6

Information visualization

6a Basics of visualization
Information visualizations are a part of everyday communications and
scholarship. These graphics have powerful rhetorical force. The visualiza-
tions are often more easily consumed than the complex research data on
which they depend. Understanding the process by which visualizations are
made helps bring into focus what they show and what they conceal.
All information visualizations are metrics expressed as graphics. The
implications of this simple statement are far ranging. Data can be very dif-
ficult to interpret in tabular form. Very few individuals are skilled at reading
spread sheets, let alone relational databases, to make sense of information.
A query might produce thousands of data points. Information visualizations
are used to make this quantitative data legible. They are particularly useful
for seeing patterns in large amounts of information, making these apparent
in a condensed form.
Anything that can be quantified (given a numerical value) can be turned
into a graph, chart, diagram, or other visualization.
Points, lines, and areas can be plotted using analog tools—paper and
colored pencils—and many of the formats used in digitally produced visu-
alizations are centuries old. The process of making graphs by hand is slow
and deliberate. Each point has to be marked, each line created by connecting
dots or using mathematical formulae, and each area calculated. At each step
of hand-drawing a graph or chart, we reflect on how it is made.
But the ease of production afforded by computational means makes it
possible to create polished and sophisticated graphics without critical reflec-
tion. We can easily overlook the fact that all parts of the process—from
creating quantified information to producing visualizations—are acts of
interpretation. In addition, the ability to read a visualization requires under-
standing the semantics of graphic formats. Visual forms create meaning,
they don’t just display it. A bar chart makes a different statement than a pie
chart, for instance, and such insights are crucial to the critical engagement
with information visualization (Lengler and Eppler 2007).
 Information visualization 87
Benefits and liabilities
To begin, consider the two components of a visualization separately—the
metrics and the graphics. Here are two versions of the same information, a
table and a bar chart:

Figure 6.1a Segment of a table and 6.1b Bar chart generated from the same informa-
tion (JD)

The table is not very complicated, it puts dates in one column and num-
ber of pages output by an author into a second one. All of the information
in it makes good sense but trying to read columns of numbers to see a
pattern in them is difficult. The chart makes clear that a steady output of
pages occurred in 1972, matched by one spike in 1971, and followed by
88 Information visualization

Figure 6.1a Continued

low output in 1973. The comparison of values is easily done in the visual
format, and if we imagine extending the table to include hundreds or thou-
sands of data points, this fact would be even more dramatically clear.
What is the relationship of the data to the visualization? In this situation,
a line of dates is charted on the x-axis and a set of values is indicated by the
y-axis. The conventions of charts make this easy to read and even intuitive
in layout. But is there an inherent visual form in the data? One interesting
exercise is to put the same data into other graphical formats to see what
happens. Here are two examples of the same data but in a line chart and a
pie chart.
We are immediately confronted with the question of what features of the
graphical display are meaningful. For instance, the continuous line on the left
graphing the dates suggests that the rate of change in the data about pages
is a significant factor. But the “number of pages” data is actually a discrete
value. While the bar chart compares the values of each segment to each other,
the line chart makes these part of a continuous process, though this is not the
case. By contrast, the pie chart suggests that each entry is part of a whole—
that the sum total of pages is significant, not the difference in their value. The
values are hard to compare, the dates are lost entirely, and the concept of
the “whole” of the author’s output has no meaning. Neither of these charts
makes the correlation of date and page output as clear as the initial bar chart.
These are both “bad” graphics (and possibly bad data as well).
The point is that nothing in the data dictates the form of the visualiza-
tion. These and a host of other charts can be generated from the same data.
 Information visualization 89

Figures 6.2 and 6.3 Other visualizations of the same data in Figure 6.1 (JD)
90 Information visualization
Any data set can be put into a pie chart, a continuous graph, a scatter plot,
a tree map, and so on. The challenge is to understand how the information
visualization creates an argument and then make use of the graphical format
whose features serve your purpose. Any sense that data have an inherent
visual form is an illusion. [See Exercise #1: A range of graphs.]
Data creation, as we noted in earlier (see Sections 2a and 2b), depends
on parameterization. As stated before, this means that anything that can be
measured, counted, or given a metric or numerical value can be turned into
data. The concept of parameterization is crucial to visualization because the
ways in which we assign value to the data will have a direct impact on the
ways it can be displayed. Visualizations are convincing by virtue of their
graphic qualities and can easily distort the data. While all visualizations are
interpretations, some are more suited to the structure of a given data set
than others.

Visualization basics
In many cases, the graphic image is an artifact of the way the decisions
about the design were made, not about the data. Understanding some basics
of the relation between graphics and metrics is essential.
Here are some fundamental guidelines for thinking about which chart
to use:

• The distinction between discrete and continuous data is one of the most
significant decisions in choosing a design. Example: in visualizing the
height of students in a class, making a continuous graph that connects
the dots makes no sense at all. There is no continuity between the height
of one student and another. Individual height is a discrete value.
• If you are showing change over time or any other variable, then a con-
tinuous graph is the right choice. Example: Change in height for indi-
vidual students over a five-year period.
• If a graph shows quantities with area, use it for percentages of a whole,
like a pie chart, not comparative value. If you increase the area of a
circle by length of the radius, or a square from the length of the side,
you are introducing distortion into the relation of the elements. This is a
common error. Example: The population in the town doubled from ten
thousand to twenty thousand in five years. The data is visualized with
two squares on a map, with the second having its sides twice the length
of the first (10,000 to 20,000). But the area of the second square four
times that of the first, not double.
• The way in which you label and order the elements in a chart will make
some arguments more immediately evident. If you want to compare
quantities, be sure they are displayed in proximity. Example: when
comparing the population size of states should you put the states in
 Information visualization 91

Figure 6.4a Meaningless graph of height among a group of girls graphed continuously

and 6.4b Graph of the change of one girl’s height changing over time (JD)

alphabetical order or put the data in size order? Which is going to make
the information more legible?
• The use of labels is crucial and their design can either aid or hinder leg-
ibility. Where are the labels? How much work are you adding to your
reader’s experience?
• Another consideration and challenge is the choice of a scale. When
values are relatively close, the scale of the chart can be kept consistent.
92 Information visualization

Figure 6.5 Classic error in which a value increases numerically but the area increases
geometrically. The quantity on the right is twice that on the left, but the
area is four times as large (JD)

But imagine the charts of date and page outputs in the example above
if in one year the author produced 2000 pages. To show this value,
the scale would need to extend to forty times its current height. The
result would be that the difference between 20 pages and 50 pages
would barely register. The legibility of the graph and patterns would
be altered. To deal with such anomalies, charts are drawn with “bro-
ken” or modified scales, leaving a gap between lower and upper values.
These gaps need to be noted and taken into account in some kind of
legend, labeling, or documentation. [See Exercise #2: Reverse engineer-
ing a visualization.]

The rhetoric of graphics

Every visualization has a history to its format (Friendly 2007). The earliest
forms of visual records seem to have been observations of the planets and
other natural cycles. Early accounting systems for tracking inventory and
also for taking census information used tabular forms. These allow easy
correlation across values. The notion of continuous graphs, line charts, and
other visual representations of information from natural or social phenom-
ena did not appear until modern times. These emphasize continuous change.
William Playfair, the 18th century statistician, is credited with the invention
of many forms of bar chart and continuous graph still in use today. Playfair
was working with what he called “Political Arithmetik,” or the tracking
of information relevant for guiding politics and policy in economic arenas
(Norman 2004–2020). Playfair’s visual solutions were very elegant as well
as highly legible. Keep in mind that the science of statistics is also relatively
modern, originating chiefly in the 17th century with techniques developed
 Information visualization 93

Figure 6.6a and 6.6b Charts showing ordering and labels: The first chart makes it
easy to find individuals by name, the second makes it easy to
compare heights and correlate with names (JD)

by the French mathematicians, Blaise Pascal and Pierre de Fermat, to gauge

the risks of gambling (Apostol 1969).
The power of visualizations has been understood for a long time. In the
19th century, the nurse and activist Florence Nightingale created a specific
format—known as the cockscomb because of its resemblance to the rooster’s
crown—to make her point about the fact that more deaths occurred among
the wounded in field hospitals than on the battlefield. She deliberately chose
94 Information visualization

Figure 6.7 The scale has to stretch to include the height of the outlier and makes it
difficult to compare the differences among the close values in the middle
range. Making a “break” in the scale could allow focus on the area in
which the meaningful information is present (JD)

a format that exaggerated this information. She used the difference in her
data values to set the length of a radius in a circular form, also known as a
polar area diagram, thus distorting the area. (This is similar to the example
of the square, above, but here the area is calculated by the standard formula
A = π r² (area = pi x square of the radius r). The contrast was dramatic, and
she won her argument.
This kind of exaggeration can be very misleading in any chart that
uses area as a feature of its graphical form. As already noted, when using
graphics that are based on area, such exaggerations are built in. This
distortion is a regular feature of information display on maps, as will be
seen ahead.
Figure 6.8 William Playfair Chart of the National Debt, The Commercial and Political Atlas, 1786 (Public domain)
Information visualization 95
96 Information visualization

Figure 6.9 Florence Nightingale, cockscomb diagrams, 1854–55 (Public domain)

Figure 6.10 Graphic variables (JD)

 Information visualization 97
Components of visualizations
When considering visualizations, a few fundamentals besides the type of
chart and the rhetoric of its impact can be useful in guiding design decisions.
The components of visualizations include axes, elements, scales, order/
sequence, values of coordinates, and the graphic variables.
Axes establish the basis for mapping values. Typically, the x-axis (left
to right) is used to graph a value that changes over time (dates) while the
y-axis is used to chart a specific value (cost of living, sea levels, etc.) against
it. These are sometimes, but rarely, augmented with a z-axis that gives depth
to the chart. However, mapping a third variable is trickier than it sounds,
and it is often easier to simply have this information displayed as a second
set of points or lines (earnings might change, for instance, as page outputs
increased in our reference example). The basic coordinate x-y system was
invented by René Descartes and is sometimes referred to as a Cartesian
grid. (The apocryphal version of the story is that he was trying to figure out
how to pinpoint the location of a fly on his ceiling (Wild Maths n.d.).) The
grid uses standard metrics that remain the same across the full extent of the
chart. One question to ask is whether you can imagine situations in which a
metric might need to change? What conditions might require an alteration
in one area of a neutral grid. Is every square the same—even if a spider is
lurking in one?
Elements are the bars, lines, points, symbols, or other features that express
value. They are always read against the axes. Even in a pie chart, the per-
centage is read in relation to an axis—this is the circumference, which forms
the 100% boundary of the whole.
Scales set the specific metric to be used—inches, feet, number of units,
dates, and so on. Scales have a start and endpoint. If I am measuring the
difference in height among a group of giant statues, all of which are over
twenty feet high but less than twenty-one feet high. Should I measure them
in light years? Millimeters? Some scales are too large or too small to be use-
ful. If I am measuring the occupancy of an airport, a scale of years might be
too large, but a scale of seconds will be too small.
Order and sequence are generally determined by data and given as logi-
cal an expression as possible. Putting the work of an author into size order
might be trivial and putting all of the paintings in the world into a single date
sequence might be meaningless. The order and sequence should be meaning-
ful to the research—and for communicating information in a visualization.
Values of coordinates are generated by the axes. But a major difference
exists between the value of crossing points—where one axis intersects
another—and continuous values within the lines, grid, or tick marks. Are
discrete or continuous values being gauged and presented?
Finally, graphic variables are the features of visual language: color, tonal
value, size, shape, orientation, position, and texture. These will be revis-
ited in the discussion of mapping but designating a variable for a specific
98 Information visualization
purpose makes good sense. Shape is very legible, so distinguishing differ-
ent data types with stars, circles, squares, triangles, and other icons in a
chart makes for legibility—provided there are not too many types of data.
Tonal value is useful for showing changes of intensity, as in heat maps. Size
generally indicates quantity, but can signal importance, particularly with
typography. Color, like shape, is very legible and makes distinctions highly
visible, as does texture. Use color to distinguish themes or topics (the height
of freshmen relative to the height of seniors). Dotted lines are easily distin-
guished from solid ones. These allow information to be carried by the visual
elements, not just their labels. Orientation should be used when a feature
of the data correlates to it—like wind direction. Position is generally deter-
mined by coordinates but can also be part of the overall design—what is
near what and why when proximity is significant.
Using graphic variables systematically increases the communicative leg-
ibility of your visualization. [See Exercise #3: Analyze the data-graphic
connection.]

Checklist for visualizations

• Assess your data: Is it composed of discrete or continuous information?
• Choose the appropriate scale: too small a scale may make the important
differences in value hard to spot and too large may exaggerate it. If out-
liers stretch the scale for a few data points, consider a gap in the scale
and an explanation.
• Is the labeling efficient for use? What order should the information
take to be meaningful and usable (alphabetical order of country names
makes them easy to find but might separate values and make them hard
to compare visually)?
• Use graphic variables carefully: shapes carry information readily, tonal
values should be used for data that has a gradient, texture has little
“meaning” in itself, and color can carry symbolic value or simply be
used for differentiation.
• Proximity of labels to values is optimal for reducing cognitive load;
make it easy for the viewer to correlate information.
• Never use changes in area to show a simple arithmetic increase in
value.
• Review the graphic to see if it contains elements that are “incidental”
artifacts of production rather than semantically meaningful ones.
• While illustrations, images, or exaggerated forms may be considered
“junk,” they can also help set a theme or tone when used effectively.

A few last thoughts

Visualizations do not usually show the lifecycle of the data. Decisions
about parameterization, even the way samples were taken and what
 Information visualization 99
elements of the data were “cleaned” up and removed are all missing from
the final visualization. Similarly, the history of the data within its insti-
tutional or research context may not be documented. Finding the source
for the information can be difficult once the visualization exists. Thus,
the question of whose authority—whose voice and point of view—is rep-
resented in the visualization can be very difficult to answer. A process
of reduction, simplification, and what is known as reification—making
a concept appear to be a thing (solid, tractable, and understandable)—
takes place in the production of visualizations. The statement, “Informa-
tion visualizations are reifications of misinformation,” suggests that the
apparent straightforward communication in a visualization should be
treated with skepticism, rather than simply accepted, in spite of the value
of these images for data presentation (Fenton 2015). In data journalism,
these concepts are referred to as “the lie factor,” and ethical practition-
ers work conscientiously to avoid misleading graphics. [See Exercise #4:
Misleading graphics.]
Recent scholarship draws attention to critical concerns in this area
of digital research. The work of feminist scholars questions some of
the assumptions about who controls the technology of production and
whose values are embodied in the information design process (D’Ignazio
and Klein 2016). A cache of hand-drawn works by the African-Ameri-
can activist, W.E.B. Du Bois, sheds light on this formerly little-known
aspect of his work and the way he made use of data visualization for
advancing critical discussions of race (Mansky 2018). Their hand-
drawn quality inflects their presentation, raising questions of equitable
access to resources. A very different approach to hand-drawn visualiza-
tions appeared in a “Dear Data” project of letters exchanges between
Georgia Lupi and Stephanie Posavec, both sophisticated information
designers who used the experiment as a way to explore the possibilities
of analog presentation (Lupi 2017). Many artists have been intrigued
by data flows and visualizations as opportunities for aesthetic investi-
gation, some of which will be touched on ahead in the discussion of
complexity.

Takeaway
Information visualizations are metrics expressed as graphics. Information
visualizations allow large amounts of (often complex) data to be depicted
visually in ways that reveal patterns, anomalies, and other features of the
data. No data has an inherent visual form. Any data set can be expressed
in any number of standard formats, but only some of these are appropri-
ate for the features of the data. Certain common errors include misuse
of area, continuity, and other graphical qualities. The rhetorical force of
visualization is often misleading. All visualizations are interpretations, not
presentations of fact. Some graphic features of visualizations are artifacts
100 Information visualization
of the display, not of the data, and can contribute to the reification of
misinformation. Understanding the language of graphics is an art that
combines conceptual insight with design acuity. Still, even a novice can
produce useful graphics with current platforms and tools. The challenge
is to produce graphics that are appropriate to the research task and com-
munication of arguments.

Exercises

Exercise #1: arange of graphs

Try various visualizations for suitability. Take one of these data sets through
a series of Microsoft Excel visualizations. Which make the data more leg-
ible? Less?

• United States AKC Registrations

http://images.akc.org/pdf/archives/AKCregstats_1885-1945.pdf
• Sugar Content in Popular Halloween Treats
www.popsugar.com/fitness/Calories-Halloween-Candy-Fun-Size-
Treats-5452936

Exercise #2: reverse engineering a visualization

Look at Google’s Public Data directory and the visualizations generated
from the files. Can you locate the basic components (axes etc.) and evaluate
them for common errors? Consider where the data comes from and what
may be missing from its visualization.
www.google.com/publicdata/directory

Exercise #3: analyze the data-graphic connection

Imagine you are collecting data from the classroom on 1) classroom use, 2)
attention span of students, 3) snack preferences, 4) age, height, and weight
comparisons in a group? For what kind of data gathered in the classroom
would you use a column chart? Browse this D3 gallery of visualizations
for other formats: https://observablehq.com/collection/@observablehq/
visualization

Exercise #4: misleading graphics

What is the concept of the “lie factor” and how is it visible at the following
link?
www.datavis.ca/gallery/lie-factor.php
 Information visualization 101
In each case consider legibility, accuracy, or the argument made by the
form. What is meant by a graphic argument?

Recommended readings
D’Ignazio and Lauren Klein. 2016. “Feminist Data Visualization.” IEEE. www.aca
demia.edu/28173807/Feminist_Data_Visualization.
Drucker, Johanna. 2011. “Humanities Approaches to Graphical Display.” Digital
Humanities Quarterly. www.digitalhumanities.org/dhq/vol/5/1/000091/000091.
html.
Lupi, Giorgia. 2017. “Data Humanism: The Revolutionary Future of Data Visuali-
zation.” PRINT. www.printmag.com, www.printmag.com/post/data-humanism-
future-of-data-visualization.

6b Networks and complex systems

The concept of a network has become ubiquitous in current culture (Zer-
Aviv 2016). Almost any connection to anything else can be called a network,
but properly speaking, a network has to be a system of elements or entities
that are connected by explicit relations. The term network is frequently used
to describe the infrastructure that connects computers to each other and to
peripherals, devices, or systems in a linked environment. While that is an
accurate description, the networks we are concerned with in digital humani-
ties are created by relationships in an information system. This might be the
connection of books to authors, paintings to collections, people in commu-
nication with each other, or objects and ideas in circulation.
Unlike other data structures we have looked at—databases, markup sys-
tems, classification systems, and so on—networks are defined by the specific
relations among elements in the system rather than simply by the content
types or components. The elements of networks are nodes (points or enti-
ties) and edges (links or relations that connect the nodes).
Good examples of networks are social networks, traffic networks, com-
munication networks, and networks of markets and/or influence. Many of
the same diagrams are used to show or map these networks, and yet, the
content of the relations and of the entities might be very different in each
case. Standardization of graphic methods can create a problem when the
same techniques are used across disciplines and/or knowledge domains, so a
critical approach to network diagrams is useful.
Technically, networks are graphs, not visualizations. The distinction
is important because graphs can include the feature of directed or undi-
rected connections. These indicate a one-way (or two-way) movement in
the connection. For example, money may flow from a parent to a child, but
more rarely flows back in the other direction. Influence may move from a
102 Information visualization
predecessor in a field to a new development, but might not flow both ways,
particularly if the author of the earlier work is deceased.
The computational process by which graphs are produced requires that data
be structured in a specific way: source > relationship > target. The vocabulary
of nodes and edges is used to differentiate entities (source and target) from their
relationships (edges). Particular features of networks are used to process the
data in relation to notions of centrality, closeness, and between-ness. Central-
ity is the measure of how important any particular node is, measured by the
number of connections and type (to or from other nodes) (Bhasin 2019). In
graph theory, which governs the description and production of networks, other
factors are gauged to assess the factors of between-ness and closeness based on
the pathways established among and through nodes. The important principle
here is that while some features of the display can be read literally (numbers of
connections and directions), the literal distance of nodes from each other in a
visualization can only be read logically. This is because the display algorithms
try to preserve the statistical features of the data but are often optimizing leg-
ibility at the same time. As with all visualizations, it is important to be careful
about reading the visual display literally. A node pulled out to a great distance
might simply be far from the center so that its label can be seen.

Sketching network concepts

You can sketch a network on paper quite easily. Imagine yourself as a
node and then draw lines to everyone you know in your immediate cir-
cles (family, friends, clubs, and groups) around you. Draw their links to
each other. Think about degrees of proximity and also connections among
the individuals in different parts of your network. How many of them are
linked to each other as well as to you? If you can code the lines that con-
nect persons to indicate something about the relationship, how does that
change the drawing? What attributes of a relationship are readily indi-
cated? Which are not? Think about the difference between how often you
exchange communications with someone and how central they are to the
exchanges among others. A parent might be someone to whom everyone
is connected, but your own communications might be more frequent with
your siblings. When a network algorithm processes data, it tries to calcu-
late these properties.
Social networks are familiar and the use of social media has intensified
our awareness of the ways social structures emerge from interconnections
among individuals. A network may or may not have emergent properties,
may or may not be dynamic, and may have varying levels of complexity.
Simple networks, like the connection of your computer to various periph-
eral devices through a wireless router in your home environment, may
exhibit very little change over time, at least little observable change. But a
network of traffic flow is more like a living organism than it is like a set of
static connections. Though nodes may stay in place, as in airline hubs and
 Information visualization 103
transfer points, the properties of the network have capacity to vary consid-
erably. This is certainly true with social networks, most of which are highly
dynamic, even volatile.

Properties of networks
Networks exhibit varying degrees of closed-ness and open-ness. Researchers
interested in complex or emergent systems are attentive to the ways bound-
ary conditions are maintained under different circumstances, helping to
define the limits of a system. Social networks are almost never closed, and
like kinship relations or communications, they can quickly escalate to a very
high volume of connections. Epidemiologists trying to track the spread of a
disease are aware of how rapidly the connections among individuals grow
exponentially in a very short period of time. Network analysis is an essential
feature of textual studies, particularly of citations and influences. Network
analysis plays a large role in policy and resource allocation as well as in
other kinds of research work.
To reiterate what was already stated, the basic elements of any net-
work are nodes and edges. The degree of agency or activity assigned to
any node and the different attributes that can be assigned to any rela-
tion or edge will be structured into the data model. The data for linked
“nodes” are understood as “source” and “target” (even though these can
be reciprocal, and also, unrelated). Edges are the connections specified
between the nodes.
For an example of this in action, look at the project, Kindred Britain,
which studies connections of about 30,000 British individuals. The project
is meant to show the many ways in which connections form through social
networks, family ties, business, and political circumstances.
Another interesting example looks at the genre of “exchange poems” that
were part of medieval Chinese culture. These had traditionally been char-
acterized by schools and styles. But new research positioned them in social
networks. To paraphrase the work of the project director, Tom Mazanec,
it turns out that the Buddhist monks in the 7th to 10th centuries of the
Tang dynasty were central “nodes” in the network of literary production
(Mazanec 2017). Graphing these has changed the way this form of Chi-
nese poetry is understood and its place in cultural and social life. Relations
between literary forms and social activity that were not noted before were
revealed through the analysis.
Art historians Pamela Fletcher and Anne Helmreich used network analy-
sis to study the London art market, and found surprising insights from sales
records and auction catalogs (Fletcher and Helmreich 2012). Artists and
styles that have not necessarily been seen as important by later art historians
turned out to play a significant role in the markets of the time, even if they
have largely vanished from the canon. [See Exercise #1: Kindred Britain, a
social network project.]
104 Information visualization

Figure 6.11 Network graph, edgelist, and nodelist (Image courtesy of Nick Schwi-
eterman) (NS)

Tools and tutorials

As is the case with most digital methods, the fundamental principles of networks
are well-established but the tools and platforms change over time. The princi-
ples that need to be understood are those of the data structure just mentioned
(source > relation > target). The relations among entities, known as edges, can
carry weight and also annotations. Network graphs are generated using statis-
tical processes. In a network, the graph is generated by statistical assessment
of frequency and weight of relations. With a massive data set—think of a day
of Twitter feeds—these calculations become highly complex. Though a simple
network can be sketched by hand (you know who is in your immediate social
circles), generating a graph of a complex data set would be almost impossible
without computation since it involves calculation of relative values across a
number of variables (frequency, weight, directionality, etc.).
One tool frequently used in digital scholarship for generating network
diagrams is Cytoscape, an open-source platform that can be downloaded
directly from the Web and installed on a laptop or desktop. Gephi is
another. Learning to use such a tool takes some time but has the advan-
tage that it is a professional level program designed to handle data at
 Information visualization 105
every scale from small to large. Understanding the data model before you
begin—what is connected to what and how are you characterizing the
links or relations—is essential if the digital tools are to be used effectively.
Conceptualizing the network before it is visualized automatically helps
keep a critical view of what the program produces. [See Exercise #2: Com-
paring network diagrams.]
Cytoscape and Gephi make use of data in a number of structured formats,
among them, some specifically designed for graphs (Rana 2018). These
include GraphML, or Graph Markup Language, standards for networks.
They readily store information about labels and attributes of the nodes and
edges. CSV, Comma Separated Values, a common format in all spreadsheets,
can be used to define nodes and edges in weighted to/from pairs (or in linked
pathways). GEXF, or Graph Exchange XML Format, was designed by the
developers of Gephi, another free, open-source tool for network produc-
tion and analysis. Networked data formats have specific requirements and
a strict syntactic structure. Creating a small data set to work through the
tutorials with your own examples is the best way to see what is happening
at each step. [See Exercise #3: Cytoscape tutorial.]
Though the data structure is critical in network diagrams, learning to
read the graphical output generated is also important. In an initial visu-
alization, especially of large datasets, networks tend to look like “hair-
balls.” They are tangles of lines connecting points, often very densely
packed into a small image. Working to open up the nodes and stretch
the edges allows insight into the ways the network is branching and
where the main areas of connection lie. As has been mentioned above,
keep in mind that a network diagram display conforms to protocols that
optimize screen space for legibility. While the relationships in a network
display are generally accurate, the literal distances on the screen are not.
Attaching semantic value—meaning—to the spatial placement of the
nodes, can be misleading.
A final challenge is visualizing dynamic systems in static form. Very few
social networks are static, though the analysis of historical materials, con-
nections, and activities may be. Information in the data set needs to be
carefully scrutinized to be sure that events from different time periods or
unrelated events are not conflated into a single graph.

Complex systems
Systems that follow non-linear processes of development are called complex.
This does not mean complicated. A complex system can be as simple as a
relationship between two people, a person and an environment, or an envi-
ronment and changing conditions (Clemens 2019). What makes it complex
is that the development of the system cannot be predicted—because the pro-
cesses are non-linear and/or non-deterministic from a statistical standpoint.
106 Information visualization
The conditions in which they emerge continue to change and elements in the
system interact in unpredictable ways. Weather systems are a paradigmatic
example of complex systems, but so are stock markets, political processes,
social relations of all kinds, and cultural activities. Who could have pre-
dicted that a conceptual artist named Marcel Duchamp would confound
the conventions of the Western art world in 1917 by displaying a urinal
upside down in an exhibit? Or that Mao Tse Tung would come to power in
the Chinese Revolution? Or that the presence of the Missions in Australia
would create an opportunity for art practices that were 20,000 years old to
become codified in the medium of paint on oil and board? (Artlandish n.d.).
These are examples of complexity at work. Many—even most—cultural
processes are complex but modeling these requires more than creation of a
data set. This work involves modeling behaviors of agents and conditions
in a system.
Information designers—and artists—have been intrigued with visualiz-
ing complexity. Art exhibitions featuring data aesthetics have become com-
mon (Remondino, Stabellini, and Tamborrini 2018). The result has been a
rich vocabulary of vivid and dynamic information visualizations—as well
as some “eye candy” that may be more seductive than meaningful (Lima
2013). The process of constructing data and formulae for visualizing com-
plexity is more complicated than it is for other visualizations (Yau 2007–
2020). [See Exercise #4: Complexity.]
Advanced network theory pays attention to emergent properties of
systems. The capacity of networks to “self-organize” using very sim-
ple procedures that produce increasingly complex results makes them
useful models for looking at many kinds of behaviors in human and
non-human systems. Networks do not have to be dynamic, but complex
systems almost always are. The study of systems theory and of networks
is relatively recent and only emerged as a distinct field of research in
the last few decades. We might argue, however, that novelists and play-
wrights have been observing social networks for much longer, as have
observers of animal behavior, weather and climate, and the movements
of heavenly bodies held in relation to each other by magnetism, gravity,
and other forces. Most dynamic phenomena are complex systems gov-
erned by non-linear processes.

Takeaway
Networks consist of nodes (entities) and edges (relations). The data
model for a network is a simple three-part formula of entity-relation-
entity. This can be structured in a spreadsheet and exported to create a
network visualization. Networks emphasize relations and connections
of exchange and influence. Refining the relations among nodes beyond
the concept of a single relation is important and so is the change of
 Information visualization 107
relations over time. Social networks change constantly, as do communi-
cation networks, and the relations among the technology that supports
a network and the psychological, social, or affective bonds can alter
independently.

Exercises

Exercise #1: Kindred Britain, a social network project

Explore the site and then discuss the selection of individuals, the character
and quality of relations, explicit assumptions and implicit ones, and the
diagrams and their rhetorical power.
http://kindred.stanford.edu/#

Exercise #2: comparing network diagrams

Go to: https://linkedjazz.org/network/ Determine what information you can
reasonably extract from this graph. Now toggle between modes. Does this
change your understanding? Or go to: www.databasic.io/en/connectthedots/
Network visualization with interactive sample data sets created by Rahul
Bhargava and Catherine D’Ignazio.

Exercise #3: Cytoscape tutorial

This manual can be accessed without downloading and goes step by step
through the basics of network graph construction. It is provided free of
charge by the people who designed and maintain the standard platform for
this work. Read through the table of contents and introduction to get ori-
ented. http://manual.cytoscape.org/en/stable/Introduction.html

Exercise #4: complexity

Look at half a dozen examples on Nathan Yau’s site: https://flowingdata.
com/about/
What are the dimensions added in complex systems that are different
from those of static visualizations? What is the correlation between graphic
expression and information?
What role does aesthetics play in these projects?

Recommended readings
Grandjean, Martin, and Aaron Mauro. 2015. “A Social Network Analysis of Twit-
ter: Mapping the Digital Humanities Community.” Cogent: Arts and Humanities
3 (1). www.tandfonline.com/doi/full/10.1080/23311983.2016.1171458.
108 Information visualization
Weingart, Scott. 2011. “Demystifying Networks, Parts I & II Journal of Digital
Humanities.” Journal of Digital Humanities 1 (1). http://journalofdigitalhumanities.

References cited
Apostol, Tom. 1969. “A Short History of Probability.” In Calculus, Vol. II. John
Wiley & Sons. http://homepages.wmich.edu/~mackey/Teaching/145/probHist.
html.
Artlandish. n.d. “Australian Aboriginal Art.” www.aboriginal-art-australia.com/
aboriginal-art-library/the-story-of-aboriginal-art/.
Bhasin, Jasin. 2019. “Graph Analytics—Introduction and Concepts of Central-
ity.” Towards Data Science. https://towardsdatascience.com/graph-analytics-
introduction-and-concepts-of-centrality-8f5543b55de3.
Clemens, Marshall. 2019. “Visualizing Complex Systems.” New England Complex
Systems Institute. https://necsi.edu/visualizing-complex-systems-science.
Fenton, William. 2015. “Humanizing Maps: An Interview with Johanna Drucker.” PC.
www.pcmag.com/news/humanizing-maps-an-interview-with-johanna-drucker.
Fletcher, Pamela, and Anne Helmreich. 2012. “Local/Global: Mapping Nineteenth-
Century London’s Art Market.” Nineteenth Century Art Worldwide 11 (3).
www.19thc-artworldwide.org/autumn12/fletcher-helmreich-mapping-the-london-
art-market.
Friendly, Michael. 2007. “DataVis.” www.datavis.ca/index.php.
Lengler, Ralph, and Martin J. Eppler. 2007. www.visual-literacy.org/periodic_table/
periodic_table.html.
Lima, Manuel. 2013. Visual Complexity: Mapping Patterns of Information.
New York, NY: Princeton Architectural Press. https://medium.com/@mslima/
visualcomplexity-com-ad9a12fa2c1a.
Lupi, Georgia. 2017. “Dear Data, the Project.” http://giorgialupi.com/dear-data.
Mansky, Jackie. 2018. “W.E.B. Du Bois’s Visionary Infographics Come Together
for the First Time in Color.” Smithsonian Magazine. www.smithsonianmag.com/
history/first-time-together-and-color-book-displays-web-du-bois-visionary-info
graphics-180970826/.
Mazanec, Tom. 2016–17. “Chinese Exchange Poems.” https://cdh.princeton.edu/
projects/chinese-exchange-poems/.
Norman, Jeremy. 2004–2020. “The History of Information.” www.historyofinfor
mation.com/detail.php?entryid=2929.
Rana, Ashish. 2018. “Getting Started with Network Data Sets.” Towards Data
Science. https://towardsdatascience.com/getting-started-with-network-datasets-
92ec54958c07.
Remondino, Chiara L., Barbara Stabellini, and Paolo Tamborrini. 2018. “Exhibition:
Visualizing Complex Systems.” https://systemic-design.net/wp-content/uploads/
2019/05/RSD7Exhibition_VisualizingComplexSystems.pdf.
Wild Maths. n.d. https://wild.maths.org/rené-descartes-and-fly-ceiling.
Yau, Nathan. 2007–2020. “Flowing Data Site.” https://flowingdata.com.
Zer-Aviv, Mushon. 2016. “If Everything Is a Network, Nothing Is a Network.”
Visualising Information for Advocacy, visualisingadvocacy.org. https://visualisin-
gadvocacy.org/node/739.html.
 Information visualization 109
Resources
Cytoscape https://cytoscape.org/.
Gephi ttps://gephi.org/.
Kindred Britain http://kindred.stanford.edu/#.
Network Graphs (Flourish Studio) https://app.flourish.studio/@flourish/network-
graph.
Social Network Graphs https://gwu-libraries.github.io/sfm-ui/posts/2017-09-08-sna.