You are on page 1of 13

A taxonomy of INTRODUCTION

graph types
Classifications have a long history in scientific
endeavours. Often simply serving to neatly
store myriad characteristics of the subject be-
ing studied, they may also be applied in a more
ROBERT RANKIN active role. A carefully designed classification
can serve to show not only the full range of
available possibilities but also the relation-
A classification of the graph types
used throughout many scientific ships between these, and in so doing acts more
fields is presented in order to as an instrument rather than simply as a 'filing
show the range of possible meth­ cabinet'.
ods of graphically representing In the field of graphic communication, the
mathematical formulae and data. criteria used to sort illustrations are quite di-
The classification is presented in
such a way that inter-relationships verse and relationships between classifications
between the various graphic are difficult to find and often only tenuously
forms is clearly demonstrated. evident. Different researchers have used dif-
It is anticipated that such a taxon­ ferent criteria to organize pictures, and these
omy will assist in the graphic criteria not only arrange pictures into different
presentation of information by
providing a designer or author
groups, but actually suggest different ways of
with a wide range of options. looking at the same pictures. Classifications
aim to be exhaustive by attempting to encom-
pass all possibilities, and in so doing serve as a
valuable resource for intending authors, as well
as a basis for further research into communica-
tion using visual materials.
For some time, classifications of graphic
forms have been proposed to serve a variety of
functions, with each case attempting to cate-
gorise the full known range of visual structures
under one all-encompassing criteria. Very
broadly speaking, these taxonomies have either
been functionally or structurally based.

FUNCTIONAL AND STRUCTURAL


CLASSIFICATIONS Author's address
Rankin Publishers
PO Box 406
Functional analyses of pictures in general tend Toowong
to focus on the intended use and purpose of the Queensland 4066 Australia

graphic material. Examples of classifications in © Robert Rankin 1990


this category include Fleming (1967),
MacDonald-Ross (1977a, 1977b), Duchastel

147
Information Design Journal 6:2 (1990), 147–159. DOI 10.1075/idj.6.2.03ran
ISSN 0142–5471 / E-ISSN 1569–979X © John Benjamins Publishing Company
ROBERT RANKIN ■ A TAXONOMY OF GRAPH TYPES

(1978), Duchastel and Waller (1979), Waller As well, it is important that the two dimen-
(1979), and Tufte (1983). By their very nature, sions of the matrix be independent.
these classifications tend to use very broad
categories, each including a range of pictorial
forms. Their categories bear no unique rela- THE PROPOSED CLASSIFICATION
tion to the physical structure of the picture.
In contrast, a classification by Knowlton (1966)
The proposed classification discussed here
tends to include both functional and structural
belongs to the structural class and utilizes the
criteria.
two dimensional approach previously men-
Pure structural analyses focus on the type
tioned. Its scope is more narrow than those
or form of picture irrespective of how it will
already discussed in that it attempts to focus
be used. Two such classifications have been
only on those graphic forms which utilize a
proposed: by Doblin (1979) and Twyman (1979,
co-ordinate space. That is, the dimensions of
1982). In both, a two dimensional matrix net-
the picture are explicitly stated and numeric-
work has been developed as a rigorous method
ally scaled. The simplest example is the well
for generating categories of structural form.
known rectilinear cartesian co-ordinate graph,
Both taxonomies attempt to cover all presenta-
utilizing orthogonal X and Y axes, on which
tion methods of information for the visual
data is plotted or a line drawn to show visually
channel, and thereby include verbal, numer-
some trend or relationship between the X and
ical and pictorial material. Differences occur
Y parameters. In all the taxonomies previously
because of the parameters chosen for each
discussed, this type of graph and its relatives
dimension of the matrix. Twyman has used
have received little attention. In the case of the
method of configuration (7 categories relating
structural analyses, this has been because the
to the physical arrangement of the visual ele-
dimensions chosen for the matrices have not
ments) and mode of symbolization (verbal,
allowed enough resolution in this category of
numerical, pictorial, schematic) to produce a
picture to fully explore the vast range graphs
matrix with 28 categories. Doblin has produced
available.
a matrix by assuming his parameters of infor-
mation content and form of presentation (ver- In many scientific, engineering, medical and
bal, numerical and visual) are independent. commercial fields of communication, it is this
graphical category which receives major use.
Generally speaking, if the dimensions of the
What is proposed here is a comprehensive
matrix and the divisions within each of these
scheme depicting the various forms of graph
dimensions are well chosen, the resulting com-
and their relationships to one another. As a
bination can produce a fairly exhaustive set of
source, it draws upon the range of graphical
possible graphic forms. Biasing can, unfortu-
forms listed in a scientific illustration survey
nately, manifest itself in two contrasting ways:
conducted by Rankin (1986). The classification
1. All categories are given the same weight becomes, as discussed earlier, not only a conve-
irrespective of whether or not visual forms nient taxonomy but also an instrument. Pro-
in a category actually exist. spective authors have the opportunity to use
2. Some categories may not have enough reso- it to choose the most appropriate method to
lution. Many diverse forms may be classified present their data. Used in conjunction with
into one element of the matrix when it may a designing model, such as that described by
well be useful to sub-divide it.

INF. DES. J. 6/2 (1990) 147-59

148
ROBERT RANKIN ■ A TAXONOMY OF GRAPH TYPES

Rankin (1989), an author's task of producing right, the number of dimensions used in the
an effective document is greatly facilitated. prime element is reduced by one per column.
To clarify the terminology used in the fol- The system can best be explained with refer-
lowing discussion, it will be necessary to ence again to Figure 1. Here, mathematical
define three terms. 'Figure' refers to the large functions with up to five variables are dis-
arrays (of which there are four presented here - played graphically. The number of variables
Figures 1 to 4), 'illustration' refers to one divi- in the function is listed in the left-hand margin
sion within the matrix of the figure (In Figure of the figure. In mathematical terms, the math-
1, there are 16 illustrations), and 'prime ele- ematical functions for the top row can be gen-
ment' is a graphic element within one of these erally written as
illustrations. Mostly, a prime element will v = f(t,x,y,z),
mean one set of x, y, z axes, of which there may meaning v is a function of t, x, y and z. That is,
be several making up an illustration. For exam- the value of the dependent variable v depends
ple, the top left hand illustration in Figure 1 on the respective values of the independent
has two prime elements. Explicitly, the prime variables t, x, y and z. Continuing down
element is defined as the most fundamental through the illustration, subsequent rows
graphic form appearing in the illustration illustrate the following simpler mathematical
which could stand alone and still be classed as functions:
a co-ordinate graph. t = f(x,y,z)
The basis of the classification centres on the z = f(x,y)
two variables used to produce the arrays of y-f(x).
illustrations in the four figures. The vertical The basic trend, therefore, in each figure, is
axis of each of the four figures depicts the num- for the same amount of information to be rep-
ber of variables of the mathematical function resented across the figure, while down the fig-
described in each illustration. In Figure 1, the ure, the information is becoming less by one
number of variables ranges from 5 in the top variable per row.
row to 2 in the bottom row. Many of the illustrations generated by this
The horizontal axis of each figure generally method are used frequently in published docu-
describes the number of dimensions represent- ments and have received common names
ed in the prime element of each illustration. which are accepted throughout the scientific
Looking at the top row of Figure 1, the first community. These names are listed in their
illustration has prime elements of three dimen- appropriate position within the array. Other
sions (X, Y, Z), while the second illustration illustrations or parts of them have no particu-
has prime elements of only two dimensions (X, lar established names. The term cylinder-like
Y). has been used to describe graphical surfaces
Using this scheme, the most complex math- which are similar to this familiar shape but
ematical functions (those with the most vari- may not necessarily be perfectly symmetrical.
ables) occupy the top row of each figure and the For the sake of simplicity and clarity, where
least complex the bottom row. The illustra- these surfaces appear in the illustrations, they
tions occupying each horizontal row describe have been drawn as perfect cylinders. The
the same mathematical function by using a same is true for the two-dimensional polar
slightly different illustrating method in each graphs where the radial parameter has been
division. Generally, progressing from left to described using the special case of a perfect

INF. DES. J. 6/2 (1990) 147-59

149
ROBERT RANKIN • A TAXONOMY OF GRAPH TYPES

circle rather than some more complex or gener- prime element, and each discrete value of V
alised curve. represented by each trend surface within a
prime element.
Moving one position to the right in the fig-
D I S C U S S I O N OF THE FIGURES ure, the same information can be described
using prime elements of only two dimensions
(X, Y) by creating a two-dimensional space (Z,
Figure 1
T) of these elements. The fifth variable (V) is
As already mentioned, Figure 1 lists the scope described by a set of contour lines within each
of rectilinear cartesian co-ordinate graphs, prime element. With the mathematical func-
showing how they relate to each other for tion described in this form, data extraction is
mathematical functions of various complexi- straightforward and accurate. As a means for
ties, and how the same information may be dis- displaying the qualitative trends, however, it
played using illustrations of different forms. is much more limited than the previous case,
as the mathematical function now tends to
Row 1
be too dissected.
Taking the top row as an example, a five
In the third column, the prime element is
vari-able function can be displayed graphically
reduced to only two-dimensional single line
by producing a linear array (along one variable,
graphs (X, Y). To accommodate the same
T) of three-dimensional graphs (of three vari-
amount of information as the preceding case,
ables, X, Y and Z) each incorporating a set of
the array must become three-dimensional (T,
trend surfaces that describe the trends of the
Z, V). Being difficult to draw, and perhaps of
fifth variable (V). Each individual trend surface
dubious use, this example is not displayed
represents a single value of V.
here, but simply described verbally. It is antici-
A reader of such a graph would be able to ob-
pated that such a graphic would be of little use
serve changes in the dependent variable V with
both quantitatively or qualitatively. It could,
changes in the other variables. After selecting
however, be envisaged being spread over sever-
firstly, say, a value for T, then X, Y and finally
al pages of a book, with the dimensions of each
Z, a value for V can be read from the graph. A
page adopting the parameters T and Z and each
practical example of this would be the mea-
page representing a value of V. The thickness of
surement of, say, pollution levels (V) at a cer-
the book actually represents the V parameter.
tain position (X, Y) within a lake at a certain
Such an elaborate illustration would have uses
depth (Z) for varying amounts (T) of a pollutant
in organizing large amounts of inter-related
being added from an outfall.
data. The final illustration of the row requires
In practical terms, however, it is unlikely
the impossible situation of representing trend
that this complex graph would be used for ac-
surfaces of the variable Y in four-dimensional
curate data extraction. It is more likely to be
space (T, Z, V, X).
used qualitatively to show only rough trends
in the relationships between the variables. As Rows 2-4
well, both the variables T and V are not contin-
Subsequent rows in the figure are derived in
uously represented. Only several discrete val-
a similar manner, albeit with mathematical
ues of each are incorporated in the illustration,
functions of diminishing complexity. An inter-
each discrete value of T represented by each
esting feature develops in the second row as

INF. DES. J. 6/2 (1990) 147-59

150
ROBERT RANKIN ■ A TAXONOMY OF GRAPH TYPES

| NUMBER OF | 1
VARIABLES
GENERAL CASE 1-D REDUCTION 2-D REDUCTION 3-D REDUCTION Rectilinear
AND IN PRIME ELEMENT IN PRIME ELEMENT IN PRIME ELEMENT
FUNCTION Cartesian
coordinate
graphs

INF. DES. J. 6/2 (1990) 147-59

151
ROBERT RANKIN • A TAXONOMY OF GRAPH TYPES

2
Polar NUMBER OF
coordinate VARIABLES GENERAL CASE GENERAL CASE 1-D REDUCTION
graphs AND
FUNCTION (2 CYCLIC VARIABLES) (1 CYCLIC VARIABLE) IN PRIME ELEMENT

INF. DES. J. 6/2 (1990) 146-59

152
ROBERT RANKIN • A TAXONOMY OF GRAPH TYPES

the prime elements are reduced in dimensions.


2-D REDUCTION 3-D REDUCTION In the final illustration, the prime element has
IN PRIME ELEMENT IN PRIME ELEMENT been reduced to a single point value in a three-
dimensional space, and hence has reverted to
the same trend surface graph form as in the
original illustration of the first column.
The bottom row of the figure is fairly sparse
but is included here to complete the picture. As
THREE-DIMENSIONAL ARRAY (T,Z,V) THE IMPOSSIBLE SITUATION discussed previously, the fact that some cate-
OF POLAR GRAPHS (0,R) OF A 5-D TREND SURFACE GRAPH
gories in the overall matrix are poorly repre-
sented or even meaningless is an artifact of the
matrix system for generating categories.
Clearly, using this system it is possible to
generate illustrations which describe even
more variables than the maximum number of
five described here in the top row of the figure.
The top left-hand illustration uses only a linear
array of prime elements. It would be possible to
extend this to a three-dimensional array and
hence incorporate more variables, but for prac-
tical reasons such an illustration would be
difficult to construct and read.

Figure 2
The illustrations of Figure 2 are similar to
those of Figure 1 except that they describe
mathematical functions which incorporate
some cyclic variables. As examples, these
could include such parameters as compass
bearing (0 to 360 degrees), time of day or season
of the year. The distinguishing feature of these
variables is that their range of values is finite
(for example, 0 to 24 hours) and cyclic,
although it is possible for their values to con-
tinually increase if each completed cycle is
added to the next.
Because of this repetitive characteristic, it is
often desirable to plot data relating to this type
of parameter on a circular graph. These graphs
are said to utilize polar co-ordinates since the
values of the variables are measured with
respect to a central pole. The position of a point
on the graph is determined by the length of the
INF. DES. J. 6/2 (1990) 147-59

153
ROBERT RANKIN ■ A TAXONOMY OF GRAPH TYPES

radius vector from the pole and the angle this The final three illustrations describe the sit-
vector makes with a fixed line through the uation for one cyclic variable where the dimen-
pole. sions of the prime element are reduced by one,
Again, functions with a maximum of five two and three dimensions respectively. These
variables (up to two of which are cyclic) are illustrations could describe the two cyclic vari-
represented. It is possible to conceptualize the able case if either T, V or Z was assumed to be
display of functions with a greater number of cyclic. However, this cyclic property would not
variables, but in practical terms these serve be visually evident as the cyclic variable would
little use. They are too complex to reproduce be plotted along a linear axis. Quantitatively
on the flat page, hence of limited use in most little is lost in doing this, but qualitatively, it
communication modes. An example of a five could be anticipated the cyclic trends may be
variable function with two parameters cyclic more difficult to comprehend.
would be the measurement of the temperature The third illustration is formed by simply
(V) at a certain depth (R) at a certain position removing the Z variable from the prime ele-
along a river (Z) at a certain time of day (0) at ment and incorporating it as one of the major
a certain time of the year ( ) . illustration dimensions. The fourth illustra-
tion is derived in a similar way by removing
.Row 1 the V variable from the prime element and cre-
The top row displays five-variable functions in ating a three dimensional main illustration
which one or two variables are cyclic. In the composed of an array of two dimensional polar
first illustration, the first cyclic variable (<>
| ) is graphs. This would be a complex and confusing
measured as the angle with the horizontal illustration and no attempt has been made to
made by the position on the circular array of represent it here. The final illustration requires
prime elements. Similarly, the second cyclic the impossible situation of representing trend
variable (0) defines the angular position within surfaces of the variable (0) in a four-dimension-
each prime element, while the variable R mea- al space (T, Z, V, R).
sures the distance out from the centre of the
prime element in the direction of (0). The vari- Rows 2-4
able Z indicates the position along the Z-axis, Subsequent rows in Figure 2 are derived in a
while the variable V is represented by concen- similar manner to those of the first row with
tric cylinder-like trend surfaces. V, together mathematical functions of diminishing com-
with the cyclic variable (), are not represented plexity. An interesting situation develops in
continuously. Only discrete values are repre- the second row as the number of variables rep-
sented, one for each prime element in the case resented in the prime element are reduced to
of ( ) and one for each trend surface in the case one. Here the prime element collapses to a
of V. point representing one value of the fourth
With the reduction of one cyclic variable, parameter in the three-dimensional space of
the circular array (parameter ) of the first il- the illustration created by the remaining three
lustration is reduced to a linear array (T param- parameters. In actual fact, the illustration has
eter) in the second illustration. This does not reverted to that shown in the second column
necessarily imply that the variable T cannot without the cyclic parameter being represented
be cyclic. In some circumstances it may be as such.
desirable to present a cyclic variable in a linear
fashion.
INF. DES. J. 6/2 (1990) 147-59

154
ROBERT RANKIN ■ A TAXONOMY OF GRAPH TYPES

Figure 3 .Rows 4 and 5


The illustrations of the top three rows of Figure The lower section of Figure 3 (Rows 4 and 5)
3 relate directly to those of Figure 1 except that describes similar quantitative information for
it is assumed the data plotted is not a continu- cases where the mathematical function
ous mathematical function but simply a num- includes some cyclic variables. The treatment
ber of data points. is similar to Rows 3 and 4 in Figure 2.

Row 1
For presentation purposes, it makes sense to Figure 4
describe a maximum of four variables in the The illustrations in Figure 4 are produced by
one illustration with a prime element of three matrixing the number of variables and number
dimensions. This is shown in the first illustra- of independent variables of the mathematical
tion of the top row. These conditions relate to function (vertical dimension) with an assort-
the first illustration of the second row of Figure ment of conditions (horizontal dimension).
1, although in this earlier case, the variation The first and second columns incorporate illus-
with the variable T is compressed onto a single trations using prime elements of three and two
prime element by superimposing trend sur- dimensions respectively, while the third col-
faces. In Figure 3, one trend surface is described umn lists special cases where some of the
in each prime element, but since we are dealing parameters are constrained by special condi-
with discrete data, bars are drawn instead of tions, namely a constant summation.
the surface. In actual fact, these bars are simply There is overlap between some illustrations
the 'posts' which would hold up the trend sur- in Figures 1 and 4. This is desirable as it shows
face which could be draped over them, interpo- where one system of classification deviates
lating the mathematical function in the spaces from another due to differing horizontal
between them. dimensions in each figure. A top row for cases
The third column shows the result of reduc- of four variables with three independent vari-
ing the prime element dimensions by one. An ables is not included as it is represented in
array of two dimensional bar graphs results Figure 1 and presents nothing new here.
which gives direct access to quantitative infor-
Row 1
mation displayed in it.
The first row introduces two new types of
The fourth column describes the data using
illustration. The transformation grid or two-
one dimension in each prime element that is
dimensional conversion scale converts two
not spatially related. This dimension is simply
independent variables into two others. The tri-
a list of values of the fourth variable laid out in
linear chart presents a method of representing
the space created by the two remaining dimen-
sions of the prime element. For the extraction four parameters where three sum to a constant,
of actual data, this design is accurate, but quali- which is the case when dealing with, for exam-
tatively only poorly displays the trends in the ple, mixtures of three substances in varying
mathematical function. proportions. These proportions influence the
value of the fourth parameter which, if quanti-
Rows 2 and 3 tative, can be plotted as a trend surface on a
Rows 2 and 3 show the illustration of functions three-dimensional graph, or as contours on
with less variables, namely 3 and 2 respectively. (or more simply, as areas within) a two-

INF. DES. J. 6/2 (1990) 147-59

155
ROBERT RANKIN ■ A TAXONOMY OF GRAPH TYPES

3
Cartesian and
polar graphs
for discrete
data

INF. DES. J. 6/2 (1990) 147-59

156
ROBERT RANKIN ■ A TAXONOMY OF GRAPH TYPES

4
NUMBER OF
Graphs for VARIABLES
special cases AND 3-D GENERAL CASE 2-D GENERAL CASE 1,2 or 3-D SPECIAL CASE
and conditions FUNCTION

(INDEP=
independent
variable)

INF. DES. J. 6/2 (1990) 147-59

157
ROBERT RANKIN ■ A TAXONOMY OF GRAPH TYPES

dimensional graph. This latter approach is mathematical functions of varying complexity


often used if the fourth parameter can only or number of parameters.
be measured qualitatively. Some of the graphs displayed here are visu-
ally quite complex and may not necessarily
Rows 2 and 3 lend themselves to useful presentation or con-
The first illustrations of both the first and struction on the printed page or static video
second rows show the case where a trend sur- screen. However, with the availability of
face is reduced to a line when the indepen- sophisticated computer graphics, these com-
dence conditions become more restricted. In plex illustrations may be better utilized within
the third row an alignment nomogram is repre- the computer environment.
sented. Not strictly a spatial presentation of The figures would be of great use during the
information, the nomogram is more a compu- design stage of a scientific document. An
tational tool for calculating the values of inter- author, having data or a mathematical concept
related variables. to illustrate, can choose what he or she consid-
ers to be the most appropriate graphic form
Rows 4 and 5 from those available in the four figures. The
Finally, the third column of the last two rows choice will depend on the intended use of the
represent functions of three and two variables illustration, be it for efficient access to data, to
with the special additive condition. In the describe a mathematical concept, or one of
three parameter case, the trilinear chart has many other graphical functions.
collapsed to an inverted conversion scale repre- What the classification does not do, of
senting one axis of a two dimensional line course, is provide guidelines for this choice,
graph displaying the third parameter. In the especially from a communication viewpoint.
two dimensional case, only the inverted con- In the communication environment, there
version scale remains. needs to be a balance between the most appro-
priate illustration to accommodate the data,
and the most appropriate illustration to convey
CONCLUSIONS the intended message to the anticipated audi-
ence. Decisions such as these lie in the domain
The four figures bring together a wide range of of the skilled communication practitioner who
commonly used types of graph by using a rea- is relying on experience, or guidance from com-
sonably exhaustive matrix method. As such, munication models for choosing the most suit-
they maybe useful as a resource for intending able illustration.
graph users. As well, they demonstrate the
inter-relationships among the graphic forms.
Generally, within one figure, the proximity of
one graphic form to another is a measure of
their similarity. Horizontally, across one row
within a figure, the matrices generally display
alternative methods for presenting the same
mathematical function. Vertically, the
columns display similar graphic forms for

INF. DES. J. 6/2 (1990) 147-59

158
ROBERT RANKIN • A TAXONOMY OF GRAPH TYPES

REFERENCES

Doblin J (1979) Rankin R (1989)


A structure for non-textual The development of an
communications. In Kolers P illustration design model
Wrolstad M and Bourna H Educational Technology
(eds) Processing of visible Research and Development
language Vol 2 37 (2): 25-46
New York: Plenum Press
Tufte E R (1983)
Duchastel P C (1978) The visual display of
Illustrating instructional quantitative information
texts Educational Technology Cheshire, Connecticut:
November: 36-39 Graphics Press

Duchastel P and Waller R Twyman M (1979)


(1979) A schema for the study of
Pictorial illustration in graphic language. In Kolers
instructional texts, P Wrolstad M and Bourna H
Educational Technology 19 (eds) Processing of visible
(11): 20-25 language Vol 1 New York:
Fleming M (1967) Plenum Press
Classification and analysis
of instructional illustration, Twyman M (1982)
AV Communication Review The graphic presentation of
15 (3): 246-256 language Information Design
Journal 3 (1): 2-22
Knowlton J Q (1966)
On the definition of picture Waller R (1979)
AV Communication Review Four aspects of graphic
14: 157-183 communication: an
introduction to the issue. In
MacDonald-Ross M (1977a) Jonassen D (ed.) The
Graphics in text Review of technology of text
Research in Education New Jersey: Educational
5: 49-85 Technology Publications

MacDonald-Ross M (1977b)
How numbers are shown
AV Communication Review
25: 359-409

Rankin R (1986)
Communicating scientific
concepts through the use of
graphs and diagrams
Unpublished doctoral thesis,
Griffith University, Brisbane

INF. DES. J. 6/2 (1990) 147-59

159