the situation in which people have collaborated for
centuries. When face-to-face, people naturally know how to collaborate and are
so used to picking up subtle cues from each other that they may do this without
even being conscious of the precise details of the underlying coordination and
communication practices that are in play. As the developers of co-located collab-
orative information visualizations, our task is to facilitate information access and
exploration without interfering with the social protocols that make collaboration
effective. However, to do this we have to understand what these social collabo-
ration practices are and specifically if there are any differences when people are
collaborating using visual information. Some factors are:
Interactive Response Rates: Information visualization has always had a lot of
requirements in that it deals with extremely large and complex data sets and
in that it can have considerable graphics requirements for these complex rep-
resentations. Adding larger screens, more screens, higher pixel counts, multiple
simultaneous inputs, and possibly multiple representations will increase compu-
tational load adding more requirements to the challenge of maintaining good
interactive rates. Thus implementations of collaborative information visualiza-
tions will have to be carefully designed for efficiency. While continued hardware
advances will mitigate this to some extent, it will be important to address issues
in both efficient data processing and fast graphic rendering.
Interaction History: A history task has been defined as a task that involves
keeping a history of actions to support undo, replay, and progressive refinement
[85]. In a collaborative scenario keeping such a history can have other benefits.
If a visualization tracks and reveals which data items have been visited and by
whom this information could be valuable for collaborators helping them under-
stand their team members’ actions, find unexplored parts of a visualization or to
confirm discoveries made by others. A visualized interaction history may support
collaboration by promoting mutual understanding of team members involvement
in the task [24] and may help keep group members aware of each others actions
as people shift from individual to shared views of the data [39]. An exploration
history can be useful in such activities as validating work done, in explaining a
discovery process to other team members, and in supporting discussions about
data explorations.
Information Access: Exactly how to handle information access is an important
collaboration issue. The main themes in the research discussion thus far have
been motivated by social protocol issues and data centric concerns. While these
have not been seen as mutually exclusive they are quite distinct ideas. The so-
cial protocol theme has made considerable use of observational studies to better
understand exactly what are the social protocols and how do they impact col-
laboration. These understandings are then used as a basis for software design.
 Creation and Collaboration 115

The data centric approach discusses factors such as who has (or does not have)
rights to which parts of the data?, who can change the scale, zoom, or rotation
settings for a shared view of the data? And how does a data item get passed
between team members (hand-off). Restriction has been suggested as a means to
stop certain members from making unsuspected global changes to the data that
might change other members’ view of the same data [75]. Similar issues pertain-
ing to workspace awareness (individual vs. shared views), artefact manipulation
(who can make which changes), and view representation have been raised [39].
Is a single shared representation adequate? Should a system allow for multiple
representations? Should the exploration on multiple representations of the same
dataset be linked or be completely independent?

Fluid Interaction: The fluidity of interactions in a shared workspace influences

how much collaborators can focus on their task rather than on the manipulation
of interface items [82]. This implies that in a collaborative information analysis
scenario, parameter changes to the presentation or representation of a dataset
should require manipulation of as few interface widgets (menus, slider, etc.) as
possible and little or no changes of input modalities (mouse, keyboard, pen,
etc.). A study on collaborative information visualization systems has similarly
reported that groups worked more effectively with a system in which the required
interactions were easier to understand [61]. This poses a challenge to information
visualization tool designers as typically a high number of parameters are required
in visualization systems to adapt to the variability in dataset complexity, size,
and user tasks.

4 Collaborative Visualization on the Web

Visual analysis is rarely a solitary activity. A business analyst may notice an

unexpected trend in a chart of sales figures – but then she’s likely to confer with
a colleague, who may share the chart with a manager, who later might present
it to executives. Such scenarios of collaboration and presentation across both
time and space are common in business and scientific visualization. Just as a
good visualization takes advantage of the power of the human visual system,
it can also exploit our natural social abilities. Accordingly, designers of visu-
alization systems should consider not only the space of visual encodings, but
mechanisms for sharing and collaboration. At minimum, systems should enable
people to communicate about what they see so they can point out discoveries,
share knowledge, and discuss hypotheses.
The social aspects of visualization have taken on new importance with the
rise of the Web. While collaboration in small groups remains ubiquitous, it is
now also possible for thousands of people to analyze and discuss visualizations
together. These scenarios are driven by the fact that users can interact remotely
from anywhere on the globe and access the system at different times. Parti-
tioning work across time and space holds the potential for greater scalability
of group-oriented analysis. For example, one decision making study found that
116 J. Heer et al.

asynchronous collaboration resulted in higher-quality outcomes – broader dis-

cussions, more complete reports, and longer solutions – than face-to-face collab-
oration [6].
Web-based collaboration around visualizations introduces new challenges for
research, as most work on collaborative visualization has been done in the context
of synchronous scenarios: users interacting at the same time to analyze scientific
results or discuss the state of a battlefield. As described in the previous section,
co-located collaboration usually involves shared displays, including large wall-
sized screens and table-top devices (e. g., [28,31]). Systems supporting remote
collaboration have primarily focused on synchronous interaction [3,14]), such
as shared virtual workspaces (e. g., [1,24]) and augmented reality systems that
enable multiple users to interact concurrently with visualized data (e. g., [25,9]).
In addition, the increasing availability of table-top and large public displays has
prompted researchers to experiment with asynchronous, co-located visualization
(same place, different time), often in the form of ambient information displays
(e. g., [21]).
In this section, we instead focus on the kind of collaboration that is most
common over the Web: remote collaboration across time and space. Our goal
is to summarize the work done to date and indicate promising research direc-
tions. We first review recent web-based systems supporting social data analysis
around visualizations, highlighting the collaborative features provided by these
systems and how they have been used in practice. We then discuss a number of
outstanding challenges for asynchronous collaborative visualization and identify
avenues for future research.

4.1 Web-Based Collaborative Visualization Systems

Though web-based collaboration around visualizations is still in its infancy, a

handful of commercial and research systems in this area have recently been
introduced. Here we discuss contemporary visualization systems that support
asynchronous collaborative analysis (shown in Fig. 2), documenting the collab-
orative features supported by these tools and initial reports of their usage.

DecisionSite Posters: Adding Collaboration to a Single-User Tool: De-

cisionSite Posters is a feature of the Spotfire product sold by TIBCO, Inc.
Users of Spotfire’s desktop-based visualization system can capture snapshots
of their analyses and publish them on an intranet as “posters.” View sharing
is supported, as each poster has a unique URL than can be easily distributed.
Each poster also sup-ports unthreaded text comments on a side panel. However,
posters do not allow annotations, limiting the ability of collaborators to point
at specific trends or outliers.
As described in [96], the communication capabilities in DecisionSite Posters
have been used in an unexpected way. Instead of engaging in complex conversa-
tions by using the comment panel, as envisioned by the system’s designers, users
have largely used the tool for presenting their findings to colleagues. The ability
 Creation and Collaboration 117

Fig. 2. Asynchronous Collaborative Visualization Systems. Clockwise from top-left:

Spotfire DecisionSite Posters, Wikimapia, Swivel, sense.us, and Many Eyes.

to create comments with pointers into the visualization provides an easy way to
choreograph a step-by-step presentation.

Swivel: Sharing Data on the Web: Swivel.com is a web site that supports
sharing and discussion around data. The service appears to be modeled on sites
such as YouTube that support sharing of other media. In keeping with this
model, Swivel allows users to upload data sets and talk about them in attached
discussion forums. In addition, the site automatically generates graphs by com-
bining columns from uploaded data sets into bar charts, pie charts, and scatter
plots. Pointing behavior on the site appears limited.
Although the graphs on Swivel are not interactive, the site provides an exam-
ple of social data analysis in action, in particular the importance of collaborative
publishing and sharing of visualizations. While there do not seem to be many
extensive conversations in Swivel’s discussion area there has been significant use
of Swivel’s graphs among bloggers to discuss statistics. In other words, it appears
that the ability to publish graphs for use in other contexts is most valuable to
Swivel’s users.

Wikimapia: Collaborative Geographic Annotation: Wikimapia.org is a

web site enabling collective annotation of geographic satellite imagery, and is
representative of similar efforts such as Google Earth and mash-ups created with
web APIs to mapping services. The site provides a zoomable browser of satellite
118 J. Heer et al.

photos across the globe, along with the ability to select geographic regions for
annotation with names and additional data (Fig. 2). View sharing is supported
through automatically updating URLs. As the view is panned or zoomed, the
current URL updates dynamically to reflect the current zoom level and latitude
and longitude values. Pointing is supported through annotations. Users can draw
rectangular and polygonal annotations, which scale appropriately as the map is
zoomed. To avoid clutter, annotations are filtered as the view is zoomed; the
viewer does not see annotations that are too small to be legible or so large they
engulf the entire display, improving the scalability of the system.
Wikimapia supports conversation using an embedded discussion technique.
Each annotation is a link to editable text. Descriptive text about a geographic
region can then be edited by anyone, similar to articles on Wikipedia. Discussion
also occurs through voting. When annotations are new, users can vote on whether
they agree or disagree with the annotation. Annotations that are voted down
are removed from the system. For instance, the small town of Yelapa, Mexico
is located on an inlet in a bay near Puerto Vallarta. However, the bay has a
number of inlets very close together. As a result, multiple conflicting annotations
for Yelapa appeared. Through voting, the incorrect regions were discarded and
the correct annotation was preserved.

sense.us: Social Data Analysis of U.S. Census Data: Sense.us is a proto-

type web application for social visual data analysis [46]. The site provides inter-
active visualizations of 150 years of United States census data, including stacked
timelines, choropleth maps, and population pyramids. With a URL-based book-
marking mechanism, it supports collaboration through doubly-linked discussion,
graphical annotations, bookmark trails, and searchable comment listings.
Discussion occurs via a doubly-linked conversation model. Searchable com-
ment listings provide links back into the visualization, while navigating in a visu-
alization automatically causes related comments to be retrieved for the current
view. By tying commentary directly to specific view states, comments become
less ambiguous, enabling remarks such as “that spike” or “on the left” to be more
easily understood. Pointing occurs through freeform graphical annotations and
view sharing is facilitated by URLs that automatically update as users navigate
the visualizations. Sense.us also allows users to collect view links in a “bookmark
trail” of view thumbnails. Users can then drag-and-drop view thumbnails from
the trail into a comment text field, thereby adding a hyperlink to the saved view
within the comment. In this way, users can provide pointers to related views and
create tours through the data.
Studies of the sense.us system revealed interesting patterns of social data
analysis. Users would make observations about the data, often coupled with
questions and hypotheses about the data. These comments often attracted fur-
ther discussion. For example, within a visualization of the U.S. labor force over
time, a spike and then decline in the number of dentists prompted discussion
ranging from the fluoridation of water to stratification within the dentistry pro-
fession, with a rise in the number of hygienists corresponding to the decline of
dentists. There was also an interesting interaction between data analysis and
 Creation and Collaboration 119

social activity. Users who tired of exploring visualizations turned their focus to
the comment listings. Reading others’ comments sparked new questions that led
users back into the visualization, stimulating further analysis. The sense.us pro-
totype was initially available on a corporate intranet which provided employees
with blogs and a social bookmarking service. Users of sense.us found ways to
publish their findings, typically by taking screenshots and then placing them on
blogs or the bookmarking service with application bookmarks. These published
visualizations drew additional traffic to the site.

Many Eyes: Web-Based Visualization and Publishing: Many-Eyes.com

[97] is web-based service that combines public data sharing with interactive
visualizations. Like social data analysis sites such as Swivel, site members can
upload data sets and comment on them. Unlike Swivel, however, Many Eyes
offers a palette of interactive visualization techniques – ranging from bar charts
to treemaps to tag clouds – that visitors may apply to any data set. Users may
post comments on the visualizations, including bookmarks for particular states.
The pointing and discussion capabilities of Many Eyes are used in a variety
of ways. The site contains some lengthy conversations around visualizations,
although the great majority of visualizations have no comments. One class of
visualizations, however, did lead to lengthy onsite discussions: visualizations that
sidestepped sober analysis and were instead playful or comical. One person, for
example, initiated a game based on a visualization of Shakespearean poetry in
which he used the highlighting mechanisms to pick out alphabetically ordered
words to make pseudo-Elizabethan epithets. These games frequently attracted
many “players.”
The Many Eyes site also can be viewed as a publishing platform, since the
visualizations that users create are publicly visible and may be linked to from
other web pages. Many bloggers have taken advantage of this, and perhaps as
a result the deepest analyses of Many Eyes visualizations have occurred as part
of blog entries that reference the site. In one example, a blog at the Sunlight
Foundation (a political reform organization) published a Many Eyes tag cloud
to analyze messages between the U.S. Congress and the U.S. Department of
Defense. The blog entry framed the results as a funny-but-sad surprise: the
most common phrases had nothing to do with current pressing issues, but rather
requests for congressional travel. In another case, a user created a visualization
of the “social network” of the New Testament. Not only was this visualization
linked to from more than 100 blog entries, but another user went to the trouble
of recording a YouTube video of himself interacting with the visualization and
narrating what he learned. These phenomena again underscore the importance
of publishing mechanisms for collaborative visualization.

Summary: The previous examples of web-based collaborative visualization

present a number of common design decisions, but also some important dif-
ferences. All systems support view sharing through URL bookmarks and enable
discussion through text comments. Furthermore, usage examples from these sys-
tems suggest that users derive great value from being able to share and embed
120 J. Heer et al.

visualizations in external media such as blogs. One salient difference between

systems is the varied forms of pointing within visualizations: selecting individ-
ual items in Many-Eyes, creating polygonal geographic regions in Wikimapia,
and drawing freeform graphics in sense.us. Another difference is the way com-
mentary is attached to visualization views. Spotfire Decision Site Posters, Swivel,
and Many Eyes all support blog-style unthreaded comments for individual visu-
alizations. In contrast, Wikimapia supports commentary attached to geographic
annotations, while sense.us provides threaded comments tied to specific states
of the visualization and retrieved dynamically during exploration.

4.2 Research Issues in Web-Based Collaborative Visualization

As described in the previous section, developers of collaborative visualization

systems face design decisions of how to support discussion, annotation, and in-
tegration with external services. Future research in asynchronous collaborative
visualization needs to provide guidance through this design space, as well as
develop novel techniques for better facilitating collaborative analysis. In this
section, we identify five areas in which we expect additional research to make
important contributions to improving the state-of-the-art: the structure and in-
tegration of collaborative contributions; engagement and incentives; coordina-
tion and awareness; pointing and reference; and presentation, dissemination,
and story-telling.

Structuring and Integrating Contributions: A fundamental aspect of suc-

cessful collaboration is an effective division of labor among participants. This
involves both the segmentation of effort into proper units of work and the al-
location of individuals to tasks in a manner that best matches their skills and
disposition. Primary concerns are how to split work among multiple participants
and meaningfully aggregate the results.
Drawing on examples such as online discussions, open source software, and
Wikipedia, Benkler [7] introduces the concepts of modularity, granularity, and
cost of integration in the peer production of information goods. Modularity refers
to how work is segmented into individual units of contribution, while granularity
refers to the scope of these units and how much effort they require. For exam-
ple, in online scenarios where incentives tend to be small and non-monetary, a
small granularity may encourage people to participate in part due to the ease
of contributing. The cost of integration refers to the effort required to usefully
synthesize contributions into a greater whole. Collaborative work will only be
effective if the cost of integration is low enough to warrant the overhead of mod-
ularization while enforcing adequate quality control. There are a number of mu-
tually inclusive approaches to handling integration: automation (automatically
integrating work through technological means), peer production (casting inte-
gration as an additional collaborative task given to trusted participants), social
norms (using social pressures to reduce vandalistic behavior), and hierarchical
control (exercising explicit moderation).
 Creation and Collaboration 121

Collaborative visualization can similarly be viewed as a process of peer pro-

duction of information goods. Stages in this process include uploading data sets,
creating visualizations, and conducting analysis. To support this process, it is
important to identify the specific forms of contribution (modules) that users
might make and how to integrate these contributions. Existing frameworks for
aiding this task include structural models of visualization design and sensemak-
ing processes [17]. As shown in Fig. 3, each of these models suggests tasks that
contribute to collaborative analysis, including data cleaning, moderation, visual-
ization specification, sharing observations, positing hypotheses, and marshaling
evidence. These concerns are given further treatment in [43].
Once modules have been identified, one can then attempt designs which re-
duce the cost structure of these tasks. Consider the issue of scale. Most of the
examples in the previous section use sequential text comments to conduct ana-
lytic discussion. However, it is unclear how well this form of communication will
scale to massive audiences. An open research problem is the creation of new forms
of managed conversation that have a lower cost of integration, enabling people to
understand and contribute to analysis without having to wade through hundreds
of individual comments. For example, Wikipedia relies on human editing coupled
with a revision management system to integrate and moderate contributions. Al-
ternatively, systems with highly structured input such as NASA ClickWorkers
[7] or von Ahn’s (2006) “games with a purpose” [98] rely on purely automated
techniques. Some middle ground between these approaches should be possible
for collaborative analysis, such as argumentation systems that model hypothe-
ses and evidence as first class objects. One example of such a system is CACHE
[11], which maintains a matrix of hypotheses and evidence, with collaborators
providing numerical measures of the reliability of evidence and the degree to
which evidence confirms or disconfirms the hypotheses. These scores can then
be averaged to form a group assessment. Other possibilities include augment-
ing graphical workspaces such as the Analysis Sandbox [107] with collaborative
authoring features or automatic merging of representations (c.f., [13]).

Engagement and Incentives: If collaborators are professionals working

within a particular context (e. g., financial analysts or research scientists) there
may be existing incentives, both financial and professional, for conducting col-
laborative work. In a public goods scenario, incentives such as social visibility or
sense of contribution may be motivating factors. Incorporating incentives into
the design process of collaborative visualization systems may increase the level
of user contributions, and could even provide additional motivation in situations
that already have well established incentive systems.
Benkler posits an incentive structure for collaborative work consisting of mon-
etary, hedonic, and social-psychological incentives [7]. Monetary incentives refer
to material compensation such as a salary or cash reward. Hedonic incentives
refer to well-being or engagement experienced intrinsically in the work. Social-
psychological incentives involve perceived benefits such as increased status or
social capital.
122 J. Heer et al.

(a) Collaborative activity might be introduced at any (b) The sensemaking model in
phase of the information visualization pipeline. [17] can be applied to identify
potential mechanisms for col-
laborative analysis (e. g., [43])

Fig. 3. Models of the Visualization Process.

Observations of social use of visualization have noted that visualization users

are attracted to data which they find personally relevant [45,96,100]. For ex-
ample, in collaborative visual analysis of the occupations of American workers
[46]), users often search for their own profession and those of their friends and
family, similar to how people search for names in the popular NameVoyager vi-
sualization [100]. The hypothesis is that by selecting data sets or designing their
presentation such that the data is seen as personally relevant, usage rates will
rise due to increased hedonic incentive. For example, geographic visualizations
may facilitate navigation to personally relevant locations through typing in zip
codes or city names, while a visualization of the United States’ budget might
communicate how a specific user’s taxes were allocated rather than only listing
total dollar amounts.
In the case of social-psychological incentives, the visibility of contributions
can be manipulated for social effects. Ling et al [56] found that users contributed
more if reminded of the uniqueness of their contribution or if given specific chal-
lenges, but not under other theoretically-motivated conditions. Cheshire [22] de-
scribes a controlled experiment finding that, even in small doses, positive social
feedback on a contribution greatly increases contributions. He also found that
visibility of high levels of cooperative behavior across the community increases
contributions in the short term, but has only moderate impact in the long term.
These studies suggest that social-psychological incentives can improve contribu-
tion rates, but that the forms of social visibility applied have varying returns.
One such incentive for visual analysis is to prominently display new discover-
ies or successful responses to open questions. Mechanisms for positive feedback,
such as voting for interesting comments, might also foster more contributions.
Finally, it is worth considering game play as an additional framework for
increasing incentives. In contrast to environments such as spreadsheets, many
 Creation and Collaboration 123

visualizations already enjoy game-like properties, being highly visual, highly in-
teractive, and often animated. Heer [42] discusses various examples in which
playful activity contributes to analysis, applying insights from an existing the-
ory of playful behavior [16] that analyzes the competitive, visceral, and teamwork
building aspects of play. For example, scoring mechanisms could be applied to
create competitive social-psychological incentives. Game design might also be
used to allocate attention, for example, by creating a team-oriented “scavenger
hunt” analysis game focused on a particular subject matter. Salen and Zim-
merman [78] provide a thorough resource for the further study of game design
concepts.

Coordination and Awareness: An important aspect of collaborative action

is awareness of others’ activities, allowing collaborators to gauge what work has
been done and where to allocate effort next [29,19]. Within asynchronous con-
texts, participants require awareness of the timing and content of past actions.
This suggests that designs should include both history and notification mecha-
nisms (e. g. [15]) for following actions performed on a given artifact or by specific
individuals or groups. Browseable histories of past action are one viable mech-
anism, as are subscription and notification technologies such as RSS (Really
Simple Syndication).
User activity can also be aggregated and abstracted to provide additional
forms of awareness. Social navigation [30] involves the use of activity traces to
provide additional navigation options, allowing users to purposefully navigate
to past states of high interest or explore less-visited regions (the “anti-social
navigation” of Wattenberg & Kriss [100]). For example, navigation cues may be
added to links to views with low visitation rates or to action items such as unan-
swered questions and unassessed hypotheses. One recent study [106] provides
evidence that social navigation cues can simultaneously promote revisitation of
popular or controversial views while also leading to a higher rate of unique dis-
coveries. Future research is needed to further develop and evaluate other forms
of awareness cues for supporting collaborative analysis.

Pointing and Reference: When collaborating around visual media, it is com-

mon for one to refer to visible objects, groups, or regions [26,12]. Such references
may be general (“north by northwest”), definite (named entities), detailed (de-
scribed by attributes, such as the “blue ball”), or deictic (pointing to an object
and saying “that one”). Hill and Hollan [47] discuss the various roles that deictic
pointing gestures can play, often communicating intents more complicated than
simply “look here”. For example, different hand gestures can communicate angle
(oriented flat hand), intervals (thumb and index finger in “C” shape), groupings
(lasso’ing a region), and forces (accelerating fist). While other forms of reference
are often most easily achieved through speech or written text, deictic reference
in particular offers important interface design challenges for collaborative visu-
alization. Nuanced pointing behaviors can improve collaboration by making it
easier to establish the object of conversation. Hill and Hollan argue for “gener-
124 J. Heer et al.

ally applicable techniques that realize complex pointing intentions” by engaging

“pre-attentive vision in the service of cognitive tasks.”
A standard way to point in a visualization is brushing: selecting and high-
lighting a subset of the data. Naturally, these selections should be sharable as
part of the state of the visualization. In addition, a palette of visual effects richer
than simple highlighting can let users communicate different intents. For exam-
ple, following time-varying values of selected items in a scatter plot is easier
when the selected items leave trails as they move over time. The selected items
and their trails are even more salient if non-selected items are simultaneously
de-emphasized. Brushing-based forms of pointing have the advantage that the
pointing action is tied directly to the data, allowing the same pointing gesture
to be reapplied in different views of the same data. As “data-aware” annotations
are machine-readable, they can also be used to export subsets of data and help
steer automated data mining [108].
Freeform graphical annotations are a more expressive method of pointing in
visualizations. Drawing a circle around a cluster of items or pointing an arrow at
a peak in a graph can direct the attention of remote viewers; at the same time,
the angle of the arrow or shape of the hand-drawn circle may communicate emo-
tional cues or add emphasis. However, while such drawing and vector graphic
annotations allow a high degree of expression, they only apply to a single view
in the visualization, without any explicit tie to the underlying data. Freeform
annotations can persist over purely visual transformations such as panning and
zooming, but they are not data-aware and may become meaningless in the face
of data-oriented operations such as filtering or drill-down. A promising research
direction is hybrid approaches that combine aspects of both brushing and graph-
ical annotation. The resulting techniques could create graphical annotations that
are tied to data points so that they can be reapplied in other views of the data.

Presentation, Dissemination, and Story-Telling: Common forms of infor-

mation exchange in group sensemaking are reports and presentations. Narrative
presentation of an analysis “story” is a natural and often effective way to commu-
nicate findings, and has been observed as a primary use of Decision Site Posters.
Furthermore, usage of Swivel, sense.us, and Many Eyes leverages external me-
dia such as blogs and social bookmarking services as additional communication
channels in which to share and discuss findings from visualizations. The challenge
to collaborative visualization is to provide mechanisms to aid the creation and
distribution of presentations. For example, sense.us [46] allows users to construct
and share trails of related views to create tours spanning multiple visualizations
and the GeoTime Stories [32] system supports textual story-telling with hyper-
links to visualization states and annotations. However, neither system yet allows
these stories to be exported outside the respective applications. In future work,
such mechanisms could be improved with support to build presentations semi-
automatically using interaction histories, export such presentations into external
media, and apply previously discussed pointing techniques. A related issue is to
enable follow-up analysis and verification for parts of the analysis story, enabling
presentations to serve as a catalyst for additional investigation.
 Creation and Collaboration 125

4.3 Summary
In this section, we introduce an emerging use of interactive visualization: collab-
orative visual analysis across space and time. The Web has opened up new possi-
bilities for large-scale collaboration around visualizations and holds the potential
for improved analysis and dissemination of complex data sets. A new class of
systems explores these possibilities, enabling web-based data access, exploration,
view sharing, and discussion around both static and interactive visualizations.
Already, these systems exhibit the promise of web-based collaboration, provid-
ing examples of collective data analysis in which group members combine their
knowledge to make sense of observed data trends and disseminate their findings.
Still, many research questions remain on how to structure collaboration. For
example, how can we move beyond simple textual comments to better scale and
integrate diverse contributions? Interested readers may wish to consult [96,46,43]
for further discussions on this topic. As described in section 2, another open
question is how to design for particular audiences. Different scenarios – includ-
ing scientific collaboration, business intelligence, and public data consumption –
involve different skill sets, scales of collaboration, and standards of quality. Going
forward, case studies in these scenarios are crucial to better tailoring visualiza-
tion tools to such varied audiences. By enabling users to collectively explore
data, share views and findings, and debate competing hypotheses, the resulting
collaborative visual analysis systems hold the potential to improve the number
and quality of insights gained from our ever-increasing collections of data.

5 Conclusion
The adoption of visualization technologies by people from different walks of life
has important implications for visualization research and development. Visual-
ization construction tools are lowering barriers to entry, resulting in end-user
created visualizations of every kind of data set imaginable. Concurrently, new
technologies enabling collaborative use of visualizations in both physical and
online settings hold the potential to change the way we explore, analyze, and
communicate. In this paper, we have sought to identify these emerging trends
and provide preliminary design considerations for advancing the state-of-the-art
of visualization and visual analytic tools.
As a parting comment, we note that the release of visualization tools “into
the wild” will undoubtedly result in a plethora of unexpected developments.
Equipped with new creation and collaboration tools, users will almost certainly
re-appropriate these technologies for unexpected purposes. Already, use of sys-
tems like Many-Eyes has revealed new genres of data-oriented play and self-
expression that complement more traditional analytic activities.
As researchers, it is imperative that we interface with these developments
in a productive fashion. It is likely that visualization tools will not only be
used in unexpected ways, but in ways we actively dislike. As new audiences
are exposed to visualization technologies, “bad” or “chart junk” visualizations
will be generated. Furthermore, visualizations will be used to support actions
126 J. Heer et al.

or points of view we may find distasteful, and any communication medium that
is sufficiently powerful to inform may also be used to lie or misrepresent. We
as a community should not be so concerned with trying to control the medium
or prevent people from lying or creating bad visualizations. As audiences get
more comfortable communicating with visualizations, we optimistically expect
the quality of visualizations and nuance of interpretation to improve.
However, this proscription does not mean that researchers should idly sit on
their hands. Rather, there will be an expanded role for visualization experts to
play. Issues of data provenance, cleaning, and integrity will force the research
community to focus on the visualization pipeline in a more holistic manner.
Supporting data at varied levels of structure will become increasingly necessary.
New genres of visualization use may require new designs and new systems to
support emerging practices, and the design of visual exploration tools that both
empower and educate will take on new importance. Consequently, the entrance
of visualization technologies into the mainstream offers a new horizon of research
opportunities.

References
1. Agrawala, M., Beers, A.C., McDowall, I., Fröhlich, B., Bolas, M., Hanrahan, P.:
The Two-User Responsive Workbench: Support for Collaboration Through In-
dividual Views of a Shared Space. In: International Conference on Computer
Graphics and Interactive Techniques (Siggraph ’97), pp. 327–332. ACM Press,
New York (1997), doi:10.1145/258734.258875
2. Ahlberg, C., Wistrand, E.: IVEE: An information visualization & exploration
environment. In: Proceedings of the IEEE Symposium on Information Visualiza-
tion, Atlanta, GA, October 1995, pp. 66–73. IEEE Computer Society Press, Los
Alamitos (1995)
3. Anupam, V., Bajaj, C.L., Schikore, D., Shikore, M.: Representations in distributed
cognitive tasks. IEEE Computer 27(7), 37–43 (1994)
4. Artero, A.O., Ferreira de Oliveira, M.C., Levkowitz, H.: Uncovering clusters in
crowded parallel coordinates visualizations. In: Proceedings of the IEEE Sympo-
sium on Information Visualization (InfoVis), pp. 81–88. IEEE Computer Society
Press, Los Alamitos (2004)
5. Baldonado, M.Q.W., Woodruff, A., Kuchinsky, A.: Guidelines for using multiple
views in information visualization. In: Proceedings of AVI ’00, pp. 110–119. ACM
Press, New York (2000), doi:10.1145/345513.345271
6. Benbunan-Fich, R., Hiltz, S.R., Turoff, M.: A comparative content analysis of face-
to-face vs. asynchronous group decision making. Decision Support Systems 34(4),
457–469 (2003)
7. Benkler, Y.: Coase’s penguin, or, linux and the nature of the firm. Yale Law
Journal 112(369) (2002)
8. Benko, H., Wilson, A.D., Baudisch, P.: Precise Selection Techniques for Multi-
Touch Screens. In: Proceedings of the Conference on Human Factors in Comput-
ing Systems (CHI’06), Montréal, Canada, April 22-27, 2006, pp. 1263–1272. ACM
Press, New York (2006), doi:10.1145/1124772.1124963
9. Benko, H., Ishak, E.W., Feiner, S.: Collaborative mixed reality visualization of
an archaeological excavation. In: IEEE International Symposium on Mixed and
Augmented Reality (ISMAR 2004), Arlington, VA, pp. 132–140 (2004)
 Creation and Collaboration 127

10. Bertin, J.: Semiology of Graphics: Diagrams Networks Maps (Translation of:
Sémiologie graphique). The University of Wisconsin Press, Madison (1983)
11. Billman, D., Convertino, G., Shrager, J., Pirolli, P., Massar, J.P.: Collaborative
intelligence analysis with cache and its effects on information gathering and cog-
nitive bias. In: Human Computer Interaction Consortium Workshop (2006)
12. Brennan, S.E.: How conversation is shaped by visual and spoken evidence. In:
Trueswell, Tanenhaus (eds.) Approaches to studying world-situated language use:
Bridging the language-as-product and language-as-action traditions, pp. 95–129.
MIT Press, Cambridge (2005)
13. Brennan, S.E., Mueller, K., Zelinsky, G., Ramakrishnan, I.V., Warren, D.S., Kauf-
man, A.: Toward a multi-analyst, collaborative framework for visual analytics. In:
IEEE Symposium on Visual Analytics Science and Technology (2006)
14. Brodlie, K.W., Duce, D.A., Gallop, J.R., Walton, J.P.R.B., Wood, J.D.: Distrib-
uted and collaborative visualization. Computer Graphics Forum 23(2), 223–251
(2004)
15. Brush, A.J., Bargeron, D., Grudin, J., Gupta, A.: Notification for shared anno-
tation of digital documents. In: Proc. ACM Conference on Human Factors in
Computing Systems (CHI’02) (2002)
16. Caillois, R.: Man, Play, and Games. Free Press of Glencoe (1961)
17. Card, S., Mackinlay, J.D., Shneiderman, B.: Readings In Information Visualiza-
tion: Using Vision To Think. Morgan Kauffman Publishers, Inc., San Francisco
(1999)
18. Card, S.K., Nation, D.: Degree-of-Interest Trees: A component of an attention-
reactive user interface. In: Proceedings of the Working Conference on Advanced
Visual Interfaces, May 2002, pp. 231–245. ACM Press, New York (2002),
http://www2.parc.com/istl/projects/uir/pubs/items/UIR-2002-11-Card-
AVI-DOITree.pdf
19. Caroll, J., Rosson, M.B., Convertino, G., Ganoe, C.H.: Awareness and teamwork
in computer-supported collaborations. Interacting with Computers 18(1), 21–46
(2005)
20. Carpendale, M.S.T., Montagnese, C.: A framework for unifying presentation
space. In: Schilit, B. (ed.) Proceedings of ACM Symposium on User Interface
Software and Technology (UIST), pp. 61–70. ACM Press, New York (2001),
http://pages.cpsc.ucalgary.ca/~sheelagh/personal/pubs/2001/carpendal
euist01.pdf, doi:10.1145/502348.502358
21. Carter, S., Mankoff, J., Goddi, P.: Building connections among loosely cou-
pled groups: Hebb’s rule at work. Journal of Computer-Supported Cooperative
Work 13(3), 305–327 (2004)
22. Cheshire, C.: Selective incentives and generalized information exchange. Social
Psychology Quarterly 70(1) (2007)
23. Chi, E.H.-H., Riedl, J.T.: An Operator Interaction Framework for Visualization
Systems. In: Wills, G., Dill, J. (eds.) Proceedings of the IEEE Symposium on
Information Visualization (InfoVis ’98), pp. 63–70. IEEE Computer Society Press,
Los Alamitos (1998)
24. Chuah, M.C., Roth, S.F.: Visualizing Common Ground. In: Proc. of the Conf. on
Information Visualization (IV), pp. 365–372. IEEE Computer Society Press, Los
Alamitos (2003)
25. Chui, Y.-P., Heng, P.-A.: Enhancing view consistency in collaborative medical
visu-alization systems using predictive-based attitude estimation. In: First IEEE
International Workshop on Medical Imaging and Augmented Reality (MIAR’01),
Hong Kong, China (2001)
128 J. Heer et al.

26. Clark, H.H.: Pointing and placing. In: Kita, S. (ed.) Pointing. Where language,
culture, and cognition meet, pp. 243–268. Lawrence Erlbaum, Mahwah (2003)
27. Cluxton, D., Eick, S.G., Yun, J.: Hypothesis visualization. In: Proceedings of the
IEEE Symposium on Information Visualization (Posters Compendium), Austin,
TX, October 2004, pp. 9–10. IEEE Computer Society Press, Los Alamitos (2004)
28. Dietz, P.H., Leigh, D.L.: Diamondtouch: A multi-user touch technology. In: Proc.
ACM Symposium on User Interface Software and Technology, pp. 219–226 (2001)
29. Dourish, P., Belotti, V.: Awareness and coordination in shared workspaces. In:
Proc. ACM Conference on Computer-Supported Cooperative Work, Toronto, On-
tario, pp. 107–114 (1992)
30. Dourish, P., Chalmers, M.: Running out of space: Models of information naviga-
tion. In: Proc. Human Computer Interaction (HCI’94) (1994)
31. Dynamics, G.: Command post of the future. Website (accessed November 2007)
32. Eccles, R., Kapler, T., Harper, R., Wright, W.: Stories in geotime. In: Proc. IEEE
Symposium on Visual Analytics Science and Technology (2007)
33. Ellson, J., Gansner, E.R., Koutsofios, E., North, S.C., Woodhull, G.: Graphviz
and Dynagraph – static and dynamic graph drawing tools. Online Documentation
(accessed November 2007)
34. Fekete, J.-D.: The Infovis Toolkit. In: Ward, M., Munzner, T. (eds.) Proceedings
of the IEEE Symposium on Information Visualization (InfoVis), pp. 167–174.
IEEE Computer Society Press, Los Alamitos (2004)
35. Forlines, C., Shen, C.: DTLens: Multi-user Tabletop Spatial Data Exploration.
In: Proc. of User Interface Software and Technology (UIST), pp. 119–122. ACM
Press, New York (2005), doi:10.1145/1095034.1095055
36. Gahegan, M., Wachowicz, M., Harrower, M., Rhyne, T.-M.: The integration of
geographic visualization with knowledge discovery in databases and geocomputa-
tion. Cartography and Geographic Information Society 28(1), 29–44 (2001)
37. Gershon, N.: What storytelling can do for information visualization. Communi-
cations of the ACM 44(8), 31–37 (2001)
38. Grimstead, I.J., Walker, D.W., Avis, N.J.: Collaborative visualization: A review
and taxonomy. In: Proceedings of the Symposium on Distributed Simulation and
Real-Time Applications, pp. 61–69. IEEE Computer Society Press, Los Alamitos
(2005)
39. Gutwin, C., Greenberg, S.: Design for individuals, design for groups: Tradeoffs
between power and workspace awareness. In: Proceedings of Computer Sup-
ported Cooperative Work (CSCW), pp. 207–216. ACM Press, New York (1998),
doi:10.1145/289444.289495
40. Hancock, M., Carpendale, S.: Supporting multiple off-axis viewpoints at a table-
top display. In: Proceedings of Tabletop, pp. 171–178. IEEE Computer Society
Press, Los Alamitos (2007)
41. Heer, J.: The flare visualization toolkit. Website (accessed November 2007)
42. Heer, J.: Socializing visualization. In: Proc. CHI 2006 Workshop on Social Visu-
alization (2006)
43. Heer, J., Agrawala, M.: Design Considerations for Collaborative Visual Analyt-
ics. In: IEEE Symposium on Visual Analytics Science and Technology (VAST),
pp. 171–178. IEEE Computer Society Press, Los Alamitos (2007), http://vis.
berkeley.edu/papers/design collab vis/2007-DesignCollabVis-VAST.pdf
44. Heer, J., Card, S.K., Landay, J.A.: prefuse: A toolkit for interactive infor-
mation visualization. In: Proceedings of the Conference on Human Factors
in Computing Systems (CHI), pp. 421–430. ACM Press, New York (2005),
doi:10.1145/1054972.1055031
 Creation and Collaboration 129

45. Heer, J., boyd, d.: Vizster: Visualizing Online Social Networks. In: Proceedings
of the IEEE Symposium on Information Visualization (InfoVis ’05), pp. 33–40.
IEEE Computer Society Press, Los Alamitos (2005)
46. Heer, J., Viégas, F.B., Wattenberg, M.: Voyagers and voyeurs: Supporting asyn-
chronous collaborative information visualization. In: Proceedings of the Confer-
ence on Human Factors in Computing Systems (CHI), pp. 1029–1038. ACM Press,
New York (2007)
47. Hill, W.C., Hollan, J.D.: Deixis and the future of visualization excellence. In:
Proc. of IEEE Visualization, pp. 314–319 (1991)
48. Isenberg, T., Neumann, P., Carpendale, S., Nix, S., Greenberg, S.: Interac-
tive annotations on large, high-resolution information displays. In: Confer-
ence Compendium of IEEE VIS, InfoVis, and VAST, pp. 124–125. IEEE
Computer Society Press, Los Alamitos (2006), http://cpsc.ucalgary.ca/
~isenberg/papers/Isenberg 2006 IAL.pdf
49. Jerding, D.F., Stasko, J.T.: The Information Mural: A Technique for Displaying
and Navigating Large Information Spaces. In: Proceedings of the IEEE Sympo-
sium on Information Visualization (InfoVis), pp. 43–50. IEEE Computer Society
Press, Los Alamitos (1995)
50. Johnson, B., Shneiderman, B.: Tree-maps: A space-filling approach to the visual-
ization of hierarchical information structures. In: Proceedings of IEEE Visualiza-
tion, pp. 284–291. IEEE Computer Society Press, Los Alamitos (1991)
51. Kerren, A., Stasko, J.T., Fekete, J.-D., North, C.J. (eds.): Information Visualiza-
tion. LNCS, vol. 4950. Springer, Heidelberg (2008)
52. Kleinmutz, D.N., Schkade, D.A.: Information Displays and Decision Processes.
Psychological Science 4(4), 221–227 (1993)
53. Kruger, R., Carpendale, S., Scott, S.D., Greenberg, S.: Roles of orientation in
tabletop collaboration: Comprehension, coordination and communication. Journal
of Computer Supported Collaborative Work 13(5–6), 501–537 (2004)
54. Kruger, R., Carpendale, S., Scott, S.D., Tang, A.: Fluid integration of rotation
and translation. In: Proceedings of Human Factors in Computing Systems (CHI),
pp. 601–610. ACM Press, New York (2005), doi:10.1145/1054972.1055055
55. Lamping, J., Rao, R., Pirolli, P.: A focus + context technique based on hyperbolic
geometry for visualizing large hierarchies. In: Proceedings of the Conference of
Human Factors in Computing Systems,CHI, pp. 401–408. ACM Press, New York
(1995), doi:10.1145/223904.223956
56. Ling, K., Beenen, G., Lundford, P., Wang, X., Chang, K., Li, X., Cosley, D.,
Frankowski, D., Terveen, L., Rashid, A.M., Resnick, P., Kraut, R.: Using so-
cial psychology to motivate contributions to online communities. Journal of
Computer-Mediated Communication 10(4) (2005)
57. Liu, Y., Gahegan, M., Macgill, J.: Increasing geocomputational interoperability:
Towards a standard geocomputation API. In: Proceedings of GeoComputation,
Ann Arbor, MI (2005)
58. Livny, M., Ramakrishnan, R., Beyer, K., Chen, G., Donjerkovic, D., Lawande,
S., Myllymaki, J., Wenger, K.: DEVise: Integrated querying and visualization of
large datasets. In: Proceedings of the International Conference on Management
of Data (SIGMOD), Tucson, AZ, pp. 301–312. ACM Press, New York (1997)
59. Mackinlay, J.D.: Automating the design of graphical presentations of relational
information. ACM Transactions on Graphics 5(2), 110–141 (1986)
60. Mackinlay, J.D., Hanrahan, P., Stolte, C.: Show me: Automatic presentation
for visual analysis. IEEE Transactions on Visualization and Computer Graph-
ics 13(6), 1137–1144 (2007)
130 J. Heer et al.

61. Mark, G., Kobsa, A.: The effects of collaboration and system transparency on
CIVE usage: An empirical study and model. Presence 14(1), 60–80 (2005)
62. Marr, D.: Vision: A Computational Investigation into the Human Representation
and Processing of Visual Information. W.H. Freeman, New York (1982)
63. Merriam-Webster: Webster’s english dictionary. Website (accessed November
2007), http://www.cs.chalmers.se/~ hallgren/wget.cgi?presentation
64. Munzner, T.: Guest Editor’s Introduction: Information Visualization. Computer
Graphics and Applications 22(1), 20–21 (2002)
65. Munzner, T., Guimbretiére, F., Tasiran, S., Zhang, L., Zhou, Y.: TreeJuxta-
poser: Scalable Tree Comparison Using Focus+Context with Guaranteed Visi-
bility. ACM Transactions on Graphics 22(3), 453–462 (2003)
66. Nacenta, M.A., Sakurai, S., Yamaguchi, T., Miki, Y., Itoh, Y., Kitamura, Y., Sub-
ramanian, S., Gutwin, C.: E-conic: a perspective-aware interface for multi-display
environments. In: Proceedings of ACM Symposium on User Interface Software
and Technology (UIST’01), pp. 279–288. ACM Press, New York (2007)
67. Neumann, P., Schlechtweg, S., Carpendale, M.S.T.: ArcTrees: Visualizing Rela-
tions in Hierarchical Data. In: Proceedings of Eurographics / IEEE VGTC Sym-
posium on Visualization (EuroVis 2005, June 1–3, 2005, Leeds, England, UK),
Aire-la-Ville. Eurographics Workshop Series, pp. 53–60. Eurographics (2005)
68. Neumann, P., Tang, A., Carpendale, S.: A Framework for Visual Information
Analysis. Technical Report 2007-87123, University of Calgary, Calgary, AB,
Canada (July 2007)
69. North, C., Shneiderman, B.: Snap-together visualization: A user interface for co-
ordinating visualizations via relational schemata. In: Proceedings of the Working
Conference on Advanced Visual Interfaces (AVI), May 2000, pp. 128–135. ACM
Press, New York (2000)
70. Olson, G.M., Olson, J.S.: Distance Matters. Human-Computer Interaction 15(2
& 3), 139–178 (2000)
71. Olston, C., Woodruff, A., Aiken, A., Chu, M., Ercegovac, V., Lin, M., Spalding,
M., Stonebraker, M.: Datasplash. In: Proceedings of the International Conference
on Management of Data (SIGMOD), Seattle, WA, June 1998, pp. 550–552. ACM
Press, New York (1998)
72. Park, K.S., Kapoor, A., Leigh, J.: Lessons learned from employing multiple per-
spectives in a collaborative virtual environment for visualizing scientific data.
In: Proceedings of Collaborative Virtual Environments (CVE), pp. 73–82. ACM
Press, New York (2000), doi:10.1145/351006.351015
73. Pinelle, D., Gutwin, C., Greenberg, S.: Task analysis for groupware usability eval-
uation: Modeling shared-workspace tasks with the mechanics of collaboration.
ACM Transaction of Human Computer Interaction 10(4), 281–311 (2003)
74. Rensink, R.A.: chapter Change Blindness. In: McGraw-Hill Yearbook of Science
and Technology, pp. 44–46. McGraw-Hill, New York (2005)
75. Meredith, Ryall, K., Shen, C., Forlines, C., Morris, F.V.R.: Beyond ”social proto-
cols”: Multi-user coordination policies for co-located groupware. In: Proceedings
of Computer-Supported Cooperative Work (CSCW), pp. 262–265. ACM Press,
New York (2004), doi:10.1145/1031607.1031648
76. Rogers, Y., Lindley, S.: Collaborating around vertical and horizontal large inter-
active displays: Which way is best? Interacting with Computers 16(6), 1133–1152
(2004)
77. Meredith, Everitt, K., Ryall, F.V.K., Esenther, A., Forlines, C., Shen, C., Ship-
man, S., Morris, R.: Identity-differentiating widgets for multiuser interactive sur-
faces. IEEE Computer Graphics and Applications 26(5), 56–64 (2006)
 Creation and Collaboration 131

78. Salen, K., Zimmerman, E.: Rules of Play: Fundamentals of Game Design. MIT
Press, Cambridge (2003)
79. Saraiya, P., North, C., Duca, K.: An Insight-Based Methodology for Evaluating
Bioinformatics Visualizations. IEEE Transactions on Visualization and Computer
Graphics 11(4), 443–456 (2005)
80. Scott, S.D., Carpendale, M.S.T., Habelski, S.: Storage bins: Mobile storage for
collaborative tabletop displays. IEEE Computer Graphics and Applications 25(4),
58–65 (2005), http://doi.ieeecomputersociety.org/10.1109/MCG.2005.86
81. Scott, S.D., Carpendale, M.S.T., Inkpen, K.M.: Territoriality in collaborative
tabletop workspaces. In: Proceedings of Computer-Supported Cooperative Work
(CSCW), pp. 294–303. ACM Press, New York (2004), http://innovis.cpsc.
ucalgary.ca/pubs/2004/Territoriality.CSCW/scott cscw2004.pdf,
doi:10.1145/1031607.1031655
82. Scott, S.D., Grant, K.D., Mandryk, R.L.: System guidelines for co-located collab-
orative work on a tabletop display. In: Proceedings of the European Conference
on Computer-Supported Cooperative Work (ECSCW), pp. 159–178. Kluwer Aca-
demic Publishers, Dordrecht (2003),
http://www.ecscw.uni-siegen.de/2003/009Scott_ecscw03.pdf
83. Seo, J., Shneiderman, B.: Knowledge discovery in high dimensional data: Case
studies and a user survey for an information visualization tool. IEEE Transactions
on Visualization and Computer Graphics 12(3), 311–322 (2006)
84. Shen, C., Lesh, N., Vernier, F.: Personal digital historian: Story sharing around
the table. ACM Interactions 10(2), 15–22 (2003)
85. Shneiderman, B.: The eyes have it: A task by data type taxonomy for information
visualizations. In: Proceedings of the IEEE Symposium on Visual Languages, pp.
336–343. IEEE Computer Society Press, Los Alamitos (1996)
86. Shum, S.B., Li, V.U.G., Domingue, J., Motta, E.: Visualizing internetworked ar-
gumentation. In: Kirschner, P.A., Shum, S.J.B., Carr, C.S. (eds.) Visualizing Ar-
gumentation: Software Tools for Collaborative and Educational Sense-Making,
December 2002, pp. 185–204. Springer, Heidelberg (2002)
87. Spence, R.: Information Visualization, 2nd edn. Pearson Education Limited, Har-
low (2007)
88. Takatsuka, M., Gahegan, M.: GeoVISTA Studio: A codeless visual programming
environment for geoscientific data analysis and visualization. Computational Geo-
science 28(10), 1131–1144 (2002)
89. Tandler, P., Prante, T., Müller-Tomfelde, C., Streitz, B., Steinmetz, R.: Con-
necTables: Dynamic Coupling of Displays for the Flexible Creation of Shared
Workspaces. In: Proceedings of User Interface Software and Technology (UIST),
pp. 11–20. ACM Press, New York (2001)
90. Tang, A., Tory, M., Po, B., Neumann, P., Carpendale, S.: Collaborative cou-
pling over tabletop displays. In: Proceedings of Human Factors in Comput-
ing Systems (CHI), pp. 1181–1290. ACM Press, New York (2006), doi:10.1145/
1124772.1124950
91. Thomas, J.J., Cook, K.A.: Illuminating the Path: The Research and Development
Agenda for Visual Analytics. National Visualization and Analytics Center (2005),
http://nvac.pnl.gov/agenda.stm
92. Tufte, E.R.: The Visual Display of Quantitative Information. Graphic Press,
Cheshire (2001)
93. van Wijk, J.J.: The value of visualization. In: Proceedings of IEEE Visualiza-
tion (VIS), pp. 79–86. IEEE Computer Society Press, Los Alamitos (2005),
http://www.win.tue.nl/~ vanwijk/vov.pdf
132 J. Heer et al.

94. Vernier, F., Lesh, N., Shen, C.: Visualization techniques for circular tabletop
interfaces. In: Proceedings of Advanced Visual Interfaces (AVI), pp. 257–263.
ACM Press, New York (2002)
95. Viégas, F.B., boyd, d., Nguyen, D.H., Potter, J., Donath, J.: Digital artifacts for
remembering and storytelling: PostHistory and social network fragments. In: Pro-
ceedings of the Hawaii International Conference on System Sciences (HICCSS),
pp. 105–111 (2004)
96. Viégas, F.B., Wattenberg, M.: Communication-minded visualization: A call to
action. IBM Systems Journal 45(4), 801–812 (2006), doi:10.1147/sj.454.0801
97. Viégas, F.B., Wattenberg, M., van Ham, F., Kriss, J., McKeon, M.: Many Eyes:
A site for visualization at internet scale. IEEE Transactions on Visualization
and Computer Graphics (Proceedings Visualization / Information Visualization
2007) 12(5), 1121–1128 (2007),
http://www.research.ibm.com/visual/papers/viegasinfovis07.pdf
98. von Ahn, L.: Games with a purpose. Computer 39(6), 92–94 (2006)
99. Ware, C.: Information Visualization – Perception for Design, 2nd edn. Morgan
Kaufmann Series in Interactive Technologies. Morgan Kaufmann Publishers, San
Francisco (2004)
100. Wattenberg, M., Kriss, J.: Designing for Social Data Analysis. IEEE Transactions
on Visualization and Computer Graphics 12(4), 549–557 (2006)
101. Weaver, C., Fyfe, D., Robinson, A., Holdsworth, D.W., Peuquet, D.J., Mac-
Eachren, A.M.: Visual analysis of historic hotel visitation patterns. In: Proceed-
ings of the Symposium on Visual Analytics Science and Technology (VAST), Bal-
timore, MD, October 31–November 2 2006, pp. 35–42. IEEE Computer Society
Press, Los Alamitos (2006)
102. Weaver, C.E.: Improvise: A User Interface for Interactive Construction of Highly-
Coordinated Visualizations. Phd thesis, University of Wisconsin–Madison (June
2006)
103. Wesche, G., Wind, J., Göbe, M., Rosenblum, L., Durbin, J., Doyle, R., Tate,
D., King, R., Fröhlich, B., Fischer, M., Agrawala, M., Beers, A., Hanrahan, P.,
Bryson, S.: Application of the Responsive Workbench. IEEE Computer Graphics
and Applications 17(4), 10–15 (1997), doi:10.1109/38.595260
104. Weskamp, M.: newsmap. Website (accessed November 2007),
http://marumushi.com/apps/newsmap/index.cfm
105. Wigdor, D., Shen, C., Forlines, C., Balakrishnan, R.: Perception of elementary
graphical elements in tabletop and multi-surface environments. In: Proceedings
of Human Factors in Computing Systems (CHI), pp. 473–482. ACM Press, New
York (2007)
106. Willett, W., Heer, J., Agrawala, M.: Scented widgets: Improving navigation cues
with embedded visualizations. IEEE Transactions on Visualization and Computer
Graphics 13(6), 1129–1136 (2007)
107. Wright, W., Schroh, D., Proulx, P., Skaburskis, A., Cort, B.: Advances in nSpace
– the sandbox for analysis. In: International Conference on Intelligence Analysis,
McLean, VA (May 2005)
108. Yang, D., Rundensteiner, E.A., Ward, M.O.: Analysis guided visual exploration
to multivariate data. In: Proc. IEEE Visual Analytics Science and Technology
(2007)
109. Yost, B., North, C.: The perceptual scalability of visualization. IEEE Transactions
on Visualization and Computer Graphics 12(5), 837–844 (2005)
110. Zhang, J., Norman, D.A.: Representations in distributed cognitive tasks. Cogni-
tive Science 18(1), 87–122 (1994)
 Creation and Collaboration 133

111. Zuk, T., Schlesier, L., Neumann, P., Hancock, M.S., Carpendale, M.S.T.: Heuris-
tics for Information Visualization Evaluation. In: Proceedings of the Workshop
Beyond Time and Errors (BELIV), held in conjunction with AVI, pp. 55–60. ACM
Press, New York (2006), doi:10.1145/1168149.1168162
 Process and Pitfalls in Writing
Information Visualization Research Papers

Tamara Munzner

Department of Computer Science,University of British Columbia

201-2366 Main Mall, Vancouver BC V6T 1Z4 Canada,
[email protected],
http://www.cs.ubc.ca/~ tmm

Abstract. The goal of this paper is to help authors recognize and avoid
a set of pitfalls that recur in many rejected information visualization
papers, using a chronological model of the research process. Selecting a
target paper type in the initial stage can avert an inappropriate choice
of validation methods. Pitfalls involving the design of a visual encoding
may occur during the middle stages of a project. In a later stage when
the bulk of the research is finished and the paper writeup begins, the
possible pitfalls are strategic choices for the content and structure of the
paper as a whole, tactical problems localized to specific sections, and
unconvincing ways to present the results. Final-stage pitfalls of writing
style can be checked after a full paper draft exists, and the last set of
problems pertain to submission.

1 Introduction
Many rejected information visualization research papers have similar flaws. In
this paper, I categorize these common pitfalls in the context of stages of the
research process. My main goal is to help authors escape these pitfalls, espe-
cially graduate students or those new to the field of information visualization.
Reviewers might also find these pitfalls an interesting point of departure when
considering the merits of a paper.
This paper is structured around a chronological model of the information
visualization research process. I argue that a project should begin with a careful
consideration of the type of paper that is the desired outcome, in order to avoid
the pitfalls of unconvincing validation approaches. Research projects that involve
the design of a new visual encoding would benefit from checking for several
middle-stage pitfalls in unjustified or inappropriate encoding choices. Another
critical checkpoint is the late stage of the project, after the bulk of the work is
done, but before diving in to writing up results. At this point, you should consider
both strategic pitfalls about the high-level structure of the entire paper, tactical
pitfalls that affect one or a few sections, and possible pitfalls in the specifics of
your approach to the results section. At a final stage, when there is a complete
paper draft, you can check for lower-level pitfalls of writing style, and avoid
submission-time pitfalls.

A. Kerren et al. (Eds.): Information Visualization, LNCS 4950, pp. 134–153, 2008.


c Springer-Verlag Berlin Heidelberg 2008
 Writing InfoVis Research Papers 135

I have chosen a breezy style, following in the footsteps of Levin and Re-
dell [22] and Shewchuk [34]. My intent is serious, but I have tried to invent
catchy – sometimes even snide – titles in hopes of making these pitfalls more
memorable. Guides to writing research papers have been written in several sub-
fields of computer science, including systems [22], software engineering [33], pro-
gramming languages [19], networking [28], and graphics [20]. Many of the pitfalls
in the middle and later project stages apply to research writing in general, not
just information visualization, and have been mentioned in one or many of these
previous papers.
My first pass at providing advice for authors and reviewers in the field of
information visualization, abbreviated as infovis, was the creation of the author
guide for the annual conference. When I was Posters Chair of InfoVis 2002, the
IEEE Symposium on Information Visualization, I read the roughly 300 reviews
of the 78 rejected papers in order to decide which to invite as poster submissions.
The experience convinced me that future paper authors would benefit from more
specific guidance. When I became Papers Chair in 2003, with co-chair Stephen
North, we completely rewrote the Call for Papers. We introduced five categories
of papers, with an explicit discussion of the expectations for each, in a guide for
authors that has been kept unchanged through the 2007 conference.
This second pass is motivated by the patterns of mistakes I saw in my two-
year term as InfoVis Papers Co-Chair where I read the over 700 reviews for all 189
submitted papers, and in personally writing nearly 100 reviews in the subsequent
three years. My discussion of paper types below expands considerably on the
previous author guide, and I provide concrete examples of strong papers for each
type. The advice I offer is neither complete nor objective; although I draw on
my experience as a papers chair, my conclusions may be idiosyncratic and reflect
my personal biases. I do not perform any quantitative analysis. Doing so in the
domain of infovis would, no doubt, be fruitful future work, given the interesting
results from software engineering [33] and human-computer interaction [17].
None of these pitfalls are aimed at any particular individual: I have seen
multiple instances of each one. Often a single major pitfall was enough to doom
a paper to rejection, although in some cases I have seen other strengths outweigh
a particular weakness. I hasten to point out that I, myself, have committed some
of the errors listed below, and despite my best efforts I may well fall prey to them
in the future.

2 Initial Stage: Paper Types

A good way to begin a research project is to consider where you want it to end.
That is, what kind of a paper do you intend to produce? That choice should
guide many of your downstream decisions, including the critical issue of how to
validate any claims you might make about your research contribution.

2.1 Validation Approaches

Many possible ways exist to validate the claim that an infovis paper has made
a contribution, including:
136 T. Munzner

– algorithm complexity analysis

– implementation performance (speed, memory)
– quantitative metrics
– qualitative discussion of result pictures
– user anecdotes (insights found)
– user community size (adoption)
– informal usability study
– laboratory user study
– field study with target user population
– design justification from task analysis
– visual encoding justification from theoretical principles

In any particular paper, the constraints of researcher time and page limits force
authors to select a subset of these approaches to validation. The taxonomy of
paper types below can provide you with considerable guidance in choosing ap-
propriate validation approaches, leading to a paper structure where your results
back up your claims. The five paper types guide the presentation of your research
by distinguishing between the following possibilities for your primary contribu-
tion: an algorithm, a design, a system, a user study, or a model.

2.2 Technique
Technique papers focus on novel algorithms and an implementation is expected.
The most straightforward case is where the research contribution is a new algo-
rithm that refines or improves a technique proposed in previous work. A typical
claim is that the new algorithm is faster, more scalable, or provides better visual
quality than the previously proposed one. The MillionVis system [5], hierarchical
parallel coordinates [6], and hierarchical edge bundling [15] are good exemplars
for this category.
Typical results to back up such a claim would be algorithm complexity anal-
ysis, quantitative timing measurements of the implementation, and a qualitative
discussion of images created by the new algorithm. Quantitative metrics of image
quality, for example edge crossings in graph layout, are also appropriate. You
need to compare these results side by side against those from competing algo-
rithms. You might collect this information through some combination of using
results from previous publications, running publicly available code, or imple-
menting them yourself. In this case, there is very little or no design justification
for whether the technique is actually suitable for the proposed problem domain
in the paper itself: there is an implicit assumption that the previous cited work
makes such arguments.
In retrospect, a better name for this category might be Algorithms. Many
authors who design new visual representations might think that a paper docu-
menting a new technique belongs in the Technique category. However, the ques-
tion to ask is whether your primary contribution is the algorithm itself, or the
design. If your algorithm is sophisticated enough that it requires several pages
of description for replicability, then you probably have a primary algorithmic
 Writing InfoVis Research Papers 137

contribution. If the algorithm itself is straightforward enough that only a brief

description is required, or if all of the techniques that you use have been ade-
quately described in previous work, then you would be better served by explicitly
writing a Design Study paper.

2.3 Design Study

Design Study papers make a case that a new visual representation is a suitable
solution for a particular domain problem. First, you should explain the target
problem. You must provide enough background that the reader can pass judge-
ment about whether your solution is good, but not so much detail that the focus
of the paper is on domain problems rather than infovis issues. Finding the right
balance is a difficult but crucial judgement call. Second, you should crisply state
the design requirements that you have determined through your task analysis.
Third, you should present your visual encoding and interaction mechanisms and
justify these design choices in terms of how well they fulfill the requirements.
Typical arguments would refer to perceptual principles and infovis theory. For
example, using spatial position to encode the most important variables and using
greyscale value rather than hue to encode an ordered variable are both very de-
fensible choices [24]. The best justifications explicitly discuss particular choices
in the context of several possible alternatives.
Fourth, you should present results that back up the claim that your approach
is better than others. Typical results include case studies or scenarios of use. De-
sign studies often document iterative design and the use of formative evaluation
for refinement. The research contribution of a design study is not typically a new
algorithm or technique, but rather a well-reasoned justification of how existing
techniques can be usefully combined. For most design studies, adoption by the
target users is valuable evidence that the system has met its goals, as are anec-
dotes of insights found with the new system that would be difficult to obtain
using previous methods.
I think this category name is still a good choice, despite the fact that great
design studies are all too rare. I argue that the field would be well served if more
authors explicitly cast their work in this category. Interesting examples that ap-
proach the design study from several angles are the cluster-calendar system [40],
ThemeRiver [11], Vizster [13], VistaChrom [21], and a hotel visitation pattern
analysis system [42].

2.4 Systems
Systems papers focus on the architectural choices made in the design of an in-
frastructure, framework, or toolkit. A systems paper typically does not introduce
new techniques or algorithms. A systems paper also does not introduce a new
design for an application that solves a specific problem; that would be a de-
sign study. The research contribution of a systems paper is the discussion of
architectural design choices and abstractions in a framework or library, not just
a single application. A good example is the prefuse systems paper [14], which
138 T. Munzner

has a discussion of the performance, flexibility, and extensibility implications of

the ItemRegistry, Action, and ActionList data structure choices. Another good
example is the systems paper on design choices made in Rivet [37] and other
systems with similar goals, such as the tradeoffs of data granularity for trans-
formations.
A systems paper can be considered as a specialized kind of design study:
one about the choices made when building a library as opposed to the choices
made when solving a visual encoding problem. Like the design study category,
key aspects of a systems paper are the lessons learned from building the system,
and observing its use. I urge authors and reviewers of systems papers to peruse
Levin and Redell’s classic on “How (and How Not) to Write a Good Systems
Paper” [22].
The category name might be a cause for confusion because the the term
system is often used interchangeably with application or implementation. The
original intent was to follow the distributed systems usage where there is a very
strong distinction between system-level and application-level work. Although a
name like Toolkit might avert that confusion, the term ’systems paper’ is such
a strong convention in computer science that I am reluctant to advocate this
change.

2.5 Evaluation
Evaluation papers focus on assessing how an infovis system or technique is used
by some target population. Evaluation papers typically do not introduce new
techniques or algorithms, and often use implementations described in previous
work. The most common approach in infovis thus far has been formal user stud-
ies conducted in laboratory setting, using carefully abstracted tasks that can be
quantitatively measured in terms of time and accuracy, and analyzed with sta-
tistical methods. A typical claim would be that the tested tasks are ecologically
valid; that is, they correspond to those actually undertaken by target users in
a target domain. A typical result would be a statistically significant main ef-
fect of an experimental factor, or interaction effect between factors. The work of
Yost and North on perceptual scalability is a good example of this subtype [44].
A different approach to studying user behavior is field studies, where a system
is deployed in a real-world setting with its target users. In these studies, the
number of participants is usually smaller, with no attempt to achieve statis-
tical significance, and the time span is usually weeks or months rather than
hours. However, the study design axes of field versus laboratory, short-term ver-
sus long-term, and size are all orthogonal. Both quantitative and qualitative
measurements may be collected. For example, usage patterns may be studied
through quantitative logging of mouse actions or eyegaze. The work of Hornbæk
and Hertzum on untangling fisheye menus is a good example of this subtype
[16]. Usage patterns can also be studied through qualitative observations during
the test itself or later via coding of videotaped sessions. Trafton et al.’s field
study of how meteorologists use visual representations is an excellent example
of the power of video coding [39].
 Writing InfoVis Research Papers 139

Plaisant’s discussion of the difficulties of evaluating infovis is thoughtful and

germane [29]. In retrospect, a better, albeit wordy, name for this category might
be Summative User Studies, since the goal is to examine the strengths of
a system or technique. Evaluation is far too broad a term – because all papers
should contain some sort of validation. Even User Studies would not be the best
choice, because formative studies are probably a better fit for the Design Study
category, where ethnographic methods are often appropriate in the task analysis
to determine design requirements or iteratively refine a design. However, these
lines are not necessarily crisp. For instance, the MILC approach advocated by
Shneiderman and Plaisant [35] could fit into either a Design Study framework, if
the emphasis is on the formative ethnographic analysis and iterative design, or
a Summative Evaluation framework, if the emphasis is on the longitudinal field
study.

2.6 Model
Model papers present formalisms and abstractions as opposed to the design
or evaluation of any particular technique or system. This category is for meta-
research papers, where the broad purpose is to help other researchers think about
their own work.
The most common subcategory is Taxonomy, where the goal is to propose
categories that help researchers better understand the structure of the space of
possibilities for some topic. Some boundaries will inevitably be fuzzy, but the
goal is to be as comprehensive and complete as possible. As opposed to a survey
paper, where the goal is simply to summarize the previous work, a taxonomy
paper proposes some new categorization or expands upon a previous one and
may presume the reader’s familiarity with the previous work. Good examples
are Card and Mackinlay’s taxonomy of visual encodings [3] and Amar et al.’s
task taxonomy [1].
A second subcategory is Formalism, for papers that present new models,
definitions, or terminology to describe techniques or phenomena. A key attribute
of these kinds of papers is reflective observation. The authors look at what is
going on in a field and provide a new way of thinking about it that is clear,
insightful, and summative. An influential example is the space-scale diagram
work of Furnas and Bederson [7], and an interesting recent example is the casual
infovis definition from Pousman et al. [31].
A third subcategory is Commentary, where the authors advocate a position
and argue to support it. Typical arguments would be “the field needs to do more
X”, “we should be pushing for more Y”, or “avoid doing Z because of these
drawbacks”. A good example is the fisheye followup from Furnas [8]. These kinds
of papers often cite many examples and may also introduce new terminology.
Model papers can provide both a valuable summary of a topic and a vocab-
ulary to more concisely discuss concepts in the area. They can be valuable for
both established researchers and newcomers to a field, and are often used as
assigned readings in courses. I think this category name is appropriate and do
not suggest changing it.
140 T. Munzner

2.7 Combinations

These categories are not hard and fast: some papers are a mixture. For example,
a design study where the primary contribution is the design might include a
secondary contribution of summative evaluation in the form of a lab or field
study. Similarly, a design study may have a secondary contribution in the form of
a novel algorithm. Conversely, a technique paper where the primary contribution
is a novel algorithm may also include a secondary design contribution in the form
of a task analysis or design requirements. However, beware the Neither Fish Nor
Fowl pitfall I discuss below.

2.8 Type Pitfalls

Carefully consider the primary contribution of your work to avoid the pitfalls
that arise from a mismatch between the strengths of your project and the paper
type you choose.

Design in Technique’s Clothing: Don’t validate a new design by providing

only performance measurements. Many rejected infovis papers are bad design
studies, where a new design is proposed but the design requirements are not
crisply stated and justification for design choices is not presented. Many of these
authors would be surprised to hear that they have written a design study, be-
cause they simply assume that the form of a technique paper is always the
correct choice. Technique papers are typically appropriate if you have a novel
algorithmic contribution, or you are the very first to propose a technique. If you
are combining techniques that have been previously proposed, then the design
study form is probably more appropriate.

Application Bingo versus Design Study: Don’t apply some random tech-
nique to a new problem without thoroughly thinking about what the problem
is, whether the technique is suitable, and to what extent it solves the prob-
lem. I define ’application bingo’ as the game where you pick a narrowly defined
problem domain, a random technique, and then write an application paper with
the claim of novelty for this particular domain-technique combination. Applica-
tion bingo is a bad game to play because an overwhelming number of the many
combinatorial possibilities lead to a bad design.
Although application bingo is admittedly a caricature, the important ques-
tion is how we can distinguish those who inadvertently play it from those who
genuinely solve a domain problem with an effective visualization. Some visualiza-
tion venues distinguish between research papers and what are called applications
or case studies. This paper category is often implicitly or explicitly considered
to be a way to gather data from a community outside of visualization itself.
Although that goal is laudable, the mechanism has dangers. A very common
pitfall is that application paper submissions simply describe an instantiation of
a previous technique in great detail. Many do not have an adequate descrip-
tion of the domain problem. Most do not have an adequate justification of why
 Writing InfoVis Research Papers 141

the technique is suitable for the problem. Most do not close the loop with a
validation that the proposed solution is effective for the target users.
In contrast, a strong design study would be rather difficult for an outsider
unfamiliar with the infovis literature to write. Two critical aspects require a
thorough understanding of the strengths and weaknesses of many visualization
techniques. First, although a guideline like “clearly state the problem” might
seem straightforward at first glance, the job of abstracting from a problem in
some target domain to design requirements that can be addressed through visu-
alization techniques requires knowing those techniques. Second, justifying why
the chosen techniques are more appropriate than other techniques again requires
knowledge of the array of possible techniques.
The flip side of this situation is that design studies where visualization re-
searchers do not have close contact with the target users are usually also weak. A
good methodology is collaboration between visualization researchers and target
users with driving problems [18](Chapter 3.4).

All That Coding Means I Deserve a Systems Paper: Many significant

coding efforts do not lead to a systems paper. Consider whether or not you have
specific architectural lessons to offer to the research community that you learned
as a result of building your library or toolkit.

Neither Fish nor Fowl: Papers that try to straddle multiple categories often
fail to succeed in any of them. Be ruthlessly clear about identifying your most
important contribution as primary, and explicitly categorize any other contribu-
tions as secondary. Then make structural and validation choices based on the
category of the single primary contribution.

3 Middle Pitfalls: Visual Encoding

If you have chosen the design route, then a major concern in the middle stages
of a project should be whether your visual encoding choices are appropriate and
justifiable.

Unjustified Visual Encoding: An infovis design study paper must carefully

justify why the visual encoding chosen is appropriate for the problem at hand.
In the case of technique papers, where the focus is on accelerating or improving a
previously proposed technique, the argument can be extremely terse and use a ci-
tation to a previous paper. But in the case of a design study, or a paper proposing
a completely new technique, your justification needs to be explicit and convincing.
One of the most central challenges in information visualization is designing the
visual encoding and interaction mechanisms to show and manipulate a dataset.
A straightforward visual encoding of the exact input data is often not suffi-
cient. In many successful infovis approaches, the input data undergoes significant
transformations into some derived model that is ultimately shown. Many weak
papers completely skip the step of task analysis. Without any discussion of the
142 T. Munzner

design requirements, it is very hard to convince a reader that your model will
solve the problem. In particular, you should consider how to make the case that
the structure you are visually showing actually benefits the target end user. For
example, many authors new to information visualization simply assert, without
justification, that showing the hyperlink structure of the web will benefit end
users who are searching for information. One of my own early papers fell prey
to this very pitfall [26]. However, after a more careful task analysis, I concluded
that most searchers do not need to build a mental model of the structure of the
search space, so showing them that structure adds cognitive load rather than
reduces it. In a later paper [25], I argued that a visual representation of that
hyperlink structure could indeed benefit a specific target community, that of
webmasters and content creators responsible for a particular site.
The foundation of information visualization is the characterization of how
known facts about human perception should guide visual encoding of abstract
datasets. The effectiveness of perceptual channels such as spatial position, color,
size, shape, and so on depends on whether the data to encode is categorical,
ordered, or quantitative [24]. Many individual perceptual channels are preatten-
tively processed in parallel, yet most combinations of these channels must be
serially searched [12]. Some perceptual channels are easily separable, but other
combinations are not [41, Chapter 5]. These principles, and many others, are
a critical part of infovis theory. The last three pitfalls in this section are a few
particularly egregious examples of ignoring this body of knowledge.
Hammer in Search of Nail: If you simply propose a nifty new technique with
no discussion of who might ever need it, it’s difficult to judge its worth. I am
not arguing that all new techniques need to be motivated by specific domain
problems: infovis research that begins from a technique-driven starting place
can be interesting and stimulating. Moreover, it may be necessary to build an
interactive prototype and use it for dataset exploration before it’s possible to
understand the capabilities of a proposed technique.
However, before you write up the paper about that hammer, I urge you to
construct an understanding what kind of nails it can handle. Characterize, at
least with some high-level arguments, the kinds of problems where your new
technique shines as opposed to those where it performs poorly.
2D Good, 3D Better: The use of 3D rather than 2D for the spatial layout of
an abstract dataset requires careful justification that the benefits outweigh the
costs [36]. The use of 3D is easy to justify when a meaningful 3D representation is
implicit in the dataset, as in airflow over an airplane wing in flow visualization or
skeletal structure in medical visualization. The benefit of providing the familiar
view is clear, because it matches the mental model of the user. However, when the
spatial layout is chosen rather than given, as in the abstract datasets addressed
through infovis, there is an explicit choice about which variables to map to
spatial position. It is unacceptable, but all too common with naive approaches
to infovis, to simply assert that using an extra dimension must be a good idea.
The most serious problem with a 3D layout is occlusion. The ability to in-
teractively change the point of view with navigational controls does not solve
 Writing InfoVis Research Papers 143

the problem. Because of the limitations of human memory, comparing something

visible with memories of what was seen before is more difficult than comparing
things simultaneously visible side by side [30]. A great deal of work has been de-
voted to the exploration of the power of multiple linked 2D views, often directly
showing derived variables [43] or using them for ordering [2,23]. In many cases,
these views are more effective than simply jumping to 3D [40]. Other difficulties
of 3D layouts are that users have difficulty in making precise length judgements
because of perspective foreshortening [38], and also that text is difficult to read
unless it is billboarded to a 2D plane [9].

Color Cacophony: An infovis paper loses credibility when you make design de-
cisions with blatant disregard for basic color perception facts. Examples include
having huge areas of highly saturated color, hoping that color coding will be
distinguishable in tiny regions, using more nominal categories than the roughly
one dozen that can be distinguishable with color coding, or using a sequential
scheme for diverging data. Using a red/green hue coding is justifiable only when
strong domain conventions exist, and should usually be redundantly coded with
luminance differences to be distinguishable to the 10% of men who are color-
blind. You should not attempt to visually encode three variables through the
three channels of red, green, and blue; they are not separable because they are
integrated by the visual system into a combined percept of color. These principles
have been clearly explained by many authors, including Ware [41, Chapter 4].

Rainbows Just like in the Sky: The unjustified use of a continuous rainbow
colormap is a color pitfall so common that I give it a separate title. The most
critical problem is that the standard rainbow colormap is perceptually nonlinear.
A fixed range of values that are indistinguishable in the green region would
clearly show change in other regions such as where orange changes to yellow
or cyan changes to blue. Moreover, hue does not have an implicit perceptual
ordering, in contrast to other visual attributes such as greyscale or saturation. If
the important aspect of the information to be encoded is low-frequency change,
then use a colormap that changes from just one hue to another, or has a single hue
that changes saturation. If you are showing high-frequency information, where
it is important to distinguish and discuss several nameable regions, then a good
strategy is to explicitly quantize your data into a segmented rainbow colormap.
These ideas are discussed articulately by Rogowitz and Treinish [32].

4 Late Pitfalls: Paper Strategy, Tactics, and Results

The time to consider the late pitfalls is after the bulk of the project work has
been done, but before starting to write your paper draft.

4.1 Strategy Pitfalls

Strategy pitfalls pertain to the content and structure of the entire paper, as
opposed to more localized tactics problems that only affect one or a few sections.
144 T. Munzner

What I Did over My Summer Vacation: Do not simply enumerate all

activities that required effort when writing a paper. Instead, make judgements
about what to discuss in a paper based on your research contributions. This
category evokes the grade-school essays many of us were asked to write each fall,
which were typically narrative and chronological. Often these Summer Vacation
papers contain too much low-level detail, in the extreme cases reading more
like a manual than a research paper. For instance, a feature that took weeks
or even months of implementation effort may only merit a few sentences, and
some features may not be sufficiently interesting to the research community to
mention at all. These papers are usually the result of authors who do not know
the literature well enough to understand what is and is not a contribution. The
solution is to plunge into a more extensive literature review.

Least Publishable Unit: Do not try to squeeze too many papers out of the
same project, where you parcel out some tiny increment of research contribution
beyond your own previous work. The determination of what is a paper-sized unit
of work is admittedly a very individual judgement call, and I will not attempt
to define the scope here. As a reviewer, I apply the “I know it when I see it”
standard.

Dense as Plutonium: Do not try to cram many papers’ worth of content

and contributions into one — the inverse pitfall to the one above. These papers
are difficult to read because the barrage of ideas is so dense. More importantly,
because you don’t have enough room to explain the full story of how you have
accomplished your work, these papers fail the reproducability test. If you are
leaving out so many critical details that the work cannot be replicated by a
sufficiently advanced graduate student, then split your writeup into multiple
papers.

Bad Slice and Dice: If you have done two papers’ worth of work and choose
to write two papers, you can still make the wrong choice about how to split up
the work between them. In this pitfall, neither paper is truly standalone, yet
both repeat too much content found in the other. Repartitioning can solve this
problem.

4.2 Tactical Pitfalls

The tactical pitfalls are localized to one or a few sections, as opposed to the
paper-level strategy problems above.

Stealth Contributions: Do not leave your contributions implicit or unsaid,

whether from intellectual sloth, misplaced modesty, or the hope that the reader
may invent a better answer than what you can provide. It is a central piece of
your job as an author to clearly and explicitly tell the reader the contributions
of your work.
 Writing InfoVis Research Papers 145

I highly recommend having a sentence near the end of the introduction that
starts, “The contribution of this work is”, and of using bulleted lists if there are
multiple contributions. More subtle ways of stating contributions, using verbs
like ’present’ and ’propose’, can make it more difficult for readers and reviewers
to ferret out which of your many sentences is that all-important contributions
statement. Also, do not assume that the reader can glean your overall contri-
butions from a close reading of the arguments in your previous work section.
While it is critical to have a clear previous work section that states how you ad-
dress the limitations of the previous work, as I discuss below, your paper should
clearly communicate your contributions even if the reader has skipped the entire
previous work section.
I find that articulating the contributions requires very careful consideration
and is one of the hardest parts of writing up a paper. They are often quite
different than the original goals of the project, and often can only be determined
in retrospect. What can we do that wasn’t possible before? How can we do
something better than before? What do we know that was unknown or unclear
before? The answers to these questions should guide all aspects of the paper,
from the high-level message to the choice of which details are worth discussing.
And yet, as an author I find that it’s hard to pin these down at the beginning of
the writing process. This reason is one of the many to start writing early enough
that there is time to refine through multiple drafts. After writing a complete
draft, then reading through it critically, I can better refine the contributions
spin in a next pass.

I Am So Unique: Do not ignore previous work when writing up your paper.

You have to convince the reader that you have done something new, and the
only way to do that is to explain how it is different than what has already been
done. All research takes place in some kind of intellectual context, and your job
as author is to situate what you have done within a framework of that context.
A good previous work section is a mini-taxonomy of its own, where you decide
on meaningful categorization given your specific topic.
Proposing new names for old techniques or ideas may sneak your work past
some reviewers, but will infuriate those who know of that previous work. This
tactic will also make you lose credibility with knowledgeable readers. If you
cannot find anything related to what you have done, it’s more likely that you’re
looking in the wrong subfield than that your work is a breakthrough of such
magnitude that there is no context. Remember to discuss not only work that
has been done on similar problems to your own, but also work that uses similar
solutions to yours that occurs in different problem domains. This advice is even
more critical if you were lax about doing a literature review before you started
your project. If you find work similar to your own, you have a fighting chance
of carefully differentiating yours in the writeup, but if a reviewer is the one to
bring it to your attention, the paper is most likely dead.

Enumeration without Justification: Simply citing the previous work is nec-

essary but not sufficient. A description that “X did Y”, even if it includes detail,
146 T. Munzner

is not enough. You must explain why this previous work does not itself solve your
problem, and what specific limitations of that previous work your approach does
address. Every paper you cite in the previous work section is a fundamental chal-
lenge to the very existence of your project. Your job is to convince a skeptical
reader that the world needs your new thing because it is somehow better than a
particular old thing. Moreover, it’s not even enough to just make the case that
yours is different – yours must be better. The claims you make must, of course,
be backed up by your validation in a subsequent results section.
A good way to approach the previous work section is that you want to tell
to a story to the reader. Figure out the messages you want to get across to the
reader, in what order, and then use the references to help you tell this story. It
is possible to group the previous work into categories, and to usefully discuss
the limitations of the entire category.

Sweeping Assertions: A research paper should not contain sweeping

unattributed assertions. You have three choices: cite your source; delete the
assertion from your paper; or explicitly tag the statement as your observation,
your conjecture, or an explanation of your results. In the last case, the assertion
is clearly marked as being part of your research contribution. Be careful with folk
wisdom that “everybody knows”. You could be mistaken, and tracking down the
original sources may change or refine your views. If you cannot find a suitable
source after extensive digging, you have stumbled upon a great topic for a future
paper! You may either validate and extend the conventional wisdom, or show
that it is incorrect.

I Am Utterly Perfect: No work is perfect. An explicit discussion of the limita-

tions of your work strengthens, rather than weakens, your paper. Papers without
a discussion of limitations, weaknesses, and implications feel unfinished or pre-
liminary. For instance, how large of a dataset can your system handle? Can you
categorize the kinds of datasets for which your technique is suitable and those
for which it is not?

4.3 Results Pitfalls

Several pitfalls on how to validate your claims can occur in the results section
of your paper.

Unfettered by Time: Do not omit time performance from your writeup, be-
cause it is almost always interesting and worth documenting. The level of detail
at which you should report this result depends on the paper type and the contri-
bution claims. For instance, a very high-level statement like “interactive response
for all datasets shown on a desktop PC” may suffice for an evaluation paper or
a design study paper. However, for a technique paper with a contribution claim
of better performance than previous techniques, detailed comparison timings in
tables or charts would be a better choice.
 Writing InfoVis Research Papers 147

Fear and Loathing of Complexity: Although most infovis papers do not

have detailed proofs of complexity, technique papers that focus on accelerating
performance should usually include some statement of algorithm complexity.

Straw Man Comparison: When comparing your technique to previous work,

compare against state-of-the-art approaches rather than outdated work. For ex-
ample, authors unaware of recent work in multilevel approaches to force-directed
graph drawing [10] sometimes compare against very naive implementations of
spring systems. At the lower level, if you compare benchmarks of your implemen-
tation to performance figures quoted from a previous publication and your hard-
ware configuration is more powerful, you should explicitly discuss the difference
in capabilities. Better yet, rerun the benchmarks for the competing algorithms
on the same machine you use to test your own.

Tiny Toy Datasets: Avoid using only tiny toy datasets in technique papers
that refine previously proposed visual encodings. While small synthetic bench-
marks can be useful for expository purposes, your validation should include
datasets of the same size used by state-of-the-art approaches. Similarly, you
should use datasets characteristic of those for your target application.
On the other hand, relatively small datasets may well be appropriate for a
user study, if they are carefully chosen in conjunction with some specific target
task and this choice is explained and justified.

But My Friends Liked It: Positive informal evaluation of a new infovis sys-
tem by a few of your infovis-expert labmates is not very compelling evidence
that a new technique is useful for novices or scientists in other domains. While
the guerilla/discount methodology is great for finding usability problems with
products [27], a stronger approach would be informal evaluation with more rep-
resentative subjects, or formal evaluation with rigorous methodology.

Unjustified Tasks: Beware of running a user study where the tasks are not
justified. A study is not very interesting if it shows a nice result for a task
that nobody will ever actually do, or a task much less common or important
than some other task. You need to convince the reader that your tasks are a
reasonable abstraction of the real-world tasks done by your target users. If you
are the designer of one of the systems studied, be particularly careful to make a
convincing case that you did not cherry-pick tasks with a bias to the strengths
of your own system.

5 Final Pitfalls: Style and Submission

After you have a full paper draft, you should check for the final-stage pitfalls.

5.1 Writing Style Pitfalls

Several lower-level pitfalls pertain to writing style.
148 T. Munzner

Deadly Detail Dump: When writing a paper, do not simply dump out all
the details and declare victory. The details are the how of what you did and do
belong at the heart of your paper. But you must first say what you did and why
you did it before the how. This advice holds at multiple levels. For the high-level
paper structure, start with motivation: why should I, the reader, care about
what you’ve done? Then provide an overview: a big-picture view of what you
did. The algorithmic details can then appear after the stage has been set. At the
section, subsection, and sometimes even paragraph level, stating the what before
the how will make your writing more clear.

Story-Free Captions: Avoid using a single brusque sentence fragment as your

caption text. Caption words are not a precious resource that you should hoard
and spend begrudgingly. Instead, design your paper so that as much of the paper
story as possible is understandable to somebody who flips through looking only
at the figures and captions. Many readers of visualization and graphics papers
do exactly this when skimming, so make your captions as standalone as possible.

My Picture Speaks for Itself: You should talk the reader through how your
visual representation exposes meaningful structure in the dataset, rather than
simply assuming the superiority of your method is obvious to all readers from
unassisted inspection of your result images. Technique and design study papers
usually include images in the results section showing the visual encodings cre-
ated by a technique or system on example datasets. The best way to carry out
this qualitative evaluation is to compare your method side-by-side with repre-
sentations created by competing methods on the same dataset.

Grammar Is Optional: Grammar is not optional; you should use correct

syntax and punctuation for a smooth low-level flow of words. If English is not
your first language, consider having a native speaker check the writing before
submitting a paper for review, and also before the final version of your paper
goes to press. I recommend Dupré’s book [4] as an excellent pitfall-oriented
technical writing guide for computer scientists.

Mistakes Were Made: Avoid the passive voice as much as possible. I call
out this particular grammar issue because it directly pertains to making your
research contribution clear. Is the thing under discussion part of your research
contribution, or something that was done or suggested by others? The problem
with the passive voice is its ambiguity: the reader does not have enough infor-
mation to determine who did something. This very ambiguity can be the lure
of the passive voice to a slothful or overly modest writer. I urge you to use the
active voice and make such distinctions explicitly.

Jargon Attack: Avoid jargon as much as possible, and if you must use it then
define it first. Definitions are critical both for unfamiliar terms or acronyms, as
well as for standard English words being used in a specific technical sense.
 Writing InfoVis Research Papers 149

Nonspecific Use of Large: Never just use the words ’large’ or ’huge’ to de-
scribe a dataset or the scalability of a technique without giving numbers to clarify
the order of magnitude under discussion: hundreds, tens of thousands, millions?
Every author has a different idea of what these words mean, ranging from 128
to billions, so be specific. Also, you should provide the size of all datasets used
in results figures, so that readers don’t have to count dots in an image to guess
the numbers.

5.2 Submission Pitfalls

Finally, I caution against pitfalls at the very end of the project, when submitting
your paper.

Slimy Simultaneous Submission: Simultaneous submission of the same work

at multiple venues who clearly request original work is highly unethical. More-
over, simultaneous submission is stupid, because it is often detected when the
same reviewer is independently selected by different conference chairs. The num-
ber of experts in any particular subfield can be quite a small set. The standard
penalty upon detection is instant dual rejection, and multi-conference blacklists
are beginning to be compiled. Finally, even if you do succeed in getting the same
work published twice, any gains you make by having a higher publication count
will be offset when you lose credibility within your field from those who actually
read the work and are annoyed to wade through multiple papers that say the
same thing.

Resubmit Unchanged: If your paper is rejected, don’t completely ignore the

reviews and resubmit to another venue without making any changes. As above,
there’s a reasonable chance that you’ll get the one of the same reviewers again.
That reviewer will be highly irritated.

6 Pitfalls By Generality

A cross-cutting way to categorize these pitfalls is by generality. Many hold true

for any scientific research paper, rather than being specific to visualization. Of
the latter, many hold true for both scientific visualization (scivis) and informa-
tion visualization (infovis). As many have lamented, the names of these subfields
are unfortunate and confusing for outsiders. The definition I use is that it’s in-
fovis when the spatial representation is chosen, and it’s scivis when the spatial
representation is given. Operationally, InfoVis split off as a sister conference from
IEEE Visualization (Vis) in 1995. At Vis, the focus is now on scivis.
The choice of paper types is specific to the InfoVis author guide, because
this categorization is not explicitly discussed in the Vis call for papers. The
first-stage type pitfalls are thus quite specific to infovis. The middle pitfalls on
visual encoding are specific to visualization. Color Cacophony and Rainbows
Just Like In The Sky certainly pertain to both infovis and scivis. Unjustified
150 T. Munzner

Visual Encoding, Hammer In Search Of Nail, and 2D Good, 3D Better focus

on issues that are more central for an infovis audience, but may well be of
benefit to scivis as well. All of the strategy pitfalls pertain to any research paper.
The result pitfalls hold for all visualization papers, and Straw Man Comparison
is general enough for all research papers. The tactical and final stage pitfalls
are very general, with two exceptions. Story-Free Captions is specific to both
visualization and computer graphics. My Picture Speaks For Itself is again most
tuned for infovis, but certainly may pique the interest of the scivis community.
Although I have framed my discussion in terms of the InfoVis author guide
paper categories, my comments also apply to infovis papers in other venues. I
argue that even if a call for papers does not explicitly state paper categories,
keeping this paper taxonomy in mind will help you write a stronger paper.

7 Conclusion
I have advocated an approach to conducting infovis research that begins with an
explicit consideration of paper types. I have exhorted authors to avoid pitfalls
at several stages of research process, including visual encoding during design,
a checkpoint before starting to write, and after a full paper draft exists. My
description and categorization of these pitfalls reflects my own experiences as
author, reviewer, and papers chair. I offer it in hopes of steering and stimulating
discussion in our field.

Acknowledgments. This paper has grown in part from a series of talks. The
impetus to begin articulating my thoughts was the Publishing Your Visualiza-
tion Research panel at the IEEE Visualization 2006 Doctoral Colloquium. I
benefited from discussion with many participants at the 2007 Dagstuhl Seminar
on Information Visualization and the 2007 Visual Interactive Effective Worlds
workshop at the Lorentz Center, after speaking on the same topic. I also appre-
ciate a discussion with Pat Hanrahan and Lyn Bartram on design studies. John
Stasko’s thoughts considerably strengthened my discussion of model papers. I
thank Aaron Barsky, Hamish Carr, Jeff Heer, Stephen Ingram, Ciarán Llachlan
Leavitt, Peter McLachlan, James Slack, and Matt Ward for feedback on paper
drafts. I particularly thank Heidi Lam, Torsten Möller, John Stasko, and Melanie
Tory for extensive discussions beyond the call of duty.

References
1. Amar, R., Eagan, J., Stasko, J.: Low-level components of analytic activity in in-
formation visualization. In: Proc. IEEE Symposium on Information Visualization
(InfoVis), pp. 111–117 (2005)
2. Becker, R.A., Cleveland, W.S., Shyu, M.J.: The visual design and control of trellis
display. Journal of Computational and Statistical Graphics 5, 123–155 (1996)
3. Card, S., Mackinlay, J.: The structure of the information visualization design
space. In: Proc. IEEE Symposium on Information Visualization (InfoVis), pp.
92–99 (1997)
 Writing InfoVis Research Papers 151

4. Dupré, L.: Bugs in Writing. Addison-Wesley, Reading (1995)

5. Fekete, J.-D., Plaisant, C.: Interactive information visualization of a million items.
In: Proc. IEEE Symposium on Information Visualization (InfoVis), pp. 117–126
(2002)
6. Fua, Y.-H., Ward, M.O., Rundensteiner, E.A.: Hierarchical parallel coordinates
for visualizing large multivariate data sets. In: Proc. IEEE Visualization Conf.
(Vis), pp. 43–50 (1999)
7. Furnas, G., Bederson, B.: Space-scale diagrams: Understanding multiscale inter-
faces. In: Proc. ACM Conf. Human Factors in Computing Systems (CHI), pp.
234–241 (1995)
8. Furnas, G.W.: A fisheye follow-up: Further reflection on focus + context. In: Proc.
ACM Conf. Human Factors in Computing Systems (CHI), pp. 999–1008 (2006)
9. Grossman, T., Wigdor, D., Balakrishnan, R.: Exploring and reducing the effects
of orientation on text readability in volumetric displays. In: Proc. ACM Conf.
Human Factors in Computing Systems (CHI), pp. 483–492 (2007)
10. Hachul, S., Jünger, M.: Drawing large graphs with a potential-field-based mul-
tilevel algorithm. In: Pach, J. (ed.) GD 2004. LNCS, vol. 3383, pp. 285–295.
Springer, Heidelberg (2005)
11. Havre, S., Hetzler, B., Nowell, L.: Themeriver(tm): In search of trends, patterns,
and relationships. In: Proc. IEEE Symposium on Information Visualization (In-
foVis), pp. 115–123 (2000)
12. Healey, C.G.: Perception in visualization, cited 14 Nov. 2007,
http://www.csc.ncsu.edu/faculty/healey/PP
13. Heer, J., boyd, d.: Vizster: Visualizing online social networks. In: Proc. IEEE
Symposium on Information Visualization (InfoVis), pp. 32–39 (2005)
14. Heer, J., Card, S.K., Landay, J.A.: prefuse: a toolkit for interactive information
visualization. In: Proc. ACM Conf. Human Factors in Computing Systems (CHI),
pp. 421–430 (2005)
15. Holten, D.: Hierarchical edge bundles: Visualization of adjacency relations in hier-
archical data. IEEE Trans. Visualization and Computer Graphics (TVCG) (Proc.
InfoVis 06) 12(5), 741–748 (2006)
16. Hornbæk, K., Hertzum, M.: Untangling the usability of fisheye menus. ACM
Trans. on Computer-Human Interaction (ToCHI) 14(2), article 6 (2007)
17. Isaacs, E., Tang, J.: Why don’t more non-North-American papers get accepted
to CHI? SIGCHI Bulletin 28(1) (1996),
http://www.sigchi.org/bulletin/1996.1/isaacs.html
18. Johnson, C., Moorhead, R., Munzner, T., Pfister, H., Rheingans, P., Yoo, T.S.:
NIH/NSF Visualization Research Challenges Report. IEEE Computer Society
Press, Los Alamitos (2006)
19. Johnson, R.E., et al.: How to get a paper accepted at OOPSLA. In: Proc. Conf.
Object Oriented Programming Systems Languages and Applications (OOPSLA),
pp. 429–436 (1996),
http://www.sigplan.org/oopsla/oopsla96/how93.html
20. Kajiya, J.: How to get your SIGGRAPH paper rejected,
http://www.siggraph.org/publications/instructions/rejected
21. Kincaid, R., Ben-Dor, A., Yakhini, Z.: Exploratory visualization of array-based
comparative genomic hybridization. Information Visualization 4(3), 176–190
(2005)
152 T. Munzner

22. Levin, R., Redell, D.D.: An evaluation of the ninth SOSP submissions; or, how
(and how not) to write a good systems paper. ACM SIGOPS Operating Systems
Review 17(3), 35–40 (1983),
http://www.usenix.org/events/samples/submit/advice.html
23. MacEachren, A., Dai, X., Hardisty, F., Guo, D., Lengerich, G.: Exploring high-D
spaces with multiform matrices and small multiples. In: Proc. IEEE Symposium
on Information Visualization (InfoVis), pp. 31–38 (2003)
24. Mackinlay, J.D.: Automating the Design of Graphical Presentations of Relational
Information. ACM Trans. on Graphics (TOG) 5(2), 111–141 (1986)
25. Munzner, T.: Drawing large graphs with H3Viewer and Site Manager. In: White-
sides, S.H. (ed.) GD 1998. LNCS, vol. 1547, pp. 384–393. Springer, Heidelberg
(1999)
26. Munzner, T., Burchard, P.: Visualizing the structure of the world wide web in
3D hyperbolic space. In: Proc. Virtual Reality Modelling Language Symposium
(VRML), pp. 33–38. ACM SIGGRAPH (1995)
27. Nielsen, J.: Guerrilla HCI: Using discount usability engineering to penetrate the
intimidation barrier. In: Bias, R.G., Mayhew, D.J. (eds.) Cost-justifying usability,
pp. 245–272. Academic Press, London (1994)
28. Partridge, C.: How to increase the chances your paper is accepted at ACM
SIGCOMM. ACM SIGCOMM Computer Communication Review 28(3) (1998),
http://www.sigcomm.org/conference-misc/author-guide.html
29. Plaisant, C.: The challenge of information visualization evaluation. In: Proc. Ad-
vanced Visual Interfaces (AVI), pp. 109–116 (2004)
30. Plumlee, M., Ware, C.: Zooming versus multiple window interfaces: Cognitive
costs of visual comparisons. Proc. ACM Trans. on Computer-Human Interaction
(ToCHI) 13(2), 179–209 (2006)
31. Pousman, Z., Stasko, J.T., Mateas, M.: Casual information visualization: Depic-
tions of data in everyday life. IEEE Trans. Visualization and Computer Graphics
(TVCG) (Proc. InfoVis 07) 13(6), 1145–1152 (2007)
32. Rogowitz, B.E., Treinish, L.A.: How not to lie with visualization. Computers In
Physics 10(3), 268–273 (1996),
http://www.research.ibm.com/dx/proceedings/pravda/truevis.htm
33. Shaw, M.: Mini-tutorial: Writing good software engineering research papers.
In: Proc. Intl. Conf. on Software Engineering (ICSE), pp. 726–736 (2003),
http://www.cs.cmu.edu/~ Compose/shaw-icse03.pdf
34. Shewchuk, J.: Three sins of authors in computer science and math (1997),
http://www.cs.cmu.edu/jrs/sins.html
35. Shneiderman, B., Plaisant, C.: Strategies for evaluating information visualization
tools: Multi-dimensional in-depth long-term case studies. In: Proc. AVI Workshop
on BEyond time and errors: novel evaLuation methods for Information Visualiza-
tion (BELIV), pp. 38–43 (2006)
36. Smallman, H.S., John, M.S., Oonk, H.M., Cowen, M.B.: Information availability in
2D and 3D displays. IEEE Computer Graphics and Applications (CG&A) 21(5),
51–57 (2001)
37. Tang, D., Stolte, C., Bosch, R.: Design choices when architecting visualizations.
Information Visualization 3(2), 65–79 (2004)
38. Tory, M., Kirkpatrick, A.E., Atkins, M.S., Möller, T.: Visualization task perfor-
mance with 2D, 3D, and combination displays. IEEE Trans. Visualization and
Computer Graphics (TVCG) 12(1), 2–13 (2006)
 Writing InfoVis Research Papers 153

39. Trafton, J.G., Kirschenbaum, S.S., Tsui, T.L., Miyamoto, R.T., Ballas, J.A., Ray-
mond, P.D.: Turning pictures into numbers: Extracting and generating informa-
tion from complex visualizations. Intl. Journ. Human Computer Studies 53(5),
827–850 (2000)
40. van Wijk, J.J., van Selow, E.R.: Cluster and calendar based visualization of time
series data. In: Proc. IEEE Symposium on Information Visualization (InfoVis),
pp. 4–9 (1999)
41. Ware, C.: Information Visualization: Perception for Design, 2nd edn. Morgan
Kaufmann/Academic Press, London (2004)
42. Weaver, C., Fyfe, D., Robinson, A., Holdsworth, D.W., Peuquet, D.J.,
MacEachren, A.M.: Visual analysis of historic hotel visitation patterns. Infor-
mation Visualization 6(1), 89–103 (2007)
43. Wilkinson, L., Anand, A., Grossman, R.: Graph-theoretic scagnostics. In: Proc.
IEEE Symposium on Information Visualization (InfoVis), pp. 157–164 (2005)
44. Yost, B., North, C.: The perceptual scalability of visualization. IEEE Trans. Vi-
sualization and Computer Graphics (TVCG) (Proc. InfoVis 06) 12(5), 837–844
(2006)
 Visual Analytics:
Definition, Process, and Challenges

Daniel Keim1 , Gennady Andrienko2 , Jean-Daniel Fekete3 , Carsten Görg4 ,

Jörn Kohlhammer5 , and Guy Melançon6
1
Department of Computer and Information Science, University of Konstanz,
78457 Konstanz, Germany,
[email protected]
2
Fraunhofer Institute for Intelligent Analysis and Information Systems(IAIS),
Schloss Birlinghoven 53754 Sankt Augustin, Germany,
[email protected]
3
Université Paris-Sud, INRIA, Bât 490,
F-91405 Orsay Cedex, France,
[email protected]
4
School of Interactive Computing & GVU Center, Georgia Institute of Technology,
85 5th St., NW, Atlanta, GA 30332-0760, USA,
[email protected]
5
Fraunhofer Institute for Computer Graphics Research,
Fraunhoferstraße 5, D-64283 Darmstadt, Germany,
[email protected]
6
INRIA Bordeaux – Sud-Ouest, CNRS UMR 5800 LaBRI,
Campus Université Bordeaux I,
351 Cours de la libération, 33405 Talence Cedex, France,
[email protected]

1 Introduction and Motivation

We are living in a world which faces a rapidly increasing amount of data to be

dealt with on a daily basis. In the last decade, the steady improvement of data
storage devices and means to create and collect data along the way influenced our
way of dealing with information: Most of the time, data is stored without filtering
and refinement for later use. Virtually every branch of industry or business,
and any political or personal activity nowadays generate vast amounts of data.
Making matters worse, the possibilities to collect and store data increase at a
faster rate than our ability to use it for making decisions. However, in most
applications, raw data has no value in itself; instead we want to extract the
information contained in it.
The information overload problem refers to the danger of getting lost in
data which may be

– irrelevant to the current task at hand

– processed in an inappropriate way
– presented in an inappropriate way

A. Kerren et al. (Eds.): Information Visualization, LNCS 4950, pp. 154–175, 2008.


c Springer-Verlag Berlin Heidelberg 2008
 Visual Analytics: Definition, Process, and Challenges 155

Due to information overload, time and money are wasted, scientific and in-
dustrial opportunities are lost because we still lack the ability to deal with the
enormous data volumes properly. People in both their business and private lives,
decision-makers, analysts, engineers, emergency response teams alike, are often
confronted with massive amounts of disparate, conflicting and dynamic infor-
mation, which are available from multiple heterogeneous sources. We want to
simply and effectively exploit and use the hidden opportunities and knowledge
resting in unexplored data sources.
In many application areas success depends on the right information being
available at the right time. Nowadays, the acquisition of raw data is no longer
the driving problem: It is the ability to identify methods and models, which can
turn the data into reliable and provable knowledge. Any technology, that claims
to overcome the information overload problem, has to provide answers for the
following problems:
– Who or what defines the “relevance of information” for a given task?
– How can appropriate procedures in a complex decision making process be
identified?
– How can the resulting information be presented in a decision- or task-oriented
way?
– What kinds of interaction can facilitate problem solving and decision mak-
ing?
With every new “real-life” application, procedures are put to the test possibly
under circumstances completely different from the ones under which they have
been established. The awareness of the problem how to understand and analyse
our data has been greatly increased in the last decade. Even as we implement
more powerful tools for automated data analysis, we still face the problem of un-
derstanding and “analysing our analyses” in the future: Fully-automated search,
filter and analysis only work reliably for well-defined and well-understood prob-
lems. The path from data to decision is typically quite complex. Even as fully-
automated data processing methods represent the knowledge of their creators,
they lack the ability to communicate their knowledge. This ability is crucial: If
decisions that emerge from the results of these methods turn out to be wrong,
it is especially important to examine the procedures.
The overarching driving vision of visual analytics is to turn the information
overload into an opportunity: Just as information visualization has changed our
view on databases, the goal of Visual Analytics is to make our way of processing
data and information transparent for an analytic discourse. The visualization of
these processes will provide the means of communicating about them, instead
of being left with the results. Visual Analytics will foster the constructive eval-
uation, correction and rapid improvement of our processes and models and -
ultimately - the improvement of our knowledge and our decisions (see Figure 1).
On a grand scale, visual analytics solutions provide technology that combines
the strengths of human and electronic data processing. Visualization becomes
the medium of a semi-automated analytical process, where humans and machines
cooperate using their respective distinct capabilities for the most effective results.
156 D. Keim et al.

Fig. 1. Tight integration of visual and automatic data analysis methods with database
technology for a scalable interactive decision support.

The user has to be the ultimate authority in giving the direction of the analysis
along his or her specific task. At the same time, the system has to provide
effective means of interaction to concentrate on this specific task. On top of
that, in many applications different people work along the path from data to
decision. A visual representation will sketch this path and provide a reference
for their collaboration across different tasks and abstraction levels.
The diversity of these tasks can not be tackled with a single theory. Visual
analytics research is highly interdisciplinary and combines various related re-
search areas such as visualization, data mining, data management, data fusion,
statistics and cognition science (among others). Visualization has to continuously
challenge the perception by many of the applying sciences that visualization is
not a scientific discipline in its own right. Even if the awareness exists, that
scientific analysis and results must be visualized in one way or the other, this
often results in ad hoc solutions by application scientists, which rarely match
the state of the art in interactive visualization science, much less the full com-
plexity of the problems. In fact, all related research areas in the context of visual
analytics research conduct rigorous, serious science each in a vibrant research
community. To increase the awareness of their work and their implications for
visual analytics research clearly emerges as one main goal of the international
visual analytics community (see Figure 2).
Because visual analytics research can be regarded as an integrating discipline,
application specific research areas should contribute with their existing proce-
dures and models. Emerging from highly application-oriented research, dispersed
research communities worked on specific solutions using the repertoire and stan-
dards of their specific fields. The requirements of visual analytics introduce new
dependencies between these fields.
 Visual Analytics: Definition, Process, and Challenges 157

Fig. 2. Visual analytics integrates scientific disciplines to improve the division of labor
between human and machine.

2 Definition of Visual Analytics

In “Illuminating the Path” [39], Thomas and Cook define visual analytics as the
science of analytical reasoning facilitated by interactive visual interfaces. In this
paper, however, we would like to give a more specific definition:
Visual analytics combines automated analysis techniques with interactive
visualizations for an effective understanding, reasoning and decision making on
the basis of very large and complex data sets.
The goal of visual analytics is the creation of tools and techniques to enable
people to:
– Synthesize information and derive insight from massive, dynamic, ambigu-
ous, and often conflicting data.
– Detect the expected and discover the unexpected.
– Provide timely, defensible, and understandable assessments.
– Communicate assessment effectively for action.
By integrating selected science and technology from the above discussed disci-
plines and as illustrated in Figure 2, there is the promising opportunity to form
the unique and productive field of visual analytics. Work in each of the partici-
pating areas focuses on different theoretical and practical aspects of users solving
real-world problems using Information Technology in an effective and efficient
way. These areas have in common similar scientific challenges and significant sci-
entific added-value from establishing close collaboration can be identified. Benefit
of collaboration between the fields is identified to be two-fold:
– Jointly tackling common problems will arrive at better results on the local
level of each discipline, in a more efficient way.
– Integrating appropriate results from each of the disciplines will lay the fun-
dament for significantly improved solutions in many important data analysis
applications.
158 D. Keim et al.

Visual Analytics versus Information Visualization

Many people are confused by the new term visual analytics and do not see a dif-
ference between the two areas. While there is certainly some overlay and some of
the information visualization work is certainly highly related to visual analytics,
traditional visualization work does not necessarily deal with an analysis tasks
nor does it always also use advanced data analysis algorithms.
Visual analytics is more than just visualization. It can rather be seen as an
integral approach to decision-making, combining visualization, human factors
and data analysis. The challenge is to identify the best automated algorithm for
the analysis task at hand, identify its limits which can not be further automated,
and then develop a tightly integrated solution with adequately integrates the best
automated analysis algorithms with appropriate visualization and interaction
techniques.
While some of such research has been done within the visualization commu-
nity in the past, the degree to which advanced knowledge discovery algorithms
have been employed is quite limited. The idea of visual analytics is to funda-
mentally change that. This will help to focus on the right part of the problem,
i.e. the parts that can not be solved automatically, and will provide solutions to
problems that we were not able to solve before.
One important remark should be made here. Most research efforts in Infor-
mation Visualization have concentrated on the process of producing views and
creating valuable interaction techniques for a given class of data (social network,
multi-dimensional data, etc.). However, much less has been suggested as to how
user interactions on the data can be turned into intelligence to tune underlying
analytical processes. A system might for instance observe that most of the user’s
attention concern only a subpart of an ontology (through queries or by repeated
direct manipulations of the same graphical elements, for instance). Why not then
use this knowledge about the user’s interest and update various parameters by
the system (trying to systematically place elements or components of interest in
center view, even taking this fact into account when driving a clustering algo-
rithm with a modularity quality criteria, for instance).
This is one place where Visual Analytics maybe differs most from Information
Visualization, giving higher priority to data analytics from the start and through
all iterations of the sense making loop. Creativity is then needed to understand
how perception issues can help bring more intelligence into the analytical process
by “learning” from users’ behavior and effective use of the visualization.

3 Areas Related to Visual Analytics

Visual analytics builds on a variety of related scientific fields. At its heart, Visual
Analytics integrates Information and Scientific Visualization with Data Manage-
ment and Data Analysis Technology, as well as Human Perception and Cognition
research. For effective research, Visual Analytics also requires an appropriate In-
frastructure in terms of software and data sets and related analytical problems
repositories, and to develop reliable Evaluation methodology (see Figure 3).
 Visual Analytics: Definition, Process, and Challenges 159

Fig. 3. Visual Analytics integrates Scientific and Information Visualization with core
adjacent disciplines: Data management and analysis, spatio-temporal data, and human
perception and cognition. Successful Visual Analytics research also depends on the
availability of appropriate infrastructure and evaluation facilities.

An example for a common problem in several of the disciplines is that of scalabil-

ity with data size. The larger the data set to be handled gets, the more difficult
it gets to manage, analyze, and visualize these data effectively. Researching ap-
propriate forms to represent large data volumes by smaller volumes containing
the most relevant information benefits each of the data management, analy-
sis, and visualization fields. On top of these individual progresses, a synergetic
collaboration of all these fields may lead to significantly improved processing
results. Consider a very large data stream. Appropriate data management tech-
nology gives efficient access to the stream, which is intelligently processed and
abstracted by an automatic analysis algorithm which has an interface to the
data management layer. On top of the analysis output, an interactive visual-
ization which is optimized for efficient human perception of the relevant infor-
mation allows the analyst to consume the analysis results, and adapt relevant
parameters of the data aggregation an analysis engines as appropriate. The com-
bination of the individual data handling steps into a Visual Analytics pipeline
leads to improved results and makes data domains accessible which are not ef-
fectively accessible by any of the individual data handling disciplines. Similar
argumentations apply to other related fields and disciplines. In many fields, vi-
sualization is already used and developed independently as a means for analyzing
the problems at hand. However, a unified, interdisciplinary perspective on using
visualization for analytical problem-solving will show beneficial for all involved
disciplines. As common principles, best practices, and theories will be developed,
these will become usable in the individual disciplines and application domains,
providing economies of scale, avoiding replication of work or application of only
sub-optimal techniques.
160 D. Keim et al.

3.1 Visualization

Visualization has emerged as a new research discipline during the last two dec-
ades. It can be broadly classified into Scientific and Information Visualization.
In Scientific Visualization, the data entities to be visualized are typically 3D
geometries or can be understood as scalar, vectorial, or tensorial fields with ex-
plicit references to time and space. A survey of current visualization techniques
can be found in [22,35,23]. Often, 3D scalar fields are visualized by isosurfaces or
semi-transparent point clouds (direct volume rendering) [15]. To this end, meth-
ods based on optical emission- or absorption models are used which visualize the
volume by ray-tracing or projection. Also, in the recent years significant work
focused on the visualization of complex 3-dimensional flow data relevant e.g.,
in aerospace engineering [40]. While current research has focused mainly on effi-
ciency of the visualization techniques to enable interactive exploration, more and
more methods to automatically derive relevant visualization parameters come
into focus of research. Also, interaction techniques such as focus&context [28]
gain importance in scientific visualization.
Information Visualization during the last decade has developed methods
for the visualization of abstract data where no explicit spatial references are
given [38,8,24,41]. Typical examples include business data, demographics data,
network graphs and scientific data from e.g., molecular biology. The data con-
sidered often comprises hundreds of dimensions and does not have a natural
mapping to display space, and renders standard visualization techniques such as
(x, y) plots, line- and bar-charts ineffective. Therefore, novel visualization tech-
niques are being developed by employing e.g., Parallel Coordinates and their
numerous extensions [20], Treemaps [36], and Glyph [17]- and Pixel-based [25]
visual data representations. Data with inherent network structure may be visual-
ized using graph-based approaches. In many Visualization application areas, the
typically huge volumes of data require the appropriate usage of automatic data
analysis techniques such as clustering or classification as preprocessing prior to
visualization. Research in this direction is just emerging.

3.2 Data Management

An efficient management of data of various types and qualities is a key com-

ponent of Visual Analytics as this technology typically provides the input of
the data which are to be analyzed. Generally, a necessary precondition to per-
form any kind of data analysis is an integrated and consistent data basis [18,19].
Database research has until the last decade focused mainly on aspects of effi-
ciency and scalability of exact queries on homogeneous, structured data. With
the advent of the Internet and the easy access it provides to all kinds of hetero-
geneous data sources, the database research focus has shifted toward integration
of heterogeneous data. Finding integrated representation of different data types
such as numeric data, graphs, text, audio and video signals, semi-structured
data, semantic representations and so on is a key problem of modern database
 Visual Analytics: Definition, Process, and Challenges 161

technology. But the availability of heterogeneous data not only requires the map-
ping of database schemata but includes also the cleaning and harmonization of
uncertainty and missing data in the volumes of heterogeneous data. Modern ap-
plications require such intelligent data fusion to be feasible in near real-time and
as automatically as possible [32]. New forms of information sources such as data
streams [11], sensor networks [30] or automatic extraction of information from
large document collections (e.g., text, HTML) result in a difficult data analysis
problem which to support is currently in the focus of database research [43].
The relationship between Data Management, Data Analysis and Visualization
is characterized such that Data Management techniques developed increasingly
rely on intelligent data analysis techniques, and also interaction and visualiza-
tion to arrive at optimal results. On the other hand, modern database systems
provide the input data sources which are to be visually analyzed.

3.3 Data Analysis

Data Analysis (also known as Data Mining or Knowledge Discovery) researches
methods to automatically extract valuable information from raw data by means
of automatic analysis algorithms [29,16,31]. Approaches developed in this area
can be best described by the addressed analysis tasks. A prominent such task
is supervised learning from examples: Based on a set of training samples, deter-
ministic or probabilistic algorithms are used to learn models for the classification
(or prediction) of previously unseen data samples [13]. A huge number of algo-
rithms have been developed to this end such as Decision Trees, Support Vector
Machines, Neuronal Networks, and so on. A second prominent analysis task is
that of cluster analysis [18,19], which aims to extract structure from data with-
out prior knowledge being available. Solutions in this class are employed to au-
tomatically group data instances into classes based on mutual similarity, and to
identify outliers in noisy data during data preprocessing for subsequent analysis
steps. Further data analysis tasks include tasks such as association rule mining
(analysis of co-occurrence of data items) and dimensionality reduction. While
data analysis initially was developed for structured data, recent research aims at
analyzing also semi-structured and complex data types such as web documents
or multimedia data [34].
It has recently been recognized that visualization and interaction are highly
beneficial in arriving at optimal analysis results [9]. In almost all data analysis
algorithms a variety of parameters needs to be specified, a problem which is
usually not trivial and often needs supervision by a human expert. Visualization
is also a suitable means for appropriately communicating the results of the au-
tomatic analysis, which often is given in abstract representation, e.g., a decision
tree. Visual Data Mining methods [24] try to achieve exactly this.

3.4 Perception and Cognition

Effective utilization of the powerful human perception system for visual analysis
tasks requires the careful design of appropriate human-computer interfaces. Psy-
chology, Sociology, Neurosciences and Design each contribute valuable results to
162 D. Keim et al.

the implementation of effective visual information systems. Research in this area

focuses on user-centered analysis and modeling (Requirement Engineering), the
development of principles, methods and tools for design of perception-driven,
multimodal interaction techniques for visualization and exploration of large in-
formation spaces, as well as usability evaluation of such systems [21,12]. On the
technical side, research in this area is influenced by two main factors: (1.) The
availability of improved display resources (hardware), and (2.) Development of
novel interaction algorithms incorporating machine recognition of the actual user
intent and appropriate adaptation of main display parameters such as the level
of detail, data selection and aggregation, etc. by which the data is presented[44].
Important problems addressed in this area include the research of perceptual,
cognitive and graphical principles which in combination lead to improved visual
communication of data and analysis results; The development of perception-
theory-based solutions for the graphical representation of static and dynamic
structures; And development of visual representation of information at several
levels of abstraction, and optimization of existing focus-and-context techniques.

3.5 Human-Computer Interaction

Human-computer interaction is the research area that studies the interaction
between people and computers. It involves the design, implementation and eval-
uation of interactive systems in the context of the user’s task and work [12].
Like visual analytics itself, human-computer interaction is a multi-disciplinary
research area that draws on many other disciplines: computer science, system
design, and behavioral science are some of them. The basic underlying research
goal is to improve the interaction between users and computers: how to make
computers more receptive to the users’ intentions and needs. Thus, the research
areas discussed in the previous section about perception and cognition are also
much related to human-computer interaction [21].
As pointed out in the introduction, visual analytics aims to combine and
integrate the strengths of computers and humans into an interactive process to
extract knowledge from data. To effectively switch back and forth between tasks
for the computer and tasks for the human it is crucial to develop an effective
user interface that minimizes the barrier between the human’s cognitive model
of what they want to accomplish and the computer’s understanding of the hu-
man’s task. The design of user interfaces focuses on human factors of interactive
software, methods to develop and assess interfaces, interaction styles, and design
considerations such as effective messages and appropriate color choice [37].

3.6 Infrastructure and Evaluation

The above described research disciplines require cross-discipline support regard-
ing the evaluation of the found solutions, and need certain infrastructure and
standardization grounding to build on effectively. In the field of information vi-
sualization, standardization and evaluation came into the focus of research only
recently. It has been realized that a general understanding of the taxonomies
 Visual Analytics: Definition, Process, and Challenges 163

regarding the main data types and user tasks [2] to be supported are highly de-
sirable for shaping visual analytics research. A common understanding of data
and problem dimensions and structure, and acceptance of evaluation standards
will make research results better comparable, optimizing research productivity.
Also, there is an obvious need to build repositories of available analysis and vi-
sualization algorithms, which researchers can build upon in their work, without
having to re-implement already proven solutions.
How to assess the value of visualization is a topic of lively debate [42,33]. A
common ground that can be used to position and compare future developments
in the field of data analysis is needed. The current diversification and dispersion
of visual analytics research and development resulted from its focus onto specific
application areas. While this approach may suit the requirements of each of
these applications, a more rigorous and overall scientific perspective will lead to
a better understanding of the field and a more effective and efficient development
of innovative methods and techniques.

3.7 Sub-communities

Spatio-Temporal Data: While many different data types exist, one of the
most prominent and ubiquitous data types is data with references to time and
space. The importance of this data type has been recognized by a research
community which formed around spatio-temporal data management and anal-
ysis [14]. In geospatial data research, data with references in the real world
coming from e.g., geographic measurements, GPS position data, remote sensing
applications, and so on is considered. Finding spatial relationships and patterns
among this data is of special interest, requiring the development of appropriate
management, representation and analysis functions. E.g., developing efficient
data structures or defining distance and similarity functions is in the focus of re-
search. Visualization often plays a key role in the successful analysis of geospatial
data [6,26].
In temporal data, the data elements can be regarded as a function of time.
Important analysis tasks here include the identification of patterns (either lin-
ear or periodical), trends and correlations of the data elements over time, and
application-dependent analysis functions and similarity metrics have been pro-
posed in fields such as finance, science, engineering, etc. Again, visualization of
time-related data is important to arrive at good analysis results [1].
The analysis of data with references both in space and in time is a chal-
lenging research topic. Major research challenges include [4]: scale, as it is often
necessary to consider spatio-temporal data at different spatio-temporal scales;
the uncertainty of the data as data are often incomplete, interpolated, collected
at different times, or based upon different assumptions; complexity of geograph-
ical space and time, since in addition to metric properties of space and time
and topological/temporal relations between objects, it is necessary to take into
account the heterogeneity of the space and structure of time; and complexity of
spatial decision making processes, because a decision process may involve hetero-
164 D. Keim et al.

geneous actors with different roles, interests, levels of knowledge of the problem
domain and the territory.

Network and Graph Data: Graphs appear as flexible and powerful math-
ematical tools to model real-life situations. They naturally map to transporta-
tion networks, electric power grids, and they are also used as artifacts to study
complex data such as observed interactions between people, or induced interac-
tions between various biological entities. Graphs are successful at turning seman-
tic proximity into topological connectivity, making it possible to address issues
based on algorithmics and combinatorial analysis.
Graphs appear as essential modeling and analytical objects, and as effective
visual analytics paradigms. Major research challenges are to produce scalable
analytical methods to identify key components both structurally and visually.
Efforts are needed to design process capable of dealing with large datasets while
producing readable and usable graphical representations, allowing proper user
interaction. Special efforts are required to deal with dynamically changing net-
works, in order to assess of structural changes at various scales.

4 The Visual Analytics Process

A number of systems for information visualization, as well as specific visual-

ization techniques, motivate their design choice from Shneiderman’s celebrated
mantra “Overview first, Filter and zoom, Details on demand”. As is, the mantra
clearly emphasizes the role of visualization in the knowledge discovery process.
Recently, Keim adjusted the mantra to bring its focus toward Visual Analytics:
“Analyze first, Show the Important, Zoom, filter and analyze further, Details
on demand”. In other words, this mantra is calling for astute combinations of
analytical approaches together with advanced visualization techniques.
The computation of any visual representation and/or geometrical embedding
of large and complex datasets requires some analysis to start with. Many scalable
graph drawing algorithms try to take advantage of any knowledge on topology
to optimize the drawing in terms of readability. Other approaches offer repre-
sentations composed of visual abstractions of clusters to improve readability.
The challenge then is to try to come up with a representation that is as faithful
as possible to avoid introducing uncertainty. We must not fall into the naı̈ve
assumption that visualization can offer a virgin view on the data: any represen-
tation will inevitably favor an interpretation over all possible ones. The solution
offered by Visual Analytics is then to let the user enter into a loop where data
can be interactively manipulated to help gain insight both on the data and the
representation itself.
The sense-making loop structures the whole knowledge discovery process
supported through Visual Analytics. A generic scenario can be given following a
schema developed by van Wijk [42], which furthermore admits to be evaluated
and measured in terms of efficiency or knowledge gained. A choice for an initial
representation and adequate interactions can be made after applying different
 Visual Analytics: Definition, Process, and Challenges 165

statistical and mathematical techniques, such as spatio-temporal data analysis or

link mining depending on the nature of the dataset under study. The process then
enters a loop where the user can gain knowledge on the data, ideally driving the
system toward more focused and more adequate analytical techniques. Dually,
interacting on the visual representation, the user will gain a better understanding
of the visualization itself commanding for different views helping him or her to
go beyond the visual and ultimately confirm hypotheses built from previous
iterations (see Figure 4).

Fig. 4. The sense-making loop for Visual Analytics based on the simple model of
visualization by Wijk [42].

5 Application Challenges
Visual Analytics is a highly application oriented discipline driven by practical
requirements in important domains. Without attempting a complete survey over
all possible application areas, we sketch the potential applicability of Visual
Analytics technology in a few key domains.
In the Engineering domain, Visual Analytics can contribute to speed-up de-
velopment time for products, materials, tools and production methods by offering
more effective, intelligent access to the wealth of complex information resulting
from prototype development, experimental test series, customers’ feedback, and
many other performance metrics. One key goal of applied Visual Analytics in
the engineering domain will be the analysis of the complexity of the production
systems in correlation with the achieved output, for an efficient and effective
improvement of the production environments.
Financial Analysis is a prototypical promising application area for Visual
Analytics. Analysts in this domain are confronted with streams of heterogeneous
information from different sources available at high update rates, and of varying
166 D. Keim et al.

reliability. Arriving at a unifying, task-centered view on diverse streams of data

is a central goal in financial information systems. Integrated analysis and visu-
alization of heterogeneous data types such as news feeds, real-time trading data,
and fundamental economic indicators poses a challenge for developing advanced
analysis solutions in this area. Research based on results from Information Vi-
sualization is regarded as promising in this case.
Socio-economic considerations often form the basis of political decision
processes. A modern society can be regarded as a complex system of interre-
lationships between political decisions and economic, cultural and demographic
effects. Analysis and Visualization of these interrelationships is promising in de-
veloping a better understanding of these phenomena, and to arrive at better
decisions. Successful Visual Analytics applications in this domain could start
being developed based on currently existing Geo-Spatial analysis frameworks.
Public Safety & Security is another important application area where Vi-
sual Analytics may contribute with advanced solutions. Analysts need to con-
stantly monitor huge amounts of heterogeneous information streams, correlating
information of varying degrees of abstraction and reliability, assessing the cur-
rent level of public safety, triggering alert in case of alarming situations being
detected. Data integration and correlation combined with appropriate analysis
and interactive visualization is promising to develop more efficient tools for the
analysis in this area.
The study of Environment and Climate change often requires the ex-
amination of long term weather records and logs of various sensors, in a search
for patterns that can be related to observations such as changes in animal pop-
ulations, or in meteorological and climatic processes for instance. These require-
ments call for the development of systems allowing visual and graphical access
to historical monitoring data and predictions from various models in search for
or in order to validate patterns building over time.
These diverse fields of applications share many problems on an abstract level,
most of which are addressed by Visual Analytics. The actual (software) solution
must be adapted to the specific needs and terminologies of the application area
and consequently, many researchers currently focus on a specific customer seg-
ment. Much can be achieved, if the European research infrastructure in this field
becomes strong enough to encourage the exchange of ideas on a broad scale, to
foster development of solutions applicable to multiple domains, achieving syn-
ergy effects.

6 Technical Challenges

The primary goal of Visual Analytics is the analysis of vast amounts of data to
identify and visually distill the most valuable and relevant information content.
The visual representation should reveal structural patterns and relevant data
properties for easy perception by the analyst. A number of key requirements
need to be addressed by advanced Visual Analytics solutions. We next outline
important scientific challenges in this context.
 Visual Analytics: Definition, Process, and Challenges 167

Scalability with Data Volumes and Data Dimensionality: Visual Ana-

lytics techniques need to be able to scale with the size and dimensionality of
the input data space. Techniques need to accommodate and graphically repre-
sent high-resolution input data as well as continuous input data streams of high
bandwidth. In many applications, data from multiple, heterogeneous sources
need to be integrated and processed jointly. In these cases, the methods need
to be able to scale with a range of different data types, data sources, and levels
of quality. The visual representation algorithms need to be efficient enough for
implementation in interactive systems.

Quality of Data and Graphical Representation: A central issue in Visual

Analytics is the avoidance of misinterpretations by the analyst. This may result
due to uncertainty and errors in the input data, or limitations of the chosen
analysis algorithm, and may produce misleading analysis results. To face this
problem, the notion of data quality, and the confidence of the analysis algorithm
needs to be appropriately represented in the Visual Analytics solutions. The user
needs to be aware of these data and analysis quality properties at any stage in
the data analysis process.

Visual Representation and Level of Detail: To accommodate vast streams

of data, appropriate solutions need to intelligently combine visualizations of
selected analysis details on the one hand, and a global overview on the other
hand. The relevant data patterns and relationships need to be visualized on
several levels of detail, and with appropriate levels of data and visual abstraction.

User Interfaces, and Interaction Styles and Metaphors: Visual Analytics

systems need to be easily used and interacted with by the analyst. The analyst
needs to be able to fully focus on the task at hand, not on overly technical or
complex user interfaces, which potentially distract. To this end, novel interaction
techniques need to be developed which fully support the seamless, intuitive visual
communication with the system. User feedback should be taken as intelligently
as possible, requiring as little manual user input as possible, which guarantees
the full support of the user in navigating and analyzing the data, memorizing
insights and making informed decisions.

Display Devices: In addition to high-resolution desktop displays, advanced

display devices such as large-scale power walls and small portable personal assis-
tant, graphically-enabled devices need to be supported. Visual Analytics systems
should adapt to the characteristics of the available output devices, supporting
the Visual Analytics workflow on all levels of operation.

Evaluation: Due to the complex and heterogeneous problem domains addressed

by Visual Analytics, so far it has been difficult to perform encompassing evalua-
tion work. A theoretically founded evaluation framework needs to be developed
which allows assessing the contribution of any Visual Analytics system toward
the level of effectiveness and efficiency achieved regarding their requirements.
168 D. Keim et al.

Infrastructure: Managing large amounts of data for visualization or analysis

requires special data structures and mechanisms, both in memory and disks.
Achieving interactivity means refreshing the display in 100ms at worst whereas
analyzing data with standard techniques such as clustering can take hours to
complete. Achieving the smooth interaction required by the analysts to perform
their tasks while providing high-quality analytical algorithms need the combi-
nation of asynchronous computation with hybrid analytical algorithms that can
trade time with quality. Moreover, to fully support the analytical process, the
history of the analysis should also be recorded and interactively edited and an-
notated. Altogether, these requirements call for a novel software infrastructure,
built upon well understood technologies such as databases, software components
and visualization but augmented with asynchronous processing, history manage-
ments and annotations.

7 Examples for Visual Analytics Applications

7.1 Visual Analytics Tools for Analysis of Movement Data
With widespread availability of low cost GPS devices, it is becoming possible to
record data about the movement of people and objects at a large scale. While
these data hide important knowledge for the optimization of location and mobil-
ity oriented infrastructures and services, by themselves they lack the necessary
semantic embedding which would make fully automatic algorithmic analysis pos-
sible. At the same time, making the semantic link is easy for humans who however
cannot deal well with massive amounts of data. In [5] we argue that by using
the right visual analytics tools for the analysis of massive collections of move-
ment data, it is possible to effectively support human analysts in understanding
movement behaviors and mobility patterns.
Figure 5 shows a subset of raw GPS measurements presented in so-called
space-time cube. The large amount of position records referring to the same
territory over a long time period makes it virtually impossible to do the analysis
by purely visual methods.
The paper [5] proposes a framework where interactive visual interfaces are
synergistically combined with database operations and computational process-
ing. The generic database techniques are used for basic data processing and ex-
traction of relevant objects and features. The computational techniques, which
are specially devised for movement data, aggregate and summarize these objects
and features and thereby enable the visualization of large amounts of informa-
tion. The visualization enables human cognition and reasoning, which, in turn,
direct and control the further analysis by means of the database, computational,
and visual techniques. Interactive visual interfaces embrace all the tools.
Thus, in order to detect and interpret significant places visited by the mov-
ing entities, the positions of stops are extracted from the data by means of
appropriate database queries. Then, clustering methods are applied to detect
frequently visited places. Interactive visual displays put the results in the spa-
tial and temporal contexts. The spatial positions of the stops can be observed on
 Visual Analytics: Definition, Process, and Challenges 169

Fig. 5. A visual display of a large amount of position records is unreadable and not
suitable for analysis.

Fig. 6. Positions of stops have been extracted from the database. By means of cluster-
ing, frequently visited places have been detected.
170 D. Keim et al.

Fig. 7. The temporal histograms show the distribution of the stops in the frequently
visited places (Figure 6) with respect to the weekly (left) and daily (right) cycles.

a map (Figure 6) or 3D spatial view. Temporal histograms (Figure 7) are used

to explore the temporal distribution of the stops throughout the time period and
within various temporal cycles (daily, weekly, etc.). These complementary views
allow a human analyst to understand the meanings or roles of the frequently
visited places.
In order to detect and interpret typical routes of the movement between the
significant places, the analyst first applies a database query to extract sequences
of position records between the stops, from which trajectories (time-referenced
lines) are constructed. Then, clustering is applied with the use of specially de-
vised similarity measures. The results are computationally generalized and sum-
marized and displayed in the spatial context (Figure 8).

7.2 Multilevel Visualization of the Worldwide Air

Transportation Network
The air transportation network has now become more dense and more complex
at all geographical levels. Its dynamic no more rests on simple territorial logics.
The challenge is to gain insightful understandings on how the routes carrying the
densest traffic organize themselves and impact the organization of the network
into sub-communities at lower levels. At the same time, subnetworks grow on
their own logic, involving tourism, economy or territorial control, and influence
or fight against each other. Because of the network size and complexity, its study
can no more rely on traditional world map and requires novel visualization. A
careful analysis of the network structural properties, requiring recent results on
small world phenomenon, reveals its multilevel community structure.
The original network is organized into a top level network of communi-
ties (Figure 9(a)). Each component can then be further decomposed into sub-
communities. Capitals such as New York, Chicago, Paris or London (Figure 9(b))
clearly attract most of the international traffic and impose routes to fly the world
around because of airline partnerships (economical logic). Asia (Figure 9(c))
clearly stands apart from these core hubs because of strong territorial ties en-
dorsed by national Asian airline companies (territorial logic). Visualization of
social networks such as the worldwide air transportation is challenged by the
necessity to scale with the growing size of network data while being able to offer
 Visual Analytics: Definition, Process, and Challenges 171

Fig. 8. A result of clustering and summarization of movement data: the routes between
the significant places.

readable visual representations and fluid interaction. Visualization today brings

the field of social sciences close to the study of complex systems and promises
to deliver new knowledge across these disciplines [7,3,10].

8 Conclusions
The problems addressed by Visual Analytics are generic. Virtually all sciences
and many industries rely on the ability to identify methods and models, which
can turn data into reliable and provable knowledge. Ever since the dawn of mod-
ern science, researchers needed to find methodologies to create new hypotheses,
to compare them with alternative hypotheses, and to validate their results. In
a collaborative environment this process includes a large number of specialized
people each having a different educational background. The ability to commu-
nicate results to peers will become crucial for scientific discourse.
Currently, no technological approach can claim to give answers to all three
key questions that have been outlined in the first section, regarding the
– relevance of a specific information
– adequacy of data processing methods and validity of results
– acceptability of the presentation of results for a given task
172 D. Keim et al.

(a) World air transportation network.

(b) USA and world hubs. (c) Asia.

Fig. 9. Multilevel Visualization of the Worldwide Air Transportation Network

 Visual Analytics: Definition, Process, and Challenges 173

Visual Analytics research does not focus on specific methods to address these
questions in a single “best-practice”. Each specific domain contributes a reper-
toire of approaches to initiate an interdisciplinary creation of solutions.
Visual Analytics literally maps the connection between different alternative
solutions, leaving the opportunity for the human user to view these options in
the context of the complete knowledge generation process and to discuss these
options with peers on common ground.

References
1. Aigner, W., Miksch, S., Müller, W., Schumann, H., Tominski, C.: Visual meth-
ods for analyzing time-oriented data. IEEE Transactions on Visualization and
Computer Graphics 14(1), 47–60 (2008)
2. Amar, R.A., Eagan, J., Stasko, J.T.: Low-level components of analytic activity in
information visualization. In: INFOVIS, p. 15 (2005)
3. Amiel, M., Melançon, G., Rozenblat, C.: Réseaux multi-niveaux: l’exemple des
échanges aériens mondiaux. M@ppemonde 79(3) (2005)
4. Andrienko, G., Andrienko, N., Jankowski, P., Keim, D., Kraak, M.-J.,
MacEachren, A., Wrobel, S.: Geovisual analytics for spatial decision support:
Setting the research agenda. Special issue of the International Journal of Geo-
graphical Information Science 21(8), 839–857 (2007)
5. Andrienko, G., Andrienko, N., Wrobel, S.: Visual analytics tools for analysis of
movement data. ACM SIGKDD Explorations 9(2) (2007)
6. Andrienko, N., Andrienko, G.: Exploratory Analysis of Spatial and Temporal
Data. Springer, Heidelberg (2005)
7. Auber, D., Chiricota, Y., Jourdan, F., Melançon, G.: Multiscale visualization of
small world networks. In: INFOVIS (2003)
8. Card, S.K., Mackinlay, J., Shneiderman, B.: Readings in Information Visualiza-
tion: Using Vision to Think. Morgan Kaufmann, San Francisco (1999)
9. Ceglar, A., Roddick, J.F., Calder, P.: Guiding knowledge discovery through in-
teractive data mining, pp. 45–87. IGI Publishing, Hershey (2003)
10. Chiricota, Y., Melançon, G.: Visually mining relational data. International Review
on Computers and Software (2005)
11. Das, A.: Semantic approximation of data stream joins. IEEE Transactions on
Knowledge and Data Engineering 17(1), 44–59 (2005), Member-Johannes Gehrke
and Member-Mirek Riedewald
12. Dix, A., Finlay, J.E., Abowd, G.D., Beale, R.: Human-Computer Interaction (.),
3rd edn. Prentice-Hall, Inc., Upper Saddle River (2003)
13. Duda, R., Hart, P., Stock, D.: Pattern Classification. John Wiley and Sons Inc.,
Chichester (2000)
14. Dykes, J., MacEachren, A., Kraak, M.-J.: Exploring geovisualization. Elsevier
Science, Amsterdam (2005)
15. Engel, K., Hadwiger, M., Kniss, J.M., Rezk-salama, C., Weiskopf, D.: Real-time
Volume Graphics. A. K. Peters, Ltd., Natick (2006)
16. Ester, M., Sander, J.: Knowledge Discovery in Databases - Techniken und An-
wendungen. Springer, Heidelberg (2000)
17. Forsell, C., Seipel, S., Lind, M.: Simple 3d glyphs for spatial multivariate data.
In: INFOVIS, p. 16 (2005)
174 D. Keim et al.

18. Han, J., Kamber, M. (eds.): Data Mining: Concepts and Techniques. Morgan
Kaufmann, San Francisco (2000)
19. Hand, D., Mannila, H., Smyth, P. (eds.): Principles of Data Mining. MIT Press,
Cambridge (2001)
20. Inselberg, A., Dimsdale, B.: Parallel Coordinates: A Tool for Visualizing Multi-
variate Relations (chapter 9), pp. 199–233. Plenum Publishing Corporation, New
York (1991)
21. Jacko, J.A., Sears, A.: The Handbook for Human Computer Interaction. Lawrence
Erlbaum & Associates, Mahwah (2003)
22. Johnson, C., Hanson, C. (eds.): Visualization Handbook. Kolam Publishing (2004)
23. Keim, D., Ertl, T.: Scientific visualization (in german). Information Technol-
ogy 46(3), 148–153 (2004)
24. Keim, D., Ward, M.: Visual Data Mining Techniques (chapter 11). Springer, New
York (2003)
25. Keim, D.A., Ankerst, M., Kriegel, H.-P.: Recursive pattern: A technique for visu-
alizing very large amounts of data. In: VIS ’95: Proceedings of the 6th conference
on Visualization ’95, Washington, DC, USA, p. 279. IEEE Computer Society
Press, Los Alamitos (1995)
26. Keim, D.A., Panse, C., Sips, M., North, S.C.: Pixel based visual data mining of
geo-spatial data. Computers &Graphics 28(3), 327–344 (2004)
27. Kerren, A., Stasko, J.T., Fekete, J.-D., North, C.J. (eds.): Information Visualiza-
tion. LNCS, vol. 4950. Springer, Heidelberg (2008)
28. Krúger, J., Schneider, J., Westermann, R.: Clearview: An interactive context pre-
serving hotspot visualization technique. IEEE Transactions on Visualization and
Computer Graphics 12(5), 941–948 (2006)
29. Maimon, O., Rokach, L. (eds.): The Data Mining and Knowledge Discovery Hand-
book. Springer, Heidelberg (2005)
30. Meliou, A., Chu, D., Guestrin, C., Hellerstein, J., Hong, W.: Data gathering tours
in sensor networks. In: IPSN (2006)
31. Mitchell, T.M.: Machine Learning. McGraw-Hill, Berkeley (1997)
32. Naumann, F., Bilke, A., Bleiholder, J., Weis, M.: Data fusion in three steps:
Resolving schema, tuple, and value inconsistencies. IEEE Data Eng. Bull. 29(2),
21–31 (2006)
33. North, C.: Toward measuring visualization insight. IEEE Comput. Graph.
Appl. 26(3), 6–9 (2006)
34. Perner, P. (ed.): Data Mining on Multimedia Data. LNCS, vol. 2558. Springer,
Heidelberg (2002)
35. Schumann, H., Müller, W.: Visualisierung - Grundlagen und allgemeine Metho-
den. Springer, Heidelberg (2000)
36. Shneiderman, B.: Tree visualization with tree-maps: 2-d space-filling approach.
ACM Trans. Graph. 11(1), 92–99 (1992)
37. Shneiderman, B., Plaisant, C.: Designing the User Interface. Addison-Wesley,
Reading (2004)
38. Spence, R.: Information Visualization. ACM Press, New York (2001)
39. Thomas, J.J., Cook, K.A.: Illuminating the Path. IEEE Computer Society Press,
Los Alamitos (2005)
40. Tricoche, X., Scheuermann, G., Hagen, H.: Tensor topology tracking: A visual-
ization method for time-dependent 2d symmetric tensor fields. Comput. Graph.
Forum 20(3) (2001)
41. Unwin, A., Theus, M., Hofmann, H.: Graphics of Large Datasets: Visualizing a
Million (Statistics and Computing). Springer, New York (2006)
 Visual Analytics: Definition, Process, and Challenges 175

42. van Wijk, J.J.: The value of visualization. In: IEEE Visualization, p. 11 (2005)
43. Widom, J.: Trio: A system for integrated management of data, accuracy, and
lineage. In: CIDR, pp. 262–276 (2005)
44. Yi, J.S., Kang, Y.a., Stasko, J.T., Jacko, J.A.: Toward a deeper understanding
of the role of interaction in information visualization. IEEE Trans. Vis. Comput.
Graph. 13(6), 1224–1231 (2007)
 Author Index

Andrienko, Gennady, 154 Kerren, Andreas, 65

Andrienko, Natalia, 46 Kohlhammer, Jörn, 154

Carpendale, Sheelagh, 19, 92 Melançon, Guy, 154

Munzner, Tamara, 134
Dykes, Jason, 65
North, Chris, 1
Fekete, Jean-Daniel, 1, 154
Purchase, Helen C., 46
Görg, Carsten, 154

Heer, Jeffrey, 92 Stasko, John T., 1, 65

Isenberg, Petra, 92 van Ham, Frank, 92

van Wijk, Jarke J, 1
Jankun-Kelly, T.J., 46
Ward, Matthew, 46
Keim, Daniel, 154 Weaver, Chris, 92