Statistics

The R Series

Graphical Data Analysis with R

Graphical Data Analysis with R shows you what information you
can gain from graphical displays. The book focuses on why you draw
graphics to display data and which graphics to draw (and uses R to
do so). All the datasets are available in R or one of its packages and
the R code is available at rosuda.org/GDA.
Graphical Data
Graphical data analysis is useful for data cleaning, exploring data
structure, detecting outliers and unusual groups, identifying trends
and clusters, spotting local patterns, evaluating modelling output,
Analysis with R
and presenting results. This book guides you in choosing graphics
and understanding what information you can glean from them. It
can be used as a primary text in a graphical data analysis course
or as a supplement in a statistics course. Colour graphics are used
throughout.
Features
• Concentrates on why graphics are drawn and what they reveal
• Emphasises the value of drawing a variety of different graphics
• Uses real datasets in R to show how graphical data analysis
works in practice
• Supplies R code for all the graphics on the author’s website
• Includes a set of exercises in each chapter to facilitate hands-on
learning
Antony Unwin is a professor of computer-oriented statistics and data
analysis at the University of Augsburg. He is a fellow of the American
Statistical Society. His research focuses on data visualisation,
especially in interactive graphics.
Unwin

Antony Unwin

K25332

w w w. c rc p r e s s . c o m

K25332_cover.indd 1 3/4/15 3:38 PM

 Graphical Data
Analysis with R

K25332_FM.indd 1 3/4/15 4:04 AM

 Chapman & Hall/CRC
The R Series

Series Editors
John M. Chambers Torsten Hothorn
Department of Statistics Division of Biostatistics
Stanford University University of Zurich
Stanford, California, USA Switzerland

Duncan Temple Lang Hadley Wickham

Department of Statistics RStudio
University of California, Davis Boston, Massachusetts, USA
Davis, California, USA

Aims and Scope

This book series reflects the recent rapid growth in the development and application
of R, the programming language and software environment for statistical computing
and graphics. R is now widely used in academic research, education, and industry.
It is constantly growing, with new versions of the core software released regularly
and more than 6,000 packages available. It is difficult for the documentation to
keep pace with the expansion of the software, and this vital book series provides a
forum for the publication of books covering many aspects of the development and
application of R.
The scope of the series is wide, covering three main threads:
• Applications of R to specific disciplines such as biology, epidemiology,
genetics, engineering, finance, and the social sciences.
• Using R for the study of topics of statistical methodology, such as linear and
mixed modeling, time series, Bayesian methods, and missing data.
• The development of R, including programming, building packages, and
graphics.
The books will appeal to programmers and developers of R software, as well as
applied statisticians and data analysts in many fields. The books will feature
detailed worked examples and R code fully integrated into the text, ensuring their
usefulness to researchers, practitioners and students.

K25332_FM.indd 2 3/4/15 4:04 AM

 Published Titles

Stated Preference Methods Using R, Hideo Aizaki, Tomoaki Nakatani,

and Kazuo Sato
Using R for Numerical Analysis in Science and Engineering, Victor A. Bloomfield
Event History Analysis with R, Göran Broström
Computational Actuarial Science with R, Arthur Charpentier
Statistical Computing in C++ and R, Randall L. Eubank and Ana Kupresanin
Reproducible Research with R and RStudio, Christopher Gandrud
Introduction to Scientific Programming and Simulation Using R, Second Edition,
Owen Jones, Robert Maillardet, and Andrew Robinson
Nonparametric Statistical Methods Using R, John Kloke and Joseph McKean
Displaying Time Series, Spatial, and Space-Time Data with R,
Oscar Perpiñán Lamigueiro
Programming Graphical User Interfaces with R, Michael F. Lawrence
and John Verzani
Analyzing Sensory Data with R, Sébastien Lê and Theirry Worch
Parallel Computing for Data Science: With Examples in R, C++ and CUDA,
Norman Matloff
Analyzing Baseball Data with R, Max Marchi and Jim Albert
Growth Curve Analysis and Visualization Using R, Daniel Mirman
R Graphics, Second Edition, Paul Murrell
Data Science in R: A Case Studies Approach to Computational Reasoning and
Problem Solving, Deborah Nolan and Duncan Temple Lang
Multiple Factor Analysis by Example Using R, Jérôme Pagès
Customer and Business Analytics: Applied Data Mining for Business Decision
Making Using R, Daniel S. Putler and Robert E. Krider
Implementing Reproducible Research, Victoria Stodden, Friedrich Leisch,
and Roger D. Peng
Graphical Data Analysis with R, Antony Unwin
Using R for Introductory Statistics, Second Edition, John Verzani

Advanced R, Hadley Wickham

Dynamic Documents with R and knitr, Yihui Xie

K25332_FM.indd 3 3/4/15 4:04 AM

K25332_FM.indd 4 3/4/15 4:04 AM
 Graphical Data
Analysis with R

Antony Unwin
University of Augsburg
Germany

K25332_FM.indd 5 3/4/15 4:04 AM

CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742
© 2015 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government works

Version Date: 20150224

International Standard Book Number-13: 978-1-4987-1524-9 (eBook - PDF)

This book contains information obtained from authentic and highly regarded sources. Reasonable
efforts have been made to publish reliable data and information, but the author and publisher cannot
assume responsibility for the validity of all materials or the consequences of their use. The authors and
publishers have attempted to trace the copyright holders of all material reproduced in this publication
and apologize to copyright holders if permission to publish in this form has not been obtained. If any
copyright material has not been acknowledged please write and let us know so we may rectify in any
future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted, reproduced,
transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or
hereafter invented, including photocopying, microfilming, and recording, or in any information stor-
age or retrieval system, without written permission from the publishers.

For permission to photocopy or use material electronically from this work, please access www.copy-
right.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC), 222
Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that pro-
vides licenses and registration for a variety of users. For organizations that have been granted a photo-
copy license by the CCC, a separate system of payment has been arranged.

Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identification and explanation without intent to infringe.
Visit the Taylor & Francis Web site at
http://www.taylorandfrancis.com
and the CRC Press Web site at
http://www.crcpress.com
Contents

Preface xi

1 Setting the Scene 1

1.1 Graphics in action . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 What is Graphical Data Analysis (GDA)? . . . . . . . . . . . . . . 5
1.4 Using this book, the R code in it, and the book’s webpage . . . . . 14

2 Brief Review of the Literature and Background Materials 19

2.1 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.2 Interactive graphics . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 Other graphics software . . . . . . . . . . . . . . . . . . . . . . . 21
2.4 Websites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.5 Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.6 Statistical texts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3 Examining Continuous Variables 27

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 What features might continuous variables have? . . . . . . . . . . . 29
3.3 Looking for features . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4 Comparing distributions by subgroups . . . . . . . . . . . . . . . . 44
3.5 What plots are there for individual continuous variables? . . . . . . 46
3.6 Plot options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.7 Modelling and testing for continuous variables . . . . . . . . . . . 48

4 Displaying Categorical Data 53

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.2 What features might categorical variables have? . . . . . . . . . . . 56
4.3 Nominal data—no fixed category order . . . . . . . . . . . . . . . 57
4.4 Ordinal data—fixed category order . . . . . . . . . . . . . . . . . 62
4.5 Discrete data—counts and integers . . . . . . . . . . . . . . . . . 66
4.6 Formats, factors, estimates, and barcharts . . . . . . . . . . . . . . 70
4.7 Modelling and testing for categorical variables . . . . . . . . . . . 71

vii
viii

5 Looking for Structure: Dependency Relationships and Associations 75

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.2 What features might be visible in scatterplots? . . . . . . . . . . . 77
5.3 Looking at pairs of continuous variables . . . . . . . . . . . . . . . 78
5.4 Adding models: lines and smooths . . . . . . . . . . . . . . . . . . 83
5.5 Comparing groups within scatterplots . . . . . . . . . . . . . . . . 86
5.6 Scatterplot matrices for looking at many pairs of variables . . . . . 88
5.7 Scatterplot options . . . . . . . . . . . . . . . . . . . . . . . . . . 92
5.8 Modelling and testing for relationships between variables . . . . . 94

6 Investigating Multivariate Continuous Data 99

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.2 What is a parallel coordinate plot (pcp)? . . . . . . . . . . . . . . . 100
6.3 Features you can see with parallel coordinate plots . . . . . . . . . 102
6.4 Interpreting clustering results . . . . . . . . . . . . . . . . . . . . 106
6.5 Parallel coordinate plots and time series . . . . . . . . . . . . . . . 108
6.6 Parallel coordinate plots for indices . . . . . . . . . . . . . . . . . 112
6.7 Options for parallel coordinate plots . . . . . . . . . . . . . . . . . 115
6.8 Modelling and testing for multivariate continuous data . . . . . . . 127
6.9 Parallel coordinate plots and comparing model results . . . . . . . 127

7 Studying Multivariate Categorical Data 131

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

7.2 Data on the sinking of the Titanic . . . . . . . . . . . . . . . . . . 132
7.3 What is a mosaicplot? . . . . . . . . . . . . . . . . . . . . . . . . 133
7.4 Different mosaicplots for different questions of interest . . . . . . . 140
7.5 Which mosaicplot is the right one? . . . . . . . . . . . . . . . . . 148
7.6 Additional options . . . . . . . . . . . . . . . . . . . . . . . . . . 149
7.7 Modelling and testing for multivariate categorical data . . . . . . . 151

8 Getting an Overview 155

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

8.2 Many individual displays . . . . . . . . . . . . . . . . . . . . . . . 159
8.3 Multivariate overviews . . . . . . . . . . . . . . . . . . . . . . . . 162
8.4 Multivariate overviews for categorical variables . . . . . . . . . . . 168
8.5 Graphics by group . . . . . . . . . . . . . . . . . . . . . . . . . . 170
8.6 Modelling and testing for overviews . . . . . . . . . . . . . . . . . 174
 ix

9 Graphics and Data Quality: How Good Are the Data? 177

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

9.2 Missing values . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
9.3 Outliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
9.4 Modelling and testing for data quality . . . . . . . . . . . . . . . . 193

10 Comparisons, Comparisons, Comparisons 197

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197

10.2 Making comparisons . . . . . . . . . . . . . . . . . . . . . . . . . 199
10.3 Making visual comparisons . . . . . . . . . . . . . . . . . . . . . 202
10.4 Comparing group effects graphically . . . . . . . . . . . . . . . . 208
10.5 Comparing rates visually . . . . . . . . . . . . . . . . . . . . . . . 212
10.6 Graphics for comparing many subsets . . . . . . . . . . . . . . . . 214
10.7 Graphics principles for comparisons . . . . . . . . . . . . . . . . . 216
10.8 Modelling and testing for comparisons . . . . . . . . . . . . . . . 218

11 Graphics for Time Series 223

11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223

11.2 Graphics for a single time series . . . . . . . . . . . . . . . . . . . 224
11.3 Multiple series . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
11.4 Special features of time series . . . . . . . . . . . . . . . . . . . . 233
11.5 Alternative graphics for time series . . . . . . . . . . . . . . . . . 237
11.6 R classes and packages for time series . . . . . . . . . . . . . . . . 238
11.7 Modelling and testing time series . . . . . . . . . . . . . . . . . . 238

12 Ensemble Graphics and Case Studies 243

12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243

12.2 What is an ensemble of graphics? . . . . . . . . . . . . . . . . . . 246
12.3 Combining different views—a case study example . . . . . . . . . 248
12.4 Case studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252

13 Some Notes on Graphics with R 257

13.1 Graphics systems in R . . . . . . . . . . . . . . . . . . . . . . . . 257

13.2 Loading datasets and packages for graphical analysis . . . . . . . . 258
13.3 Graphics conventions in statistics . . . . . . . . . . . . . . . . . . 258
13.4 What is a graphic anyway? . . . . . . . . . . . . . . . . . . . . . . 259
13.5 Options for all graphics . . . . . . . . . . . . . . . . . . . . . . . 261
13.6 Some R graphics advice and coding tips . . . . . . . . . . . . . . . 264
13.7 Other graphics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
13.8 Large datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
13.9 Perfecting graphics . . . . . . . . . . . . . . . . . . . . . . . . . . 273
x

14 Summary 275

14.1 Data analysis and graphics . . . . . . . . . . . . . . . . . . . . . . 275

14.2 Key features of GDA . . . . . . . . . . . . . . . . . . . . . . . . . 276
14.3 Strengths and weaknesses of GDA . . . . . . . . . . . . . . . . . . 276
14.4 Recommendations for GDA . . . . . . . . . . . . . . . . . . . . . 277

References 279
General index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
Datasets index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
Preface

Graphical Data Analysis is useful for data cleaning, exploring data structure, de-
tecting outliers and unusual groups, identifying trends and clusters, spotting local
patterns, evaluating modelling output, and presenting results. It is essential for ex-
ploratory data analysis and data mining. There are several fine books on graphics us-
ing R, such as “ggplot2” [Wickham, 2009], “Lattice” [Sarkar, 2008], and “R Graph-
ics” [Murrell, 2011]). These books concentrate on how you draw graphics in R. This
book concentrates on why you draw graphics and which graphics to draw (and uses
R to do so).
The target readership includes anyone carrying out data analyses who wants to
understand their data using graphics. The book can be used as the primary textbook
for a course in Graphical Data Analysis or as an accompanying text for a statistics
course. Prerequisites for the book are an interest in data analysis and some basic
knowledge of R.
The main aim of the book is to show, using real datasets, what information graph-
ical displays can reveal in data. Seeing graphics in action is the best way to learn
Graphical Data Analysis. Gaining experience in interpreting graphics and drawing
your own data displays is the most effective way forward.
The graphics shown in the book are a starting point. Sometimes more graph-
ics could have been drawn, and alternative graphics could always have been drawn.
Readers may have their own ideas of how best to present certain features of the
datasets. Although each graphic reveals information contained in its dataset, it is
likely that in every case there is more to be discovered. It is certainly one of the aims
of each analysis to find out as much as possible about the data. The graphics are not
drawn for their own sake, they are drawn to reveal and convey information.
A central idea underlying this book is that many graphics should be drawn. The
aim should not have to be to draw a single graphic that summarises everything that
can be said about the data. That is too difficult, if not impossible. The aim is to find
a number of graphics, maybe even a large number of them, where each contributes
something to the overall picture. Just as many photographs of the same object taken
from different angles in different lights make it easier for us to grasp a whole object,
datasets should be visualised in many different ways.
The emphasis is on exploring datasets first and on presenting results second.
Graphical Data Analysis is about using graphics to find results. One way to think
about this is to imagine you are looking at a new package in R and it uses a dataset
you are not familiar with for the examples in the help. What does the dataset look
like? How would you go about finding out what features it has, and how that might
affect the use of the methods in the package? What information can you find graphi-

xi
xii Graphical Data Analysis with R

cally in the data that a modelling approach should also find? What graphical displays
are there that help you understand the results of other people’s models, such as the
examples given on the help page? This presupposes an active interest on the part
of the reader. Roland Barthes, the French structuralist, referred to readerly texts and
writerly texts. In a writerly text the reader takes an active role in the construction of
meaning. I hope the readers of this book will take an active role in thinking about
what graphics show, what information can be gleaned from them, and why they were
chosen.
As every dataset used is available in R or one of its packages, information about
them can usually be found on the relevant help page, including which variables of
what types are involved and how big the dataset is. Ideally there should be a descrip-
tion of why and how it was collected, with references to original sources. Context is
important for interpreting results and you have to know your dataset and its prove-
nance. A well-developed sense of curiosity is very helpful in data analysis.
Graphical Data Analysis is an attractive way of working with data. It encourages
you to look at many different aspects and to investigate in many different directions.
You can be surprised by what you uncover and even by which graphic turns out to be
most effective in revealing information. Your results are easy to show to others and
are easy to discuss with others.
For any result found graphically, we should try to check what statistical sup-
port there is for it, just as we use graphics to review the results of our statistical
modelling. Graphical Data Analysis and more traditional statistical approaches com-
plement each other very well and we should take advantage of this.

Acknowledgements
No book on R should omit thanking Robert Gentleman, Ross Ihaka, and all the
many R contributors. They have made analysis of data much easier for the rest of
us. Thanks also to Hadley Wickham for all his R packages (sometimes referred to as
the Hadleyverse), especially for ggplot2, and to Yihui Xie for knitr, a major help in
keeping this book in order. Particular thanks are due to Bill Venables for his words
of wisdom and for R advice and code. If any of the book’s code looks elegant, then
it must be Bill’s, and if it looks clumsy, it is certainly mine.
Dennis Freuer, Urs Freund, Katrin Grimm, Harold Henderson, Ross Ihaka, Kary
Myers, Alexander Pilhöfer, Maryann Pirie, Friedrich Pukelsheim, Christina Sanchez,
Günther Sawitzki, Rolf Turner, Chris Wild, and Aisen Yang read one or more chap-
ters and made many helpful suggestions for improvement, some of which I have been
able to adopt. John Kimmel was an encouraging and efficient publisher, who organ-
ised several constructively critical reviewers, including Di Cook, Michael Friendly,
and Ramnath Vaidyanathan. I would also like to thank the Statistics Department at
the University of Auckland for a stimulating and sociable environment in which to
work on this book during my sabbatical.
Preface xiii

Finally I would like to thank my family for never asking me when the book would
be finished and for many other kindnesses.

Augsburg, December 2014 Antony Unwin

1
Setting the Scene

‘What is the use of a book’, thought Alice, without pictures or conversations?’

Lewis Carroll

1.1 Graphics in action

Female

0
Male

0
160 180 200 220
Speed (km/hr)

FIGURE 1.1: Histograms of speeds reached at the 2011 World Speed Skiing Champi-
onships. Source: www.fis-ski.com. There were more male competitors than females,
yet the fastest group of females were almost as fast as the fastest group of males. The
female competitors were all either fast or (relatively) slow—or were they?

1
2 Graphical Data Analysis with R

The code for producing Figure 1.1 is:

library(ggplot2); library(ggthemes)
data(SpeedSki, package = "GDAdata")
ggplot(SpeedSki, aes(x=Speed, fill=Sex)) + xlim(160, 220) +
geom_histogram(binwidth=2.5) + xlab("Speed (km/hr)") +
facet_wrap(~Sex, ncol=1) + ylab("") +
theme(legend.position="none")

The 2011 World Speed Skiing Championships were held at Verbier in Switzer-
land. Figure 1.1 shows histograms of the speeds reached by the 12 female and 79
male competitors. As well as emphasising that there were many more competitors in
the men’s competition than in the women’s, the plots show that the fastest person was
a man and that a woman was slowest. What is surprising (and more interesting) is
that the fastest women were almost as fast as the fastest men and that there were two
distinct groups of women, the fast ones and the slow ones. There also appear to be
two groups of men, although the gap between them is not so large. All of this infor-
mation is easy to see in the plots and would not be readily apparent from statistical
summaries of the data.
A little more investigation reveals the reason for the groupings: There are actu-
ally three different events, Speed One, Speed Downhill, and Speed Downhill Junior.
Figure 1.2 shows the histograms of speed by event and gender. We can see that Speed
One is the fastest event (competitors have special equipment), that no women took
part in the Downhill, and that there was little variation in speed amongst the Juniors.
The reason for the two female groups is now clear: They took part in two different
events. The distribution of the men’s speeds is affected by the inclusion of speeds for
the Downhill event and by the greater numbers of men who competed. It is interesting
that there is little variation in speed amongst the 7 women who competed in the Speed
One event, compared to that of the 39 men who took part. The women were faster
than most of the men.
The code for the plots takes a little getting used to. On the one hand the informa-
tion would still have been visible with less coding, although perhaps not so clearly.
Setting sensible scale limits, specifying meaningful binwidths, and aligning graphics
whose distributions you want to compare one above the other with the same size and
scales all help.
On the other hand, the plots might have benefitted from more coding to make
them look better: adding a title, choosing different colours, or specifying different
tick marks and labelling. That is more a matter of taste. This book is about data anal-
ysis, primarily exploratory analysis, rather than presentation, so the amount of coding
is reduced. Sometimes defaults are removed (like the legends) to reduce unnecessary
clutter.
The Speed Skiing example illustrates a number of issues that will recur through-
out the book. Graphics are effective ways of summarising and conveying informa-
tion. You need to think carefully about how to interpret a graphic. Context is im-
portant and you often have to gather additional background information. Drawing
several graphics is a lot better than just drawing one.
Setting the Scene 3

ggplot(SpeedSki, aes(Speed, fill=Sex)) +

geom_histogram(binwidth=2.5) + xlab("Speed (km/hr)") +
ylab("") + facet_grid(Sex~Event) +
theme(legend.position="none")

Speed Downhill Speed Downhill Junior Speed One

8

Female
4

0
8

Male
4

0
160 180 200 160 180 200 160 180 200
Speed (km/hr)

FIGURE 1.2: Histograms of speeds in the 2011 World Speed Skiing Championships
by event and gender. There was no Speed Downhill for women. The few women
taking part in the fastest event, Speed One, did very well, beating most of the men.

1.2 Introduction
There is no complex theory about graphics. In fact there is not much theory at all, and
so the topic is not covered in depth in books or lectures. Once the various graphics
forms have been described, the textbooks can pass on to supposedly more difficult
topics such as proving the central limit theorem or the asymptotic normality of max-
imum likelihood estimates.
The evidence of how graphics are used in practice suggests that they need more
attention than a cursory introduction backed up by a few examples. If we do not
have a theory which can be passed on to others about how to design and interpret
informative graphics, then we need to help them develop the necessary skills using a
range of instructive examples. It is surprising (and sometimes shocking) how casually
graphics may be employed, more as decoration than as information, more for reasons
of routine than for reasons of communication.
It is worthwhile, as always, to check what the justly famous John Tukey has to
say. In his paper [Tukey, 1993] he summarised what he described as the true purpose
of graphic display in four statements:

1. Graphics are for the qualitative/descriptive—conceivably the semiquantitative—

never for the carefully quantitative (tables do that better).
4 Graphical Data Analysis with R

2. Graphics are for comparison—comparison of one kind or another—not for ac-

cess to individual amounts.
3. Graphics are for impact—-interocular impact if possible, swinging-finger impact
if that is the best one can do, or impact for the unexpected as a minimum—-but
almost never for something that has to be worked at hard to be perceived.
4. Finally, graphics should report the results of careful data analysis—-rather than
be an attempt to replace it. (Exploration—to guide data analysis—-can make
essential interim use of graphics, but unless we are describing the exploration
process rather than its results, the final graphic should build on the data analysis
rather than the reverse.)
These are all excellent points, although the last one implies he is emphasising graph-
ics displays for presentation. It is notable that Tukey writes of ‘the final graphic’ as
if there might only be one. In practice he commonly used more than one.
Data analysis is a complex topic. Conclusions are seldom clear-cut, and there are
often several alternative competing explanations. Graphics are part of the data analy-
sis process; both the choice of graphics displays and how they are viewed can have an
important influence on any conclusions drawn. The subject matter of the study will
also affect how any graphics are interpreted. A positively correlated collection of
points in a scatterplot may be taken as evidence of a useful association (e.g., income
as a function of years of experience) or as evidence of insufficient agreement (e.g.,
where two methods of measuring the same quantity are compared). In the first case
an outlier may be unusual, but nevertheless perfectly plausible, while in the second
it may be a clear indication of a faulty measurement.
To appreciate how much can be revealed by even the simplest graphics, it is useful
to think of all the different forms a graphic might take and what they might tell you
about the data. Consider a barchart showing the frequencies of the three categories
of a variable:

• The bars could all be the same height (as you might expect in a scientific study
with three groups).

• The bars might have slightly different heights (possibly suggesting some missing
values in a scientific study).
• One of the bars might be very small, suggesting that that category is either rare
(a particular illness perhaps) or not particularly relevant (support for a minor
political party).
• The bars might not follow an anticipated pattern (sales in different regions or the
numbers of people with various qualifications applying for a job).
• ...

There is literally no limit to the number of possibilities once you take into account
the different settings the data may have come from. This means that you need to gain
Setting the Scene 5

experience in looking at graphics to learn to appreciate what they can and cannot
show.
As with all statistical investigations it is not only necessary to identify potential
conclusions, there has to be enough evidence to support the conclusions. Tradition-
ally this has meant carrying out statistical tests. Unfortunately there are distinct limits
to testing. A lot of insights cannot easily be directly tested (Does that outlying clus-
ter of points really form a distinctive group? Is that distribution bimodal?) and even
those that can be require restrictive assumptions for the tests to be valid. Additionally
there is the issue of multiple testing. None of this should inhibit us from testing when
we can, and occasionally a visually tentative result can be shown to have such a con-
vincingly small p-value that no amount of concerns about assumptions can cast much
doubt on the result. The interplay of graphics with testing and modelling is effective
because the two approaches complement each other so well. The only downside is
that while it is usually feasible to find a graphic which tells you something about the
results of a test, it is not always possible to find a test which can help you assess a
feature you have discovered in a graphic.

1.3 What is Graphical Data Analysis (GDA)?

It is simplest to see what GDA can do by looking at a few examples. These exam-
ples are all just initial looks at the datasets to give the flavour of GDA, not complete
analyses. In each case we will draw graphical displays of a dataset to reveal some
of the information in the data. Graphics are good for showing structure and for com-
municating results. They are generally easier to interpret than tables (which are good
for providing exact values) or statistical reports (which are good for giving estimates
and formal comparisons) and convey more qualitative information.
For each example more than one graphic has been drawn. In general it is always
better to draw many graphics, offering many different views, to ensure you get as
much information out of a dataset as you can. This is part of the open-ended nature
of GDA; it is a process, in which you pursue multiple ideas in parallel, just as any
investigative process should be. As a result, a graphical data analysis of these exam-
ples in practice would involve drawing a lot more graphics, checking to see if there
are other features of interest, comparing different versions of various graphics to see
which ones work best, and finally settling on a group of graphics to summarise the
analysis.
Drawing graphical displays of data is not about selecting the one best graphic,
it is about selecting the best set of graphics. In the past drawing graphics required
extensive effort and producing one graphic by hand took a lot of time, so it made
sense to draw only a few. Nowadays we can draw graphics very quickly, many of
the basic options are sensibly chosen by software defaults, and the overall quality
is extremely high. It is still a good idea to be selective about which graphics you
keep and especially about which graphics you show to others, but the main aim is
6 Graphical Data Analysis with R

to uncover information and there is every reason to draw more graphics rather than
fewer when doing GDA.
With presentation graphics you prepare one graphic for many potential viewers.
You need experience in deciding which graphic to present and expertise in how to
draw it well. With GDA you prepare many graphics for one viewer, yourself, and
your aim is to uncover the information hidden in the data. You need expertise in
choosing a set of informative graphics and experience in interpreting graphics.

The Iris dataset

The dataset contains information on three species of iris. There are 50 plants for
each species and measurements in cms for four attributes of each of the 150 plants.
Originally the dataset was used by Fisher to illustrate linear discriminant analysis
[Fisher, 1936]. (It has been used for all manner of other analyses since.) Figure 1.3
shows a default histogram of petal length, one of the four attributes. There appear to
be two clearly different groups in the dataset.

ggplot(iris, aes(Petal.Length)) + geom_histogram()

20
count

10

0
2 4 6
Petal.Length

FIGURE 1.3: A histogram of petal lengths from Fisher’s iris dataset. The data divide
into two distinct groups.

We can look at a plot of the two petal attributes together, petal length and petal
width. Figure 1.4 shows that there is a very strong relationship between these two
attributes, providing further convincing evidence of at least two distinct groups of
flowers. The colouring by species shows that the lower group are all setosa, that the
upper group is made up of both versicolor and virginica flowers, and that these two
groups are moderately well separated by their petal measurements.
The iris dataset is so well known that many readers will be familiar with this in-
formation. Imagine, however, that you wanted to present this information to someone
who did not already know it. Are there better ways than simple graphics?
Setting the Scene 7

library(ggthemes)
ggplot(iris, aes(Petal.Length, Petal.Width, color=Species)) +
geom_point() + theme(legend.position="bottom") +
scale_colour_colorblind()

2.5 ! !!

! !

!!!! ! ! ! !

! ! !

!!!! ! !

2.0 !!!! ! !

!! ! !

!! ! !! ! ! !

! !

! ! ! !

1.5
Petal.Width

! !!! !!!

! ! !!! !

! !!!!!!!

!! ! ! !

! !!

1.0 ! ! ! !!

0.5 !

! !!! !

!!! !

! !!!!!! !

! !!

0.0
2 4 6
Petal.Length

Species ! setosa ! versicolor ! virginica

FIGURE 1.4: A scatterplot of petal lengths and petal widths from Fisher’s iris dataset
with the flowers coloured by species. The two variables are highly correlated and
separate setosa clearly from the other two species. The colours used do not reflect
the real colours of the species, which are all fairly similar.
8 Graphical Data Analysis with R

Student Admissions at UC Berkeley dataset

The data concern applications to graduate school at Berkeley for the six largest de-
partments in 1973 classified by admission and sex. One of the reasons to study the
data was to see whether there was any gender bias in the admission of students.
Figure 1.5 shows separate barcharts for the three variables (department, gender, and
whether admitted or not).

library(gridExtra)
ucba <- as.data.frame(UCBAdmissions)
a <- ggplot(ucba, aes(Dept)) + geom_bar(aes(weight=Freq))
b <- ggplot(ucba, aes(Gender)) + geom_bar(aes(weight=Freq))
c <- ggplot(ucba, aes(Admit)) + geom_bar(aes(weight=Freq))
grid.arrange(a, b, c, nrow=1, widths=c(7,3,3))

750
2000 2000
count

count

count
500

1000 1000
250

0 0 0
A B C D E F Male Female AdmittedRejected
Dept Gender Admit

FIGURE 1.5: Numbers of applicants for Berkeley graduate programmes in 1973 for
the six biggest departments. The departments had different numbers of applicants.
Overall more males applied than females and fewer applicants were admitted than
rejected.

It is immediately apparent that different numbers applied to the departments,

ranging from about 600 to 900, that there were more male applicants than female
ones, and that more applicants were rejected than admitted. None of this is exactly
rocket science and all these comments could equally well have been derived from
summary statistics (with the additional benefit of knowing that the reported range of
departmental applicant numbers was actually from 584 to 933).
Graphics make these points directly and give an overview that is easier to remem-
ber than sets of numbers. Graphics are good for qualitative conclusions and often that
is what is primarily wanted. Of course, precise numbers may be useful as well, and
the two approaches complement one another.
The ordering of the departments is alphabetic, based on the coding used, and
other orderings may be more informative. The widths of the plots were controlled
to avoid wide bars, a common issue with default charts. The vertical scales might
also have been made equal for all three charts (they are close for the last two by
chance) to reflect that they display the same cases. Then the differences between the
departments would have been downplayed.
Setting the Scene 9

The main aim of the study was to examine the acceptance and rejection rates
by gender. For the six departments taken together the acceptance rate for females
was just over 30% and for males just under 45%, suggesting that there may have
been discrimination against females. Results by department are shown in Figure 1.6,
where the widths of the bars are proportional to the numbers in the respective groups..
In four of the six departments females had a higher rate of acceptance. This is an
example of Simpson’s paradox.

library(vcd)
ucb <- data.frame(UCBAdmissions)
ucb <- within(ucb, Accept <-
factor(Admit, levels=c("Rejected", "Admitted")))
doubledecker(xtabs(Freq~ Dept + Gender + Accept, data = ucb),
gp = gpar(fill = c("grey90", "steelblue")))

Accept

Rejected

Admitted
Male Fe Male F Male Female Male Female Ma Female Male Femal Gender
A B C D E F Dept

FIGURE 1.6: Acceptance rates at Berkeley by department and gender. Overall

males were more likely to be accepted, but four of the six departments were more
favourable to females. A doubledecker plot has been used to show the numbers of
each gender applying to the departments. Relatively fewer females applied to depart-
ments A and B, while relatively fewer males applied to departments C and E.
10 Graphical Data Analysis with R

Pima Indians diabetes dataset

This dataset has been used often in the machine learning literature and it can
be found on the web in the excellent UCI library of machine learning datasets
[Bache and Lichman, 2013]. The version from R used here, Pima.tr2 in the MASS
package, is a training dataset of size 300 with zero values recorded as NA’s (a stan-
dard missing value code). There are six continuous variables and default histograms
are shown for all of them in Figure 1.7.
The distributions of three variables (plasma glucose, blood pressure, and body
mass index) look roughly symmetric. The variable skin thickness has at least one
outlier and the diabetes pedigree function distribution is skew, possibly with outliers.
The age histogram shows that most women were young with an age distribution like
half of a classical age pyramid.

data(Pima.tr2, package="MASS")
h1 <- ggplot(Pima.tr2, aes(glu)) + geom_histogram()
h2 <- ggplot(Pima.tr2, aes(bp)) + geom_histogram()
h3 <- ggplot(Pima.tr2, aes(skin)) + geom_histogram()
h4 <- ggplot(Pima.tr2, aes(bmi)) + geom_histogram()
h5 <- ggplot(Pima.tr2, aes(ped)) + geom_histogram()
h6 <- ggplot(Pima.tr2, aes(age)) + geom_histogram()
grid.arrange(h1, h2, h3, h4, h5, h6, nrow=2)

25
20
30 20

15
15
20
count

count

count

10
10
10
5 5

0 0 0
50 100 150 200 50 75 100 0 25 50 75 100
glu bp skin

25 50
60

20 40

15 40 30
count

count

count

10 20
20
5 10

0 0 0
20 30 40 50 0.0 0.5 1.0 1.5 2.0 2.5 20 40 60
bmi ped age

FIGURE 1.7: Histograms of the six continuous variables in Pima.tr2. There are a
few possible outlying values. Two of the variables have skew distributions.
Setting the Scene 11

Rather than drawing a lot of histograms to get an impression of the variables, it

takes up less space to draw boxplots. Since R automatically uses the same scale for
all boxplots in a window, they have to be standardised in some way first and this can
be achieved using the scale function. Figure 1.8 shows the result with the outliers
coloured in red for emphasis.

library(dplyr)
PimaV <- select(Pima.tr2, glu:age)
par(mar=c(3.1, 4.1, 1.1, 2.1))
boxplot(scale(PimaV), pch=16, outcol="red")

!
6

!
4

! !
! ! ! !
!
! ! !
! !
!
!
2
0
!2

glu bp skin bmi ped age

FIGURE 1.8: Scaled boxplots of the six continuous variables in a version of the Pima
Indians dataset in R. A couple of big outliers are picked out, as is a low outlier on bp
(blood pressure). The distributions of the last two variables, ped (diabetes pedigree)
and age, are skewed to the right.

There are several outliers (including a couple of extreme ones) and boxplots are
better for showing that than histograms. The last two variables are clearly not sym-
metric. Two facts should be borne in mind: The boxplots are not to be compared with
one another, drawing them all together in one plot is primarily a time and space sav-
ing exercise. The scaling just transforms each variable to have a mean of zero with a
standard deviation of one and nothing more, so the same points are identified as out-
liers as would be in the equivalent unscaled plots. This display tells us a little about
the shapes of the distributions, but not much, and nothing about the missing values
in the data, a potentially important feature. Of course, the histograms told us nothing
about the missing values either. Plots for missing values are discussed in §9.2.
The two sets of displays, histograms and boxplots, have given us a lot of infor-
mation about the variables in the dataset. A scatterplot matrix, as in Figure 1.9, tells
us even more.
12 Graphical Data Analysis with R

We can see that only two variables are strongly associated, bmi and skin, and
that that association would be even better were it not for the outlying skin thickness
measurement. All of this is valuable information, which helps us to understand the
kind of data we are dealing with.

library(GGally)
ggpairs(PimaV, diag=list(continuous=’density’),
axisLabels=’show’)

200

150 Corr: Corr: Corr: Corr: Corr:

glu

100
0.253 0.221 0.218 0.0919 0.284

!
!
! !
100 !
!
! !!!! !

Corr: Corr: Corr: Corr:

! ! ! ! !
! !! !! ! !!!! ! !!!
!!!
80
!!! ! !!
!!
!! !!
!!!!
! !!!!!!
!! !! !!
!!
!!
!
!
!!
bp

! !!!
!! !!
!!!! ! !
!!
!!!!!
!!
!!
!!!
!
!!
!!!
!!
! ! !
!!! !!
! !! ! !
0.264 0.279 !0.0382 0.399
!!! !
!
!
!!!
!
!
!!
!
!
!!
!!
!
!
!
!!
!
!
!!
!
!!
!!
! !!!!
!
!!!! !!!
! !
!!!
!
! ! ! !
60 !
!!!!!
!!
!!!
!! !
!! !! !
!
!!!!!
!!!!!!!
! !!
!!! ! ! !!
! !!! !!! !
! ! !! ! ! ! ! !
! !!
40 ! !

100 ! !

75
Corr: Corr: Corr:
skin

! !
50 ! ! !! !!

0.66 0.0914 0.255

!!!!!! !!! !!! !
!
!!!!
!!
!!!! !
! !!!
!!
!
!
!
!
!!! ! ! !! ! !!!!
!!!!!! ! !!!
!!!!
!!
! !!!!
!!!! ! !!!!! !!!
!!!!
!! !!!
! ! !!! ! ! ! !! ! !!
!! !! !!!
!!!!! ! !!!! ! ! !!!!!! !!!!
! ! !
25 !! ! !!
!
!! !
!!! !! ! !!!!!
!!!!!
!
!
!!
!
!!!
!
!
!
!!!!
! !
!!! !!!
!!! !!
!!!
! !!
!
!!!
!!!
!
!
!
!
!!!
! !
!
!!
!!!!
!!!! !
!! ! !! ! ! ! ! !
!!!!!!!! ! !!!!!!
!!!!
!! !
!
!
!!!
!
! !!!!
! !
!
! ! ! !! ! !!!
!!!!!! !! !! ! !
! ! ! !
!! ! ! ! ! ! !!! ! !!
! !
! !
50 !! !
! ! !
! !! !!
! !!!
!! !! !
! ! !! ! !!! !!!!! ! !
Corr: Corr:
! !!
!
!!!!!! !! ! ! ! !
!! !!! !! ! !!!! !!
40 ! ! ! !!
! !
!! !!!
! !! ! !!!!! !
!
! !!!! !! !
! ! !! ! ! ! !! !
!!!
! !
bmi

!!!!
! !
!! !!!
!! ! ! ! !!!!!!!
! !
!
!
!!! ! !!!
!!
! !
!!!!
!
!
!! !!
!!!!! !
! ! !! ! ! !!!! !! !
!!!!!!! !!! !
!
!!
!!
!!
! !
!!!!
!!!! ! ! !! ! !!!! ! !!
!!!!! !! !
!
!!! ! !
0.148 0.068
! !!! !!! !
!!
! ! !
! !
! !
!! !!!!!!! !! ! !
! !
!!!
!! !!
! !
! !!! !! ! !
!! !!! !!
!!! !!! ! !!!
!! !!
! ! !! !!!! ! !!! !
!!
!
!!
!!!!! !
30 ! !!!!!!!!!! !
! !
!!! ! !! !!!!
!!
!! !!!
! !!!
!!!!
!!!
!
!! !
!!
!
!
!!
!
!!
!!
!
! ! !!!! !!! ! ! ! ! ! !! ! !
!!
!!
! !!!
!!! !!!! !!!! !! !!!!!!!
!!
!
!
!!!
!! !!
!!!!
! !! ! !
!!
!
!
!!
!
!
!
!
!!
!! !
!!
!
!!! !!!!!
!! !!!!!!
! !! !!! !
! !!
!!!! !
!! !!!!
!!!
!!
!
!
!
!!!!! !
! !! !
!!
!!!
!
!!
! !
!
!!! ! !!!
!
! ! !
!!!!!!!! !!!!
!!
!!
!!
!!!
!
!
! !
!!!
!!!!!!! ! !!! ! ! !!!! ! !!!!! ! !!!!
!
20 ! !! ! ! !
! ! !
!
!
! ! ! ! !!! !! ! !!
!!
!
!
!! !
! ! ! !
2.0
Corr:
! ! ! !
1.5 ! ! !! ! ! !
ped

! !! ! !! !
! !!!! ! ! ! !!
! ! ! ! ! !!
1.0
!0.0523
! ! !! ! !
!!!!! ! !! !! ! ! !!!! ! !! !! ! ! !!
! !!! ! !! ! !!!!!! ! !
!!!! ! !!! !!! !!!!!!!!!! !! !!!!!!! ! ! !! !!!! ! !! !!
!!! !
!!!!!
! !
!! !!
! !! ! !! ! ! !
!!! !
! !!
!!
! ! !
! ! !!
! !
!!!!! ! !!!
!!! ! ! !
! ! ! ! ! ! ! ! !!! ! !!
! !
! ! ! !!
!!
! ! !
! ! !!
!!! !
! ! !
! !!!
0.5 !! !
!
!!!! ! ! !!! ! !
! ! !! ! !
! !!!!!!! !
!! !
! ! !!!!
!! !!!! ! !! !!!
!! ! !!! !! !!!!! ! ! !! ! !!!
! !!
!!
!
! !!
!!
!!! !
!!
!! ! !!
!!!!! !! ! ! !
!! !
!! !!
!
! ! ! !!!
! !!
!
!
!
!
!
!!
!
!!
!
!
!
! !
!!!!!!! !!
!
!!!!
!!
!!!!!!
!
!
!!!!!
!
!
!
! !!!
!!
! !
!
!!
! !!!!
!!
!
! !!!
! !!
!!!!!! !!!!! !!!!!!
!!!
!! !!
!!
!!!!
!!!
!!
! !!!
!! !
! !
!!!! ! ! !!!!
!!!
!
!!
!
!!
!
!
!!
!
!
!
!
! !
!!
!
!!
! !!
!
!!!
!! !!
!!!!!
!
!
!
!
!!
!
!! !
!
!!
!!
!
! !!!! !
!! !!! ! !
!!
!!!!!
!
!
!
!
!
!!
!
!!!
!
!
!
!
!
!
!!!
!
!
!
! !!
!!!
!
!
!!
!!!
!
!!
!!!!!
! ! !!!
!!! ! !!!
!!!!!!
!!!
!!
!!
!!!
!
!
!
!!!
!!
!!
!!
!!
!
!!!! ! !!!
!!
!!
!!!!
!
!
!!
!
!!
!
!
!
!
!!!
!
!!! !
!!!
!!!!!
!
!
!
!!
!!
!!
!
!
!! !!
!!
!
!
!
!
!!
!
!!
!
!!
!
!!
!!
!
!
!
!
!!!
!!! !
! !
! !
! !!
! ! ! ! !!
! !
!
!!!
!! !! !! !!!! ! !
!!!! ! ! !!
!
! !!!
!
!!
! !!! !
!
!!
! ! !
!! !!!
0.0 !! ! !
!
! ! !!!
! !
! ! ! !!
!! !
!!!!!! !!!

70 !
!! !!
!
! ! !
!
! ! ! !
60 ! !! ! !! !
! !
! ! ! !! ! !! !
!! !
! ! !!! !! ! ! !!!! ! !! ! ! !!!! !!
! ! !!
!! !! ! !! !! ! ! ! ! ! !! ! ! !
! ! !! !
!! ! ! ! ! ! !! ! !! !
!! !! !! ! !!! ! !
50 ! ! ! !! ! !!! ! ! !!!! ! ! !!!!! ! ! !!
age

!!!!! !! !! !!!
! ! ! ! !!! ! !! !!!!! !
! !
!
! ! !
!!
! !! !
!! ! !! ! ! ! !! !!!
!! !! !! !!! ! !! ! !
!
!!
!
!! ! ! ! !!
!!!!! !! !!!
! !! !
!!! ! !!
40
! !! ! ! !! ! !!
!
! !!! !!! !! ! !!! ! !! !! !! ! !
! !!! !! !! !! !
!!!
! !! ! !! !!! !!!! !!! ! ! !!
!!!
!!
!
!! !! !
! !
!!! !
! ! ! !! !!!!!!!
!! ! !! !!!!! !! !!
!
! ! !! ! !
! !
! !
! !!! ! ! ! !!! !!!!! !! ! !! ! !!!
! !! !! !! !! ! !!! !!
!!
!!! ! !!
!! !!! !!
! ! !! !!!!!
!!!!!! !!!!!
! !! !!
!! !!
!! !
!!!!
!
!! !! ! !! !!!! ! !! !!! !! ! ! !!!!!!
!
30 !!!
!! ! ! ! ! !! !! !! !!! !
!!
!
!
!
!
!
!!!
!!! !
!
!! ! !! ! ! !!!
!!! !
!!!
! !!!
!!! !! !!! ! ! !!!! ! ! !
! !!!!!!!!
!!
! !!!! ! !! !! !! !
!!
! !
!!!!!
!!!
! !!! !!
!! !
!! !!
!! !!!!!!!!
! !! ! !!!! !
!!! ! !!!!!!! !
!
! !!! ! !! !! !!!
!!!!! !!! !!!! ! !!!
!!
! !!
!
!!
!!
!!!
! !
!!!!!!
!!!! !
!!
!!
!
!!
!
!
!
!
!!!
!!
!!! !!!!!!
!!!
!
! ! !!! !!
!!!
!!!!
!
!
!!!!
! !! !!! !
!!!!
!!
!
!
!! !
!
!
!!!!
!
!
!!
!
!!
!!
!
!
!!!!
!
!! !!
!
!!!
!!!!!!!
!!
!
!
!!
! !! !
! !!! !! !
! !!!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!!!!
!!! !!!! ! !
!! !!
!!!
!!! !
! ! !! ! ! !! !!
!! ! ! !! !!! !!!!!
!!
! ! !!
!!! !
!! !
!!!!!!!!!
! !! ! ! !!! !!
!!!!!!!!!!!!!
! !!
!!!!! !
!!!! !!!! ! !!
! !
!! !! ! !!!!!! !!
!!!!!
!!! !
!!!!!
20 ! ! !!!
!! ! !
!! ! !!
!! ! ! !!!!!! !
!! !! !! ! ! ! !
!! !!!
! !! !! !
! !!!!
!
!
!!!
! !!!!
! !!!
!!
!!
!!!
! !
! !!! !!!
!!!!!!!
!!!
!!!!
!!!!
!
!!!!!! ! !!
!
!!
!
!!!
!
!!
!!
!!
!
!!
!
!!!
!!! !!! !!!!!
!!!
!!!!
!
!
!!!
!! !!
!
!
! !
!!
!!! ! ! !!
! !! !!
!
!!
!!
!!
!!
!
!!
!!!
!
!!
!
!!
!
!!!
!! !!

50 100 150 20040 60 80 100 25 50 75 10020 30

glu bp skin bmi40 50 0.0 0.5 1.0
ped1.5 2.0 20 30 40 50 60 70
age

FIGURE 1.9: A scatterplot matrix of the six continuous variables for the same Pima
Indians dataset. Only one of the scatterplots shows a strong association (bmi and
skin). The two extreme outlying values, one skin and one ped measurement,
make it harder to see what is going on, as there is less space left for the bulk of the
data.
Setting the Scene 13

GDA in context
GDA does not stand on its own. As has already been said (and should probably
be repeated several times more), any result found graphically should be checked
with statistical methods, if at all possible. Graphics are commonly used to check
statistical results (residual plots being the classic example) and statistics should be
used to check graphical results. You might say that seeing is believing, but testing is
convincing.
Graphics are for revealing structure rather than details, for highlighting big dif-
ferences rather than identifying subtle distinctions. Edwards, Lindman and Savage
wrote of the interocular traumatic test; you know what the data mean when the con-
clusion hits you between the eyes [Edwards et al., 1963]. They were referring to sta-
tistical analyses in general, but it is particularly relevant for graphics. Their article
continued: “The interocular traumatic test is simple, commands general agreement,
and is often applicable; well-conducted experiments often come out that way. But the
enthusiast’s interocular trauma may be the skeptic’s random error. A little arithmetic
to verify the extent of the trauma can yield great peace of mind for little cost.”
Approximate figures suffice for appreciating structure, there is no need to provide
meticulous accuracy. If, however, exact values are needed—and they often are—then
tables are more useful. Graphics and tables should not be seen as competitors, they
complement one another. With printed reports sometimes difficult choices have to
be made about whether to include a graphic or a table. With electronic reports the
question becomes how to include both gracefully and effectively.
Graphical Data Analysis is obviously appropriate for observational data where
the standard statistical assumptions that are needed for model building may not hold.
It can also be valuable for analysing experimental data. There may be patterns by time
or other factors that were not expected. In medical studies, the balance of control and
treatment groups has to be checked.
This book concentrates on exploratory graphics, using graphics to explore
datasets to discover information. Experience gained in looking at and interpreting
exploratory graphics will be valuable for looking at all kinds of graphics associated
with statistics, including diagnostic graphics (for checking models) and presentation
graphics (for displaying results).
The importance of data analysis in general is sometimes underplayed, because
there is little formal theory and because the results that are found may appear obvi-
ous in retrospect. Effective graphical analysis makes things seem obvious, the effort
involved in making the graphical analysis effective is not so obvious. In his poem
“Adam’s Curse” W. B. Yeats wrote of the amount of work that went into getting a
line of poetry right:
Yet if it does not seem a moment’s thought,
Our stitching and unstitching has been naught.
That is an appropriate analogy here.
14 Graphical Data Analysis with R

1.4 Using this book, the R code in it, and the book’s webpage
A graphic is more than just a picture and every display in the book should convey
some information about the dataset it portrays. There should be some description of
what you can see accompanying every graphic. In order to ensure that all discussions
of graphics are either on the same page as the graphic itself or on the opposite page,
gaps have been left on some pages. This is intentional, as however irritating gaps may
be, it always seems more irritating having to turn pages backwards and forwards to
flip between displays and descriptions.
Like anything else, using graphics effectively is mostly a matter of practice. Study
and criticise the examples. Test out the code yourself—it is all available on the book’s
webpage, rosuda.org/GDA. You can experiment with it while you are reading
the book, just copy and paste the code into R. Vary the size and aspect ratio of
your graphics, vary the scaling and formatting, vary the colours used. Draw lots of
graphics, see what you get, and decide what is most effective for you in making it
easy to recognise information.
Work through the exercises at the ends of the chapters. All these exercises use
datasets available in R or in one of the packages associated with R. There are no
purely technical exercises, they all require consideration of the context involved. The
goal is not just to draw graphics successfully, but to interpret the resulting displays
and deduce information about the data. Some exercises are more open-ended than
others and you should not expect definitive answers to all of them. The best approach
is to try several versions of each graphic and to work with sets of graphics of different
types, not just with individual ones. Doing the exercises is highly recommended—
to become experienced in carrying out Graphical Data Analysis, you need to gain
experience in looking at graphics.
There are far too many R packages to load them all. It makes sense to ask read-
ers to load the packages that are used more often in the book instead of repeatedly
referring to them in the text. Please ensure you have the following packages loaded:
ggplot2 for graphics based on ideas from “The Grammar of Graphics”. Most of the
book’s graphics use these ideas.
gridExtra for arranging graphics drawn with ggplot2.
ggthemes for its colour blind palette.
dplyr for advanced and transparent data manipulation capabilities.
GGally for additional graphics in ggplot2 form, including parallel coordinate plots.
vcd for a range of graphics for categorical data.
extracat for multivariate categorical data graphics and for missing value patterns.
Some of these packages load further packages via a namespace. To check the state
of your R session you can use the function sessionInfo().
Setting the Scene 15

Loading packages in advance will mean that the functions in these packages can
be immediately used, and that any datasets supplied with the packages are to hand.
For datasets in other packages there are two cases to consider. With LazyData
packages datasets can be accessed without loading the package, as long as you have
the package installed. With (most) other installed packages a data statement is
needed, an R function for making a dataset available. To avoid repetition, datasets
are generally only loaded once in each chapter.
Many graphics are improved by an appropriate choice of window size, informa-
tive labelling, sensible scaling, good positioning (for instance in multiple graphics
displays) and other details, which are more about presentation than the graphic it-
self and may require much more code than you might expect. This book is about
exploratory graphics so the code is mainly restricted to the graphics essentials.
You may find some graphics too small (or too big). If so, redraw them yourself
and experiment to find what looks best to you. Space and design restrictions in a
printed book can hamper displays. For some of the graphics the code includes ad-
justments to improve the look of the default versions. Graphics for exploration are
usually only on display temporarily, while graphics for presentation, especially in
print, are more permanent. Nevertheless a little enhancing helps to avoid occasional
displeasing elements in exploratory graphics. It is often a matter of taste and you
should develop your own style of graphics for your own use. For presenting to others
you need to think of their needs and expectations as well. Some general advice on
coding graphics in R is given in Chapter 13.
Code listings for every plot are given in the book and on the book’s webpage for
downloading. The code is not explained in detail, so if an option choice puzzles you,
check the help file for the function, especially the examples there. With R there are
always several ways of achieving the same goal and you may find you would have
done things differently. The end result is the important thing.
Graphical functions in R can offer very many options and working out what ef-
fects they have, especially in combination, can be complicated. It would be nice to be
able to say that all functions follow the same rules with similarly defined parameters.
Languages are rarely consistent like that, and R is no exception.
No formal statistical analyses are carried out in this text, as there are already many
fine books covering statistical modelling. A list of suggested references is given at the
end of Chapter 2 and there are a few remarks at the end of each chapter. Readers are
encouraged to look for statistical methods to complement their graphical analyses.
You should be able to find the tools to do so in R and its many packages. There is
often a number of ways offered to carry out particular analyses, each with its own
advantages (and possibly disadvantages), so no recommendations are made here.
The final exercises in each chapter are labelled “Intermission” and are intended
to be a break and a distraction. Perhaps they would have been better labelled “And
now for something completely different”. At any rate, I hope they lead you to some
interesting visual discoveries and to developing your visual skills in many other di-
rections.
16 Graphical Data Analysis with R

Main points
1. Graphical Data Analysis uses graphics to display and interpret data to reveal the
information in a dataset. It is an exploratory tool rather than a confirmatory one.
2. Simple plots can reveal useful information about datasets.
Figure 1.1 showed a surprising feature of the Speed Skiiing Championships and
Figure 1.2 explained it. Figure 1.3 showed the two groups of flowers with quite
different petal lengths. Figure 1.5 showed that more males than females applied
to the Berkeley graduate program. Figure 1.8 showed that there are some extreme
outliers in the Pima Indians dataset.
3. Scales and formatting of plots are important.
Using the same scales for the two histograms in Figure 1.1 and aligning them
above one another is essential for conveying the information in the plots effec-
tively. Figure 1.3 is an informative histogram for the iris variable, as we can
see that there are two groups and the distributions of lengths within them. Bar-
charts with different numbers of categories, but from the same dataset, have dif-
ferent default scales and so care is necessary with interpretations across plots
(Figure 1.5). Comparing distribution forms for differently scaled variables needs
some standardisation first (Figure 1.8).
4. Different plots give different views of the data.
While Figure 1.3 displays the distribution of petal lengths in the iris dataset,
Figure 1.4 shows the close relationship between petal length and petal width.
Figure 1.7 shows the distribution shapes of the Pima Indian variables and Fig-
ure 1.8 emphasises outliers. Figure 1.9 shows that only two of the variables in
the dataset appear to be strongly related.
Setting the Scene 17

Exercises
More detailed information on the datasets is available on their help pages in R.
1. Iris
How would you describe this histogram of sepal width?

ggplot(iris, aes(Sepal.Width)) +
geom_histogram(binwidth=0.1)

2. Pima Indians
Summarise what this barchart shows:

ggplot(Pima.tr2, aes(type)) + geom_bar()

3. Pima Indians
Why is the upper left of this plot of numbers of pregnancies against age empty?

ggplot(Pima.tr2, aes(age,npreg)) + geom_point()

4. Estimating the speed of light

There are 100 estimates of the speed of light made by Michelson in 1879, com-
posed of 5 groups of 20 experiments each (dataset michelson in the MASS pack-
age).
(a) What plot would you draw for showing the distribution of all the values
together? What conclusions would you draw?
(b) What plots might be useful for comparing the estimates from the 5 different
experiments? Do the results from the 5 experiments look similar?

5. Titanic
The liner Titanic sank on its maiden voyage in 1912 with great loss of life. The
dataset is provided in R as a table. Convert this table into a data frame using
data.frame(Titanic).
(a) What plot would you draw for showing the distribution of all the values
together? What conclusions would you draw?
(b) Draw a graphic to show the number sailing in each class. What order of
variable categories did you choose and why? Are you surprised by the dif-
ferent class sizes?
(c) Draw graphics for the other three categorical variables. How good do you
think these data are? Why are there not more detailed data on the ages of
those sailing? Even if the age variable information (young and old) was
accurate, is this variable likely to be very useful in any modelling?
18 Graphical Data Analysis with R

6. Swiss
The dataset swiss contains a standardized fertility measure and various socio-
economic indicators for each of 47 French-speaking provinces of Switzerland in
about 1888.
(a) What plot would you draw for showing the distribution of all the values
together? What conclusions would you draw?
(b) Draw graphics for each variable. What can you conclude from the distribu-
tions concerning their form and possible outliers?
(c) Draw a scatterplot of Fertility against % Catholic. Which kind of
areas have the lowest fertility rates?
(d) What sort of relationship is there between the variables Education and
Agriculture?
7. Painters
The dataset painters in package MASS contains assessments of 54 classical
painters on four characteristics: composition, drawing, colour, and expression.
The scores are due to the eighteenth century art critic de Piles.
(a) What plot would you draw for showing the distribution of all the values
together? What conclusions would you draw?
(b) Draw a display to compare the distributions of the four assessments. Is it
necessary to scale the variables first? What information might you lose, if
you did? What comments would you make on the distributions individually
and as a set?
(c) What would you expect the association between the scores for drawing and
those for colour to be? Draw a scatterplot and discuss what the display
shows in relation to your expectations.

8. Old Faithful
The dataset faithful contains data on the time between eruptions and the duration
of the eruption for the Old Faithful geyser in Yellowstone National Park, USA.
(a) Draw histograms of the variable eruptions using the functions hist
and ggplot (from the package ggplot2). Which histogram do you prefer
and why? ggplot produces a warning, suggesting you choose your own
binwidth. What binwidth would you choose to convey all the information
you want to convey in a clear manner? Would a boxplot be a good alterna-
tive here?

(b) Draw a scatterplot of the two variables using either plot or ggplot. How
would you summarise the information in the plot?

9. Intermission
Van Dyck’s Charles I, King of England, from Three Angles belongs to the Royal
Collection in Windsor Castle. What is gained from having more than one view
of the King?
2
Brief Review of the Literature and
Background Materials

A picture shows me at a glance what it takes dozens of pages of a book to

expound.

Ivan Turgenev (Fathers and Sons)

Summary
Chapter 2 reviews some of the available literature on graphics for data analysis and
statistics, gives a brief overview of alternative software to R for graphics (as at the
time of writing), and explores what relevant material is available on the web: discus-
sions, graphics and datasets. Finally there is a list of texts covering statistical models
that might be used in conjunction with the graphical approach described in the book.

2.1 Literature review

Progress in graphics is bedevilled by the lack of theory. Bertin’s classic “Semiology
of Graphics”, originally published in French in 1967, recently reissued by ESRI Press
[Bertin, 2010], and Wilkinson’s “Grammar of Graphics” [Wilkinson, 2005] are two
of the few works that attempt to make substantial contributions. Bertin is not an easy
read and is now old-fashioned in many ways, being essentially from a pre-computing
age, but the book contains many interesting ideas. It has a strong geographic empha-
sis. Wilkinson describes a formal structure and includes many attractively drawn dis-
plays, although the reasons for drawing the graphics are seldom discussed. Hadley
Wickham developed his R package ggplot2 using the “Grammar of Graphics” and
has shown how useful Wilkinson’s grammar can be in practice, that it is more than a
theoretical construct.
There are far too many books to mention every one that offers good advice on
how to draw graphics (let alone to mention all that offer any advice). Tufte’s books,
especially his first one [Tufte, 2001], are an excellent starting point, containing many

19
20 Graphical Data Analysis with R

attractive and instructive examples as well as some cautionary ones. Tufte offers gen-
eral principles and discusses both graphics for data analysis and statistics and also
what is now called Information Visualisation. Cleveland’s advice ([Cleveland, 1993]
and [Cleveland, 1994]) is more specific and directed at statistical graphics. Wainer
has written a number of enlightening books ([Wainer, 1997], [Wainer, 2004] and
[Wainer, 2009]) illustrating with a range of real examples how graphics can be used
to reveal information in data. [Robbins, 2005] is worth looking through for prac-
tical advice and [Few, 2012], while taking a business-oriented line, supplies force-
ful opinions in a strongly argued text. For analysts working in the Life Sciences
[Krause and McConnell, 2012] offers sound advice and discussion of many real ap-
plications.
For graphics in R there is the book [Murrell, 2005], which appeared in an exten-
sively revised second edition in 2011. It covers the full range of R graphics and in
particular it explains the details of Murrell’s grid graphics package, which he im-
plemented as a more structured alternative to the original graphics system in R. The
book is fairly technical, but essential for anyone wanting to write graphics using grid
and valuable for anyone wanting guidance on how graphics in R work. Wickham’s
ggplot2 package and Sarkar’s lattice package are based on grid. The additional
power and flexibility of grid comes at the expense of occasional slowness. Wick-
ham [Wickham, 2009] and Sarkar [Sarkar, 2008] have both written useful books on
their packages, although given the continual improvements and changes being made,
it is probably best to consult the packages’ websites rather than the books.
If you want a specific form of graphic that is not directly available or want to
amend or embellish a particular graphic, and these books are not enough, then one
of the R cookbooks may help. There are at least two for graphics, [Mittal, 2011] and
[Chang, 2012]. In addition, there is the German book [Rahlf, 2014], which provides
extensive code examples for producing elegant images. Finally, given that drawing
graphics often requires restructuring the data first, it may be helpful to consult one of
the many textbooks and resources available for the R language itself.
For graphics not specifically tied to R there have been a number of interesting
books in recent years. The visualization volume [Chen et al., 2008] in the Hand-
book of Computational Statistics series offers a collection of articles from authors
with many different views of data visualization. The book “Graphics of Large
Datasets” [Unwin et al., 2006] considers the problems of visualising datasets that
are a lot bigger than the datasets graphics texts usually discuss. [Inselberg, 2009]
treats parallel coordinate plots in detail, emphasising their geometric properties.
[Cook and Swayne, 2007] covers dynamic graphics with special reference to the
GGobi software for rotating plots. In [Theus and Urbanek, 2007] the authors give
an overview of interactive graphics for data analysis including several in-depth case
studies. To see how much progress has been made, it is worth looking at “Graph-
ical Exploratory Data Analysis” [DuToit et al., 1986]. Both the ease of producing
graphics and the quality of graphics produced have improved enormously in the last
twenty-five years.
As this book is about graphics for data analysis and statistics, the emphasis
is on the relevant statistical literature. There is additionally an extensive litera-
Brief Review of the Literature and Background Materials 21

ture on Information Visualisation, which overlaps in interest, although not always

in approach, with the books mentioned already. The books [Spence, 2007] and
[Bederson and Shneiderman, 2003] are both worth consulting.
There are important factors which affect graphical displays that have nothing to
do with data analysis per se. Colour, perception, psychology all play critical roles
and effective design should take account of them [Ware, 2008]. There are valuable
books on design in general such as [Norman, 1988]. Design is difficult, and much
of the theory in these areas is still too far removed from practice to have a direct
influence. Theories can explain why something does not work well and they can
provide rational support for principles. That does not mean that they can offer specific
advice on how to tackle particular tasks. Graphics are still very much a matter of taste.

2.2 Interactive graphics

It would be misleading to talk about GDA without at least some discussion of inter-
active graphics. Interaction means being able to interact directly with the statistical
components of a graphic display. This makes graphical analyses faster and more flex-
ible. The corresponding disadvantages are the difficulty of recording what has been
done and the lack of presentation quality reproduction. This will change, and the
increasing use of the web for presenting analyses will likely have a big influence.
There are interesting problems to be solved, notably how to formalise interactive
graphics, how to provide an intuitive interface for the software, and how to provide
results of analyses interactively in a structured form for occasional users. Querying
values, zooming in, and reformatting graphics are sensible starting points and much
can be achieved with RStudio’s Shiny. More is possible, in particular the linking of
several associated graphics for the same dataset, which the RCloud project of a group
at AT&T provides [Urbanek et al., 2014].
Dynamic graphics, where the displays are animated in some way, for instance ro-
tating through low-dimensional projections of high-dimensional data, is an attractive
subset of interactive graphics.

2.3 Other graphics software

Lots of people draw their graphics in Excel, and why not? For many purposes it is a
very useful tool. It offers a large variety of alternative graphic forms and considerable
numbers of options for amending the appearance of the graphics. Since it is easy to
record data in a spreadsheet, Excel is a natural tool for small projects, even though it
offers only a limited range of statistical methods. As with all software, it is better for
some graphics than others (histograms are difficult to draw in Excel).
22 Graphical Data Analysis with R

The ‘big’ statistics packages like SAS and SPSS provide extensive graphics ca-
pabilities and, like Excel, aim more for preparing presentation graphics than ex-
ploratory graphics. They are capable of handling large and complex datasets, provide
extensive and substantial statistical capabilities, and can be used to set up regularly
operating analysis systems as well as to carry out one-off studies. There are also
commercial packages for presentation graphics, some more closely associated with
statistical tools than others.
Although R offers interactive graphics resources through the iplots, rggobi, and
ggvis packages, it is more for static graphics. There have been a number of commer-
cial interactive graphics softwares, notably Data Desk, JMP, Spotfire, and Tableau.
They all have their strengths and weaknesses. Data Desk was very innovative for its
time and is still impressive. JMP has the most powerful statistical tools of the group
and offers a tightly integrated system. Spotfire arose out of work in Shneiderman’s
group in Maryland. Tableau is the newest of the group and originated from work at
Stanford. On the research side of interactive graphics there are tools like Mondrian
and GGobi. All these software packages have been for general applications. Recently
there have been some tools designed for particular applications made available on the
web, such as Gapminder for displaying scatterplots animated over time or Wordle for
word clouds.
The main advantage of concentrating on R is the integrated access to R’s exten-
sive range of statistical models and tools.

2.4 Websites
The web changes so quickly that anything written to-day is liable to be out-of-date
tomorrow. Nevertheless, several of these websites have been around a while and there
are many crosslinks between them. New sites that are any good will doubtless quickly
be linked, and once you have found a starting point you should be able to find further
interesting sites. Failing that, there is always Google or whatever may take over from
them.
A number of websites have sprung up around the theme of data visualization
encouraging contributions from readers. Both the FlowingData [Yau, 2011] and
Junk Charts [Fung, 2011] websites discuss visualisations of data and have many
interesting examples, often taken from the media. Statistical Graphics and More
[Theus, 2013] is a similar site. Many Eyes [IBM, 2007] lets users create visualisa-
tions of datatsets on their site and upload their own datasets for others to visualise.
Unsurprisingly the resulting graphics are of a very mixed standard. Visualizing.org
[Viz, 2011] sees itself as a forum, where readers can discuss visualisation, contribute
graphics, and find data. It appears to be more for designers than statisticians. The
British site Improving data visualisation for the public sector [OCSI, 2009] wants to
do just that and includes examples, case studies, and guides.
Tufte includes a discussion page on his website [Tufte, 2013], covering visualisa-
Brief Review of the Literature and Background Materials 23

tion issues amongst other topics. Gelman’s blog Statistical Modelling, Causal Infer-
ence, and Social Science [Gelman, 2011] often includes debates about graphics and
on which alternatives readers prefer. There are sites, which are more Infoviz oriented
than statistics oriented, such as eagereyes [Kosara, 2011] and information aesthetics
[Vande Moere, 2011].
The Gallery of Data Visualisation [Friendly, 2011] has many classic graphics and
helpful supporting historical information. The choice of graphics on display is mildly
idiosyncratic, although none the worse for that, and there are enough splendid graph-
ics to suit everyone’s taste. The R Graph Gallery used to provide a large number of
graphics drawn in R and the code used to draw them. Visitors to the site could vote on
how good they thought the individual graphics were and it was curious to see which
were rated highly and which poorly. The voting probably said more about the vot-
ers than about R graphics. The site is no longer maintained. So many websites have
sprung up around R that it is impossible to keep track and it would be inadvisable
to make firm recommendations given the speed with which things change. All that
can be said is that if you are looking for good advice, it is almost certainly available
somewhere, just be cautious and check carefully any advice you intend to use.

2.5 Datasets
Many datasets are used in this book, often several times. You need to gain experience
in using various kinds of graphics with the same data, and in using graphics for
different contexts with different kinds of data. All the datasets are already available
in R, in one of its packages, or in the package GDAdata accompanying the book.
The sources of the datasets and definitions of the variables in them can be found on
the corresponding R help pages. To see where a dataset is referred to or analysed
by other users of R, you can try one of the search functions on CRAN (http:
//cran.r-project.org).
There are some excellent datasets in R and readers can easily find them to ex-
periment with the data themselves. Unfortunately, there is the disadvantage that the
datasets are not always fully documented and that they are sometimes only provided
in a cut-down form, for no apparent reason. It is a pity that some R examples are
more just for showing how commands work rather than for illustrating how and why
you might want to use those commands. More could be done [Unwin et al., 2013].
On the other hand, sorting out the necessary background information and sources for
a dataset can involve a lot of hard work (e.g., see the discussions in §3.3 on Gal-
ton’s and Pearson’s family height datasets). Successful data analysis requires sound
knowledge of context, so that you can make sense of your results.
A number of datasets are available in several different packages, occasionally
under different names and in various versions or formats. For instance, the Titanic
dataset used in the book is the one from the datasets package. You will also find ti-
tanic (COUNT, prLogistic, msme), titanic.dat (exaxtLoglinTest), titan.Dat (elrm),
24 Graphical Data Analysis with R

titgrp (COUNT), etitanic (earth), ptitanic (rpart.plot), Lifeboats (vcd), TitanicMat

(RelativeRisk), Titanicp (vcdExtra), TitanicSurvival (effects), Whitestar (alr4), and
one package, plotrix, includes a manually entered version of the dataset in one of
its help examples. The datasets differ on whether the crew is included or not, on the
number of cases, on information provided, and on formatting. Versions with the same
names in different packages are not identical.
Collecting data requires considerable effort. The individual experiments carried
out by Michelson to estimate the speed of light over one hundred years ago are a
classic example (cf. Exercise 4 in Chapter 1). There used to be plenty of effort in-
volved in loading datasets into a computer for analysis as well (although far less
than needed for gathering the original data). It is amusing to read the instructions
for the 1983 DataExpo competition organised by the ASA’s Committee on Statistical
Graphics: “Because of the Committee’s limited (zero) budget for the Exposition, we
are forced to provide the data in hardcopy form only (enclosed). (Sorry!) There are
406 observations on the following 8 variables...”
In the early years of the web there were a few sites like Statlib at Carnegie
Mellon University or UCI’s Machine Learning repository that collected datasets
for free use by others. The datasets tended to be small and primarily for teach-
ing purposes (Statlib) or for studying algorithms (UCI). In recent years the situa-
tion has totally changed. It is astonishing how much data is now provided on the
web. Statistical offices, health departments, election offices, and other official bodies
make substantial amounts of data available (for instance, the US Government’s site
www.data.gov). Sports data are often collected and put up on the web (e.g., the
splendid Estonian decathlon webpage www.decathlon2000.com). The British
newspaper the Guardian runs a Datastore [Guardian, 2011] where they publish
many datasets of public interest ranging from AIDS statistics round the world to
more parochial issues like the expenses claimed by British MP’s.
Some academic journals require contributors to make their data publicly avail-
able. For instance, in 2014 the American Statistical Association’s information for
authors for its main journal JASA included this rule: “Whenever a dataset is used, its
source should be fully documented and the data should be made available as an on-
line supplement. Exceptions for reasons of security or confidentiality may be granted
by the Editor.” [ASA, 2014] We can expect further progress in this direction.
There are R packages to assist you in downloading data from websites that offer
their data in particular formats. The book “Data Technologies” [Murrell, 2009] gives
advice on gathering and organising web data. When working with web datasets (or
indeed with any dataset collected by someone else), it is always a good idea to check
the source of the data and to ensure you have sufficient background information. Ide-
ally you should know the aims of the original study from which the data came, how
the variables are defined, how the data were collected, and what editing or cleaning
of the data has been carried out. Investing time in analysing data that turn out to be
flawed is a bad use of your time.
Brief Review of the Literature and Background Materials 25

2.6 Statistical texts

Although this book is about graphics for data analysis, there is no claim being made
that graphics are enough on their own, far from it. Statistical models are essential for
checking ideas found through graphics just as graphics are important for checking
results from models.
There are many excellent statistics textbooks that can be recommended for statis-
tical modelling (although perhaps not always for statistical graphics). The following
is a personal selection that should cover the models mentioned in this book and more
besides.
Sound introductory texts include [Freedman et al., 2007], [De Veaux et al., 2011],
and [Maindonald and Braun, 2010]. Other books like [Gelman and Hill, 2006],
[Davison, 2008], and [Fahrmeir et al., 2013] assume more knowledge on the part of
the reader. All are excellent books. The classic “Modern Applied Statistics with S”
[Venables and Ripley, 2002] is still worth reading for its smooth integration of theory,
software, and application. Overlaps of statistics with machine learning are well han-
dled by the splendid [Hastie et al., 2001]. Recommended texts covering specific ar-
eas of statistics are [van Belle et al., 2004] (Biostatistics), [Tutz, 2012] (Categorical
Data Analysis), [Kleiber and Zeileis, 2008] (Econometrics), [Lumley, 2010] (Survey
Analysis), [Ord and Fildes, 2013] (Time Series), and, for those wanting a Bayesian
approach, [Gelman et al., 2013].
3
Examining Continuous Variables

Normality is a myth; there never was, and never will be, a normal distribu-
tion.

Roy Geary (“Testing for Normality”)1

Summary
Chapter 3 looks at ways of displaying individual continuous variables.

3.1 Introduction
A continuous variable can in theory take any value over its range. (In practice data for
continuous variables are generally rounded to some level of measurement accuracy.)
Many different plots have been suggested for displaying data distributions of contin-
uous variables and they all have the same aim: to display the important features of
the data. Some plots emphasise one feature over another, some are very specialised,
some require more deciphering than others. Perhaps the reason there are so many is
that there are so many different kinds of features that might be present in the data.
Two possible approaches are to use a range of different plot types or to use a
variety of different plots of the same type. In this book the latter approach is mostly
favoured and the plots in this chapter are primarily histograms and boxplots. Readers
should use what they feel most comfortable with and what works for them. For any
plot types you do use, make sure you have plenty of experience in interpreting them.
Look at examples of your plots for all kinds of different datasets and get to know
both the ways the kinds of features you are interested in are represented and what
kinds of structure are emphasised. Textbooks rarely have the space for showing a
wide variety of graphics, you have to draw your own.

1 He added “This is an overstatement from the practical point of view. . . ” [Geary, 1947]

27
28 Graphical Data Analysis with R

Figure 3.1 shows a histogram of the percentage share of the vote won by Die
Linke in each of the 299 constituencies in the 2009 Bundestag election. Amongst the
German political parties there are two on the left of the political spectrum, the SPD,
similar to the Labour party in the UK, and Die Linke, literally ‘The Left’, a party
even further to the left. Before drawing the histogram, a new variable is calculated to
give the percentage support in each constituency. The binwidth is chosen to be 1 to
give bins that are easy to interpret. Thus the height of each bin shows the number of
constituencies with a particular percentage of Die Linke supporters. Although they
are a party on the left, they are usually given a purple colour in displays of results, as
red is used for the SPD.
The party had little support in most of the country, but managed over 10% in
a fifth of the constituencies. The areas where they did well turned out to be in the
new Bundesländer (the old East Germany), in East Berlin, although not in West
Berlin, and in the small Bundesland called Saarland, where their party leader in
that election was based, Oscar Lafontaine. You might have guessed at a pronounced
regional distribution given the form of the histogram, but not that the country would
be so precisely divided up.

data(btw2009, package = "flexclust")

btw2009 <- within(btw2009, Linke2 <- 100*LINKE2/valid2)
ggplot(btw2009, aes(Linke2)) + geom_bar(binwidth = 1,
fill = "mediumpurple") + ylab("") +
xlab("Percentage voter support for Die Linke in 2009")

40

20

10 20 30 40
Percentage voter support for Die Linke in 2009

FIGURE 3.1: Percentage support by constituency for Die Linke party in the German
election of 2009. The graphic suggests a division of the constituencies into two dis-
tinct groups. Further investigation revealed that the constituencies where Die Linke
did poorly are all in the old Western Germany, including West Berlin, but excluding
Saarland, where the party leader was based.
Examining Continuous Variables 29

3.2 What features might continuous variables have?

There might be
Asymmetry the distribution is skewed to the left or right, for instance distributions
of income.
Outliers there are one or more values that are far from the rest of the data.

Multimodality the distribution has more than one peak, e.g., both variables in the
Old Faithful geyser data.
Gaps there are ranges of values within the data where no cases are recorded. This
can happen with exam marks, where there are no cases just below the pass mark.
Heaping some values occur unexpectedly often. Birthweights are a good example
[Clemons and Pagano, 1999]. Perhaps there are more important things to do than
to weigh the newborn baby to the nearest gram or ounce . . .

Rounding only certain values (perhaps integers) are found, e.g., age distributions.
Impossibilities values outside the feasible range, for instance negative ages. There is
a version of the Pima Indians dataset discussed in 1.3 in the UCI machine learn-
ing repository [Bache and Lichman, 2013] in which there are blood pressure and
skin thickness measurements of zero.
Errors values that look wrong for one reason or another. In a German car insurance
dataset there were drivers whose year of birth gave them ages of less than 16,
so they could not have a licence. It was suggested that this might be possible, as
insuring someone who never drove would enable them to build up a history of
several years of no claims!
Graphics are good for displaying the features that make up the shapes of data
distributions. They can provide more and different kinds of information than a set of
summary statistics. Obviously it is best to use both approaches.
With a single variable the mean is usually the most important statistic and perhaps
no statistical test is used as often as the t-test for testing means. A t-test can be used
if the underlying data are from a normal distribution. For small datasets (and the t-
test is intended specifically for small samples) data from a normal distribution can
look very non-normal, which is why tests of normality have low power and provide
little support for t-tests. Fortunately the t-test is fairly robust against non-normality.
This should not prevent anyone from at least checking whether the data have some
seriously non-normal feature before carrying out analyses. That can best be done
graphically.
30 Graphical Data Analysis with R

3.3 Looking for features

This section discusses a number of different datasets, mainly using histograms, to
see which features might be present and how they can be found. Whichever display
you favour, it is important to study lots of examples to see the various forms of the
graphic that can arise and to get experience in interpreting them.

Galton’s heights

Children Parents

150 200

150
Frequency

Frequency

100

100

50
50

0 0

62 64 66 68 70 72 74 64 66 68 70 72

FIGURE 3.2: Histograms of child and midparent heights with default binwidths. The
plots have been drawn separately, so the scales are different. Both distributions are
roughly symmetric and there is more spread amongst the children. The binwidth has
been set to 1 inch. If the data were in cms, then another binwidth would be more
suitable, perhaps 2.5 cms.
data(galton, package="UsingR")
ht <- "height (in)"
par(mfrow=c(1,2), las=1, mar=c(3.1, 4.1, 1.1, 2.1))
with(galton, {
hist(child, xlab=ht, main="Children", col="green")
hist(parent, xlab=ht, main="Parents", col="blue")})

Galton famously developed his ideas on correlation and regression using data
which included the heights of parents and children. The dataset galton in the pack-
age UsingR includes data on heights for 928 children and 205 ‘midparents’. Each
parent height is an average of the father’s height and 1.08 times the mother’s height.
The daughters’ heights are also said to have been multiplied by 1.08. Note that there
is one midparent height for each height of a child in this dataset, so that many
midparent heights are repeated. (For anyone interested in investigating a detailed
dataset with full family information, including the sex and order of the children, see
Examining Continuous Variables 31

[Hanley, 2004] and that author’s webpage.) Figure 3.2 shows default histograms of
the two variables.
Both distributions are vaguely symmetric and there appear to be no outliers. It
is difficult to compare the histograms as the scales in the two plots are different.
Although the histograms appear to have different binwidths they are actually the
same, thanks to R looking for sensible breaks, in this case integers, and only drawing
bins for the range of the data. To look for gaps or heaping, it can be useful to draw
histograms with very thin bins. Barcharts would also work, although not for large
datasets with many individual values. Dotplots are good for gaps, although not for
heaping.
From Figure 3.3, histograms with binwidths of 0.1, we can see that only a limited
number of values are used for height (and if you go to the source of this version of
the dataset given in the R help you find that the data were taken from a table, so
individual values are not provided). In both histograms there appear to be narrower
gaps between values at the boundaries. Drawing up tables of the ‘raw’ data confirms
this and reveals the curiosity that the parent values almost all end in .5 while the child
values almost all end in .2. It would be better to use Hanley’s version of the dataset.
The function truehist was used for Figure 3.3, as the binwidth can be set directly.
(hist uses the number of cells.) As it regards histograms as density estimates, the
y-axis scale no longer shows the frequencies.

1.5 2.0

1.5
1.0
1.0

0.5
0.5

0.0 0.0

62 64 66 68 70 72 74 64 66 68 70 72

FIGURE 3.3: Histograms of child and midparent heights drawn with a narrow bin-
width to demonstrate how few height values actually occur. Clearly the data were
aggregated or reported to a limited level of precision. The plots have been drawn
separately, so the scales are different.
par(mfrow=c(1,2), las=1, mar=c(3.1, 4.1, 1.1, 2.1))
with(galton, {
MASS::truehist(child, h=0.1)
MASS::truehist(parent, h=0.1)})
32 Graphical Data Analysis with R

To actually compare the distributions a number of plots could be drawn, for in-
stance parallel boxplots. Keeping with the histogram theme, Figure 3.4 displays the
histograms one above the other with equal scales and binwidths. The x-axis scale
limits were chosen to include the full range of the data and the y-axis limits were
chosen by inspection.
In interpreting the data it should be borne in mind that, as Hanley points out,
Galton obtained the data "through the offer of prizes" for the "best Extracts from
their own Family Records" [Hanley, 2004], so the sample is hardly a random one;
that the data have been rounded for tabulation; and that family sizes vary from 1 to
15, so that one midparent data point occurs 15 times in the dataset and 33 points
occur only once. The medians have been marked by red vertical lines. Although
the means are very similar, the parents’ distribution is clearly less variable. This is
because each parent value is an average of two values. Interestingly (and perhaps √
even suspiciously) the standard deviation of the children, 2.52, is almost exactly 2
times the standard deviation of the parents, 1.79.
You might think that if the individual parent values were available and the genders
of the children were known, as is the case with the full Galton data set available on
Hanley’s webpage, then the height data distributions would be neatly bimodal with
one peak for females and one for males. They are not. Apparently height distributions
are rarely like that, as is discussed in [Schilling et al., 2002], which includes some
nice photographic examples of putative bimodal height distributions constructed by
getting people to stand in lines according to their heights.
Examining Continuous Variables 33

200

150

100

50

0
60 65 70 75
child

200

150

100

50

0
60 65 70 75
parent

FIGURE 3.4: Histograms of child and midparent heights drawn to the same scale
and with medians marked in red. Parents’ heights are spread less and their median is
slightly higher.
c1 <- ggplot(galton, aes(child)) + geom_bar( binwidth=1) +
xlim(60, 75) + ylim(0, 225) + ylab("") +
geom_vline(xintercept=median(galton$child),
col="red")
p1 <- ggplot(galton, aes(parent)) + geom_bar( binwidth=1) +
xlim(60, 75) + ylim(0, 225) + ylab("") +
geom_vline(xintercept=median(galton$parent),
col="red")
grid.arrange(c1, p1)
34 Graphical Data Analysis with R

Some more heights—Pearson

There is another dataset of father and son heights in the UsingR package. This
stems from Karl Pearson and includes 1078 paired heights. His paper with Alice
Lee [Pearson and Lee, 1903] includes a detailed description of how the data were
collected. Families were invited to provide their measurements to the nearest quar-
ter of an inch with the note that “the Professor trusts to the bona fides of each
recorder to send only correct results.” The dataset was used in the well-known text
[Freedman et al., 2007] and a scatterplot of the data is on the book’s cover. As the
data are given to five decimal places and as each one of the father’s heights is a
unique value (even though we know there must have been repeats), someone must
have jittered the data.
The father and son height distributions can be displayed as histograms with
overlaid density estimates (Figure 3.5). Densities can only be successfully over-
laid when the histogram scales are densities instead of frequencies, which is why
y=..density.. is needed. Both distributions look fairly normal, as we might ex-
pect, and, if anything, an initial look would suggest that the heights of the sons look
more normal than the heights of the fathers.

Sons Fathers

0.15 0.15

0.10 0.10

0.05 0.05

0.00 0.00
60 65 70 75 80 60 65 70 75 80
ht (inches) ht (inches)

FIGURE 3.5: Histograms and overlaid density estimates of the heights of fathers and
sons from the father.son dataset in the UsingR package. Both distributions look fairly
normal. For comparison purposes the plots have been common-scaled.
data(father.son, package="UsingR")
c2 <- ggplot(father.son, aes(sheight)) +
geom_histogram(aes(y = ..density..), binwidth=1) +
geom_density() + xlim(58, 80) + ylim(0, 0.16) +
xlab("ht (inches)") + ylab("") + ggtitle("Sons")
p2 <- ggplot(father.son, aes(fheight)) +
geom_histogram(aes(y = ..density..), binwidth=1) +
geom_density() + xlim(58, 80) + ylim(0, 0.16) +
xlab("ht (inches)") + ylab("") +
ggtitle("Fathers")
grid.arrange(c2, p2, nrow = 1)
Examining Continuous Variables 35

You have to be careful about jumping to conclusions here, or as Pearson put it

in his 1903 paper with Lee: “It is almost in vain that one enters a protest against the
mere graphical representation of goodness of fit, now that we have an exact measure
of it.” Whether we would regard Pearson’s test as an “exact measure” nowadays is
neither here nor there. For the time, he was justified in his polemically demanding
both graphical and analytic approaches. Since those early days, analytic approaches
have often been too dominant, both are needed.
If you really want to look at normality, then Q-Q plots are best (Figure 3.6).
These suggest that the fathers’ heights are more normal than the sons’! Comparing
these graphical impressions with some tests (shapiro.test for Shapiro-Wilk and
the further five normality tests in nortest) gives a range of p-values for the fathers
from 0.42 to 0.20 (so none are significant) and for the sons from 0.07 to 0.01 (where
four of the six tests are significant).

par(mfrow=c(1,2), las=1, mar=c(3.1, 4.1, 1.1, 2.1))

with(father.son, {
qqnorm(sheight, main="Sons", xlab="",
ylab="", pch=16, ylim=c(55,80))
qqline(sheight)
qqnorm(fheight, main="Fathers", xlab="",
ylab="", pch=16, ylim=c(55,80))
qqline(fheight)})

Sons Fathers
80 80
! !
!!!!
!!
75 75
!
!
!
!! !
!!
!
!
!
! !!!!!
!
!
!!
!
!!
!
! !!
!!
!
!!
! !
!!
!
!!
!
!
!
!
!!
!
! !!
!
!!
!
!
!!
!
!
!
!!
!
!
! !
!
!!
!
!!
!
!
!
!
!
!!
!
!
!! !
!
!
!!
!
!!
!
!
! !
70 70
!!
!
!
!! !
!
!
!!
!
!
!!
!
!!
!
! !!
!
!
!!
!
!
!
!!
!
!
!!
!
! !!
!
!
!!
!
!
!!
!
!
!!
!
! !
!
!
!!
!
!
!
!!
!
!
!!
!
! !
!
!
!!
!
!
!
!!
!
!
!!
! !
!
!
!!
!
!
!
!
!
!!
!
!
!!
! !
!
!!
!
!
!
!!
!
!
!! !
!!
!
!
!!
! !
!
65 65
!
!
!!
!
!
! !
!
!
!!
!
!
!
!
!
!!
!
!! !
!
!
!!
!
!
!!
!
!! !!
!
!
!!
!
!
!
!!
!
!!
!
! !
!
!
!!
!
!
!
!!
! !!
!
!
!!
!
!
!
!! !!
!
!
!!
!
!
!
! !
!!
!
!
!!
!
! !
!
!!
!
!
!
60 60
! !!
!
!!! !!!!
! ! !

55 55

!3 !2 !1 0 1 2 3 !3 !2 !1 0 1 2 3

FIGURE 3.6: Q-Q plots of the heights of fathers and sons from the dataset father.son
with lines added going through the 25% and 75% quantiles. The distribution of the
fathers’ heights now looks more normal than the distribution of the sons’ heights,
because of the deviations in the upper and lower tails for the sons.

It is curious that these two famous datasets on heights of parents and children,
galton and father.son, have been altered in opposite ways: Some heights in galton
have been rounded and some heights in father.son have been jittered. It is always
better to make the raw data available, even if they need to be amended, ‘just in case’.
36 Graphical Data Analysis with R

Scottish hill races (best times)

The hills dataset in the package MASS contains record times from 1984 for 35
Scottish hill races. The total height gained and the race distance are also in-
cluded. The dataset is well known, as it is used in the introductory chapter of
[Venables and Ripley, 2002]. The DAAG package includes some later hills data with
more races in various datasets. Figure 3.7 shows a boxplot of the record times for the
races in the original dataset. Four races required much longer times than the others
and the distribution is skewed to the right (look at the position of the median in the
box). If you look at three histograms produced by some default histograms, you will
find that the truehist default is poor, the hist default is a little better, although
it still does not pick out the outliers clearly, while the qplot histogram (a so-called
quick option in ggplot2) does direct your attention to the outliers, while at the same
time being too detailed for the races with faster record times. Formulating default
settings for histograms is hard and it is simpler and more informative to draw a range
of histograms with different settings.

par(mfrow=c(1,1), mar=c(3.1, 4.1, 1.1, 2.1))

with(MASS::hills,
boxplot(time, horizontal=TRUE, pch=16, ylim=c(0, 220)))

! ! ! !

0 50 100 150 200

FIGURE 3.7: A boxplot of the record times for the hills dataset. The distribution is
skew to the right with a few high outliers, races which must have been more demand-
ing than the others.

How are the variables in the Boston dataset distributed?

The Boston housing data is a dataset from 1978 that has been analysed many times.
There are two versions in R, the original one in MASS and a corrected one in DAAG,
in which 5 median house values have been ‘corrected’. There are 14 variables for 506
areas in and around Boston. The main interest lies in the median values of owner-
occupied homes by area and here we will use the original dataset. Default histograms
(drawn with either hist(medv) or truehist(medv)) hint that there might be
some interesting structure and that other binwidths or graphics might be useful, while
Figure 3.8, drawn with ggplot, highlights two features: there are surprisingly many
Examining Continuous Variables 37

ggplot(MASS::Boston, aes(medv)) + geom_bar() + ylab("") +

xlab("Median housing value (thousands of dollars)")

40

20

10 20 30 40 50
Median housing value (thousands of dollars)

FIGURE 3.8: A histogram of the median values of owner-occupied homes in the

Boston dataset. The large frequency of the final bin is unusual and there appears to
be a sharp decline in the middle.

areas in the final bin and there is a sudden drop in the counts round about the middle.
A binwidth matching the data units, say 1 or 2, would be better.
At this stage you really need to examine the data directly, either with a table
or with a histogram with a narrow binwidth. A table (table(medv)) is not so
helpful in this case, but at least it tells us that the data are reported to one deci-
mal place (in thousands of dollars). A histogram with a binwidth of 1 (say with
truehist(medv, h=1)) confirms that there are very many values in the last bin
at 50, but does not tell us much more about the break around 25.
In a full analysis it would now be a good idea to look at the rest of the dataset
to see if features observed in the median values variable are related to features in
other variables. That will be done in a later chapter, but in the meantime consider
Figure 3.9, which displays histograms of all the variables. You can readily see the
variety of possible forms a histogram may take and thinking about what each might
mean is the subject of Exercise 2 at the end of the chapter. The binary variable chas
has been included, not because histograms are good for binary variables, but because
it is coded numerically. It is easier to leave it in, and the histogram does show the
information that only a few areas have a river boundary.
The code needs some explanation. The melt function creates a new dataset with
all the data in one variable called BostonValues and a second variable called
BostonVars defining which of the original variables a value comes from. The
second line of code draws a set of histograms, one for each of the original variables,
using facetting, where the option scales="free" ensures that each display is
individually scaled.
38 Graphical Data Analysis with R

crim zn indus chas

300 300 400

100
300
200 200
50 200
100 100
100
0 0 0 0
0 25 50 75 0 25 50 75 100 0 10 20 30 0.00 0.25 0.50 0.75 1.00
nox rm age dis
50 80 60
40 60
60
30 40
40
40
20
20 20
10 20

0 0 0 0
0.4 0.5 0.6 0.7 0.8 0.9 4 5 6 7 8 9 0 25 50 75 100 5 10
rad tax ptratio black
150
200
100 100
100 150

100
50 50 50
50

0 0 0 0
0 5 10 15 20 25 200 400 600 12.5 15.0 17.5 20.0 22.5 0 100 200 300 400
lstat medv
40

30 40

20
20
10

0 0
0 10 20 30 40 10 20 30 40 50

FIGURE 3.9: Histograms of 14 variables from the Boston housing dataset. There are
several different histogram forms, each telling a separate story. Default binwidths,
dividing each variable’s range by 30, have been used. Other scalings could reveal
more information and would be more interpretable.
library(tidyr)
B2 <- gather(MASS::Boston, BosVars, BosValues, crim:medv)
ggplot(B2, aes(BosValues)) + geom_histogram() + xlab("") +
ylab("") + facet_wrap(~ BosVars, scales = "free")

Plots like these are not ideal. In particular, the default scaling of 30 bins works
better for some than for others. Nevertheless, this is a quick and easy way to get
an overview of the data and you can always redraw any individual plots which you
think deserve a closer look. Note that the vertical scales vary from maxima of 40 to
over 400. If you plot the histograms individually, choosing binwidths and scale limits
are the main decisions to be taken. Occasionally the anchorpoint, where the first bin
starts, and whether the bins are open to the left or to the right can make a difference.
The latter becomes an issue if many points lie on bin boundaries. Compare the two
Examining Continuous Variables 39

histograms produced by default by hist (open to the left) and truehist (open to
the right) for the variable ptratio.

with(Boston, hist(ptratio))
with(Boston, truehist(ptratio))

Histograms of datasets with rounded values need to be checked for these effects. And
in case you are wondering, ggplot is open to the left like hist.
It is worth considering what other plots of the variable medv might show. Here
are some you could look at:

• Boxplots

boxplot(Boston$medv, pch=16)

• Jitterered dotplots

stripchart(Boston$medv, method="jitter", pch=16)

• Stem and leaf plots

stem(Boston$medv)

• Average shifted histograms

library(ash)
plot(ash1(bin1(Boston$medv, nbin=50)), type="l")

• Density estimates with a rugplot

d1 <- density(Boston$medv)
plot(d1, ylim=c(0,0.08))
rug(Boston$medv)
lines(density(Boston$medv, d1$bw/2), col="green")
lines(density(Boston$medv, d1$bw/5), col="blue")

Some of these work much better than others in revealing features in the data. There
is no optimal answer for how you find out information, it is only important that you
find it. Note that what density estimates show depends greatly on the bandwidth used,
just as what histograms show depends on the binwidth used, although histograms also
depend on their anchorpoint. One graphic may work best for you and another may
work best for someone else. Be prepared to use several alternatives.
40 Graphical Data Analysis with R

Hidalgo stamps thickness

The dataset Hidalgo1872 is available in the package MMST, which accompanies
the textbook “Modern Multivariate Statistical Techniques” [Izenman, 2008]. The
dataset was first discussed in detail in 1988 in [Izenman and Sommer, 1988]. A keen
stamp collector called Walton von Winkle had bought several collections of Mexican
stamps from 1872-1874 and measured the thickness of all of them. The thickness of
paper used apparently affects the value of the stamps to collectors, and Izenman’s in-
terest was in looking at the dataset as a mixture of distributions. There are 485 stamps
in the dataset and for some purposes the stamps may be divided into two groups (the
years 1872 and 1873-4). We shall examine the full set here first.
The aim is to investigate what paper thicknesses may have been used, assum-
ing that each shows some kind of variability. Figure 3.10 displays a histogram
with a binwidth of 0.001 (the measurements were recorded to a thousandth of a
millimeter) and two density estimates, one using the bandwidth selected by the
density function and one using the direct plug-in bandwidth, dpik, described
in [Wand and Jones, 1995], used in the kde function of package ks.
The histogram suggests there could be many modes, seven or even eight, depend-
ing on how you interpret the three peaks to the left of the distribution. The density
estimate from density suggests there are two, and the estimate from kde suggests
perhaps seven. In Izenman and Sommer’s 1988 paper they found seven modes with
a nonparametric approach and three with a parametric one. This is a complex issue
and it underlines the value of looking at more than one display.
 80

60

40
Examining Continuous Variables

20

0.06 0.08 0.10 0.12

FIGURE 3.10: A histogram and two density estimates of stamp thicknesses from the Hidalgo1872 dataset. Both the histogram and the
red density estimate suggest there are at least 5 modes, implying different production runs or locations.
library(KernSmooth)
data(Hidalgo1872, package="MMST")
par(las=1, mar=c(3.1, 4.1, 1.1, 2.1))
with(Hidalgo1872, {
hist(thickness,breaks=seq(0.055,0.135,0.001), freq=FALSE, main="", col="bisque2", ylab="")
lines(density(thickness), lwd=2)
ks1 <- bkde(thickness, bandwidth=dpik(thickness))
lines(ks1, col="red", lty=5, lwd=2)})
41
42 Graphical Data Analysis with R

How long is a movie?

Most of the datasets in R are not very big. Examples provided by package authors are
more for illustration and for showing how methods work than for carrying out full
scale data analyses of large datasets. The effort involved in preparing and making
available a large dataset in a usable form should never be underestimated.
Nevertheless there are some largish datasets to be found, for instance the dataset
movies in ggplot2 with 58, 788 cases and 24 variables (which on no account should
be confused with the small dataset of the same name in UsingR with 25 cases and
4 variables, nor indeed with the larger more up-to-date version with 130, 456 films
in the package bigvis). Strangely, rather modestly, the help page says there are only
28, 819 cases. One of the variables is the movie length in minutes and it is interest-
ing to look at this variable in some detail, starting with the default histogram (Fig-
ure 3.11) and boxplot (Figure 3.12) using ggplot2.

ggplot(movies, aes(length)) + geom_bar() + ylab("")

60000

40000

20000

0
0 2000 4000
length

FIGURE 3.11: A histogram of film length from the movies dataset. There must be at
least one extreme outlier to the right distorting the scale.

The histogram is hopeless, except that it does imply there must be at least one
very high value to the extreme right, even if the resolution of the plot is not good
enough to see it (with over 58, 000 cases in the single bar on the left, one case is
never going to be visible on its own, a typical problem for large datasets).
The boxplot does what boxplots do best, telling us something more about the
outliers. It appears there are two particularly gross ones, one implying a film length
of over three and a half days and the other with a length of about two days. Although
you would think these would have to be errors, it is always best to check, if possible.
With the help of something like

s1 <- filter(movies, length > 2000)

print(s1[, c("length", "title")], row.names=FALSE)
Examining Continuous Variables 43

ggplot(movies, aes("var", length)) + geom_boxplot() +

xlab("") + scale_x_discrete(breaks=NULL) + coord_flip()

!
!
!
! !
!!
!!
!
!!
!!
!
!
!!
!!
!
!
!!
!
!
!!
!
!!
!
!
!!
!
!
!!
!
!!!!
!
!!!
!!!! ! ! ! ! !

0 1000 2000 3000 4000 5000

length

FIGURE 3.12: A boxplot of film length from the movies dataset. (An artificial x
aesthetic, here "var", is needed when drawing single boxplots with ggplot2.) There
are two very long films, longer than two days, and many that are much longer than
the average film.

you can find out the titles of the films and their length and then check if they really
exist by Googling them. It turns out they do! Slightly bizarrely, the second longest
film in the dataset is called “The Longest Most Meaningless Movie in the World”. It
must have been a shock for the makers to discover that their film wasn’t. Incidentally,
this version of the dataset is no longer up-to-date and there are now some even longer
films.
Clearly the extreme outliers should be ignored, and for exploring the main distri-
bution of movie lengths it makes sense to set some kind of upper limit. Over 99% are
less than three hours in length, so we will restrict ourselves to them. Having got rid
of the outliers, another boxplot does not make a lot of sense and a stem and leaf plot
or a dotplot (whether jittered or not) is clearly out of the question. A density estimate
could be interesting, but might miss possible heaping or gaps, if these features are
present. Figure 3.13 shows a histogram with the natural binwidth of 1 minute. (Any
other value might miss some feature of the raw data and with such a large dataset
there can be no concerns about having too many bins, you just have to ensure that
the plot window is wide enough.) Several features now stand out:
1. There are few long films (i.e., over two and a half hours).
2. There is a distinct group of short films with a pronounced peak at a length of 7
minutes.
3. There is, unsurprisingly, a sharp peak at a length of 90 minutes. Perhaps we
would have expected an even sharper peak.
4. There is clear evidence of round numbers being favoured; as well as a length of
90 minutes, you also find 80, 85, 95, 100, and so on.
44 Graphical Data Analysis with R

3000

2000

1000

0 50 100 150
Movie lengths in minutes

FIGURE 3.13: A histogram of film lengths up to 3 hours from the movies dataset.
The most frequent value is 90 minutes and values are often rounded to the nearest 5
minutes. The most frequent time for short films is 7 minutes.
ggplot(movies, aes(x = length)) + xlim(0,180) +
geom_histogram(binwidth=1) +
xlab("Movie lengths in minutes") + ylab("")

3.4 Comparing distributions by subgroups

When a population is made up of several different groups, it is informative to com-
pare the distribution of a variable across the groups. Boxplots are nearly always best
for this, as they make such an efficient use of the space available. Figure 3.1 showed
the voting support of Die Linke party in Germany and claimed that the form was due
to regional differences. Figure 3.14 shows boxplots of the support by Bundesland and
you can see that apart from Berlin, which divides into East and West, the boxplots
split into two distinct groups, basically the new Bundesländer plus Saarland (higher
support) and the old Bundesländer minus Saarland (lower support). Although the
boxplots convey a lot of information in a limited space, they have two disadvantages:
If a distribution is not unimodal (e.g., Berlin), you can’t see that, and it is difficult to
convey information on how big the different groups are. Neither is serious, but you
need to be aware of them. The option varwidth sets the boxplot widths propor-
tional to the square root of the number of cases. Another alternative is to draw an
accompanying barchart for the numbers in each group (cf. Figure 4.1).
Figure 3.14 shows a range of possible boxplot shapes. There are outliers above
and below, whiskers of different lengths, and asymmetries of various kinds.
 40

30

20 !
Examining Continuous Variables

! !
!
!
10 !
!

BW BY BE BB HB HH HE MV NI NW RP SL SN ST SH TH
Bundesland

FIGURE 3.14: Boxplots of Die Linke support by Bundesland. Standard abbreviations for the Bundesländer have been added to the
dataset to make the labels readable. In the old East (BB, MV, SN, ST, TH) Die Linke were relatively strong, in the old West they were
weak, apart from in Saarland (SL). Berlin (BE), made up of East and West, straddles both groups. Boxplot widths are a function of
Bundesland size.
btw2009 <- within(btw2009, Bundesland <- state)
btw2009 <- within(btw2009, levels(Bundesland) <- c("BW", "BY", "BE", "BB",
"HB", "HH", "HE", "MV", "NI", "NW","RP", "SL", "SN", "ST", "SH", "TH"))
ggplot(btw2009, aes(Bundesland, Linke2)) + geom_boxplot(varwidth=TRUE) + ylab("")
45
46 Graphical Data Analysis with R

3.5 What plots are there for individual continuous variables?

To display continuous data graphically you could use a
histogram grouping data into intervals, and drawing a bar for each interval, shows
the empirical distribution.
boxplot displaying individual outliers and robust statistics for the data, useful for
identifying outliers and for comparisons of distributions across subgroups.
dotplot plotting each point individually as a dot, good for spotting gaps in the data.
rugplot plotting each point individually as a line, often used as an additional plot
along the horizontal axis of another display.
density estimate plotting an estimated density of the variable’s distribution, so
more like a model than a data display.
distribution estimate showing the estimated distribution function, useful for com-
paring distributions, if one is always ‘ahead’ of another.

Q-Q plot comparing the distribution to a theoretical one (most often a normal dis-
tribution).
And there are other possibilities too (e.g., frequency polygon, P-P plot, average
shifted histogram, shorth plot, beanplot).
R’s default for plot is to draw a scatterplot of the variable against the case index.
This can be useful (e.g., showing if the data have been sorted in increasing order or
that the first few values or last few values are different from the others), mostly it
is not. Different analysts may favour different kinds of displays, for instance I like
histograms and boxplots. Pronounced features will probably be visible in all plots.
For more subtle effects the best approach in exploratory analysis is to draw a
variety of plots. There is some general advice to follow, such as histograms being
poor for small datasets, dotplots being poor for large datasets and boxplots obscuring
multimodality, but it is surprising how often even apparently inappropriate graphics
can still reveal information. The most important advice remains—which is why it is
now repeated—to draw a variety of plots.
If data are highly skewed it may be sensible to consider transforming them, per-
haps using a Box-Cox transformation. Graphical displays can help you appraise the
effectiveness of any transformation, but they cannot tell you if they make sense. You
should consider the interpretation of a transformed variable as well as its statistical
properties.
Examining Continuous Variables 47

3.6 Plot options

• Binwidths (and anchorpoints) for histograms
There is an intriguing and impressive literature on the data-driven choice of
binwidths for histograms. [Scott, 1992] and articles by Wand (for example,
[Wand, 1997]) are reliable sources. In practice there are often good reasons for
choosing a particular binwidth that is not optimal in a mathematical sense. The
data may be ages in years, or times in minutes, or distances in miles. Using a
non-integer binwidth may be mathematically satisfying, but can conceal useful
empirical information. It is important to remember that histograms are for pre-
senting data; they are poor density estimators. There are far better approaches for
estimating a possible density generating the data. And it is worth bearing in mind
that methods for determining optimal histogram binwidth assume a given anchor-
point, i.e., the starting point of the first bin. Both display parameters should really
be used for optimisation. In his package ggplot2 Wickham does not attempt to
find any ‘optimal’ choice, but uses 30 bins and prints a message explaining how
to change the binwidth. That is a practical solution.
• Unequal binwidths
When they introduce histograms, some authors like to point out the possibility of
using unequal binwidths. While the idea is theoretically attractive, it is awkward
to apply in practice and the displays are confusing to interpret. If you still want
to do it, consider a variable transformation instead.
• Bandwidth for density estimates
Binwidth is crucial for histograms and bandwidth is crucial for density estimates.
There are many R packages offering different bandwidth formulae and it is not
obvious which to recommend. It is more effective to experiment with a range of
bandwidths. Since you can overlay several density estimates on a single plot, it
is easy to compare them, just use different colours to make them stand out.
• Boxplots
Tukey’s definition of a boxplot distinguishes between outliers (over 1.5 times
the box length away from the box) and extreme outliers (over 3 times the box
length away from the box), while many boxplot displays do not. In fact, there are
frustratingly many different boxplot definitions around, so you should always
confirm which one is used in any plot. Some do not mark outliers, some use
standard deviations instead of robust statistics, there are all sorts of variations.
A set of boxplots in the same window can be either boxplots for the same variable
with one for each subgroup or boxplots for different variables. It is necessary to
know which type you have in front of you. Boxplots by group must have the
same scale and could be drawn with their width a function of the size of the
group. Boxplots of different variables may have different scales and each case
appears in each boxplot (apart from missing values), so that there is no need to
consider different widths.
48 Graphical Data Analysis with R

3.7 Modelling and testing for continuous variables

1. Means
The most common test for continuous data is to test the mean in some way, either
against a standard value, or in comparison to the means of other variables, or by
subsets. Mostly a t-test is used. It would be invidious to select a reference here as
there are so many texts covering the topic. Alternatively medians may be tested,
especially in conjunction with using boxplots.
2. Symmetry
[Zheng and Gastwirth, 2010] discusses several tests of symmetry about an un-
known median and also proposes bootstrapping to improve the power of the tests.
3. Normality
There are a number of tests for normality (e.g., Anderson-Darling, Shapiro-Wilk,
Kolmogorov-Smirnov). These have low power for small samples and may be
rather too powerful for really large samples. A large sample will tend to have
some feature that will lead to rejection of the null hypothesis. There is a book
on testing for normality [Thode Jr., 2002] and there is an R package nortest
which offers five tests to add to the Shapiro-Wilk test offered in the stats package
that comes with R. Tests assess the evidence as to whether there has been some
departure from normality, while graphics, especially Q-Q plots, help identify the
degree and type of departure from normality.
4. Density estimation
There are too many packages in R which offer some form of density estimation
or other for it to be possible to list them all. They fit density estimates, but do
not test. Choose the one (or ones) that you think are good and use it (or them).
Bear in mind that densities for variables with strict boundaries (e.g., no negative
values) need special treatment at the boundaries. At least one of the R packages,
logspline, offers an option for this problem. Most do not.

5. Outliers
The classic book on outliers [Barnett and Lewis, 1994] describes many tests for
outliers, mostly for univariate distributions and individual cases. How useful they
may be depends on the particular application. As the book counsels, you need
to watch out for both masking (one group of outliers prevents you recognising
another) and swamping (mistaking standard observations for outliers).
6. Multimodality
Good and Gaskin introduced the term ‘Bump-Hunting’ for looking for modes
in an oft-cited article [Good and Gaskins, 1980]. The dip test for testing for uni-
modality was proposed in [Hartigan and Hartigan, 1985] and it is available in R
in the appropriately named package diptest.
Examining Continuous Variables 49

Main points
1. There are lots of different features that can arise in the frequency distributions of
single continuous variables (e.g., Figure 3.9).
2. There is no optimal type of plot and no optimal version of a plot type. Look at
several different plots and several different versions of each (cf. §3.3).
3. Natural binwidths based on context are usually a good choice for histograms
(e.g., Figures 3.1 and 3.13).
4. Histograms are good for emphasising features of the raw data, while density es-
timates are better for suggesting underlying models for the data (e.g., Figures 3.5
and 3.10).

5. Boxplots are best for identifying outliers (Figure 3.12) and for comparing distri-
butions across subgroups (Figure 3.14).
50 Graphical Data Analysis with R

Exercises
1. Galaxies
The dataset galaxies in the package MASS contains the velocities of 82 planets.
(a) Draw a histogram, a boxplot, and a density estimate of the data. What in-
formation can you get from each plot?
(b) Experiment with different binwidths for the histogram and different band-
widths for the density estimates. Which choices do you think are best for
conveying the information in the data?
(c) How many plots do you think you need to present the information? Which
one(s)?
2. Boston housing
The dataset is called Boston from the package MASS.
(a) Figure 3.9 displays histograms of the 14 variables. How would you describe
the distributions?
(b) For which variables, if any, might boxplots be better? Why?
3. Student survey
The data come from an old survey of 237 students taking their first statistics
course. The dataset is called survey in the package MASS.
(a) Draw a histogram of student heights and overlay a density estimate of the
data. Is there evidence of bimodality?
(b) Experiment with different binwidths for the histogram and different band-
widths for the density estimates. Which choices do you think are best for
conveying the information in the data?
(c) Compare male and female heights using separate density estimates that are
common scaled and aligned with one another.
4. Movie lengths
The movies dataset in the package ggplot2 was introduced in §3.3.
(a) Amongst other features we mentioned the peaks at 7 minutes and 90 min-
utes. Draw histograms to show whether these peaks existed both before and
after 1980.
(b) One variable, Short says whether a film was classified as a ‘short’ film
(‘1’) or not (‘0’). What plots might you draw to investigate which rule was
used to define ‘short’ and whether the films have been consistently classi-
fied? (Hint: make sure you exclude the high outliers from your plots or you
won’t see anything!)
5. Zuni educational funding
The zuni dataset in the package lawstat seems quite simple. There are three
pieces of information about each of 89 school districts in the U.S. State of New
Mexico: the name of the district, the average revenue per pupil in dollars, and
Examining Continuous Variables 51

the number of pupils. This apparent simplicity hides an interesting story. The
data were used to determine how to allocate substantial amounts of money and
there were intense legal disagreements about how the law should be interpreted
and how the data should be used. Gastwirth was heavily involved and has written
informatively about the case from a statistical point of view, [Gastwirth, 2006]
and [Gastwirth, 2008]. One statistical issue was the rule that before determining
whether district revenues were sufficiently equal, the largest and smallest 5% of
the data should first be deleted.
(a) Are the lowest and highest 5% of the revenue values extreme? Do you
prefer a histogram or a boxplot for showing this?
(b) Having removed the lowest and highest 5% of the cases, draw a density es-
timate of the remaining data and discuss whether the resulting distribution
looks symmetric.
(c) Draw a Q-Q plot for the data after removal of the 5% at each end and com-
ment on whether you would regard the remaining distribution as normal.
6. Non-detectable
The dataset CHAIN from the package mi includes data from a study of 532 HIV
patients in New York. The variable h39b.W1 records, according to the R help
page, ‘log of self reported viral load level at round 6th (0 represents undetectable
level)’. Using the function table(CHAIN$h39b.W1) you can find out that
there are 188 cases with value 0, i.e., with undetectable levels. Further exami-
nation of the dataset (using mi.info(CHAIN), for instance) reveals that ad-
ditionally 179 cases have missing values for h39b.W1. What plots would you
draw to show the distribution of the variable CHAIN$h39b.W1?
(a) with the 0 values?
(b) without the 0 values?
7. Diamonds
The set diamonds from the package ggplot2 includes information on the weight
in carats and price of 53, 940 diamonds.
(a) Is there anything unusual about the distribution of diamond weights? Which
plot do you think shows it best? How might you explain the pattern you
find?
(b) What about the distribution of prices? With a bit of detective work you
ought to be able to uncover at least one unexpected feature. How you dis-
cover it, whether with a histogram, a dotplot, a density estimate, or what-
ever, is unimportant, the important thing is to find it. Having found it, what
plot would you draw to present your results to someone else? Can you think
of an explanation for the feature?
8. Intermission
Albrecht Dürer’s preparatory drawing Praying Hands is in the Albertina Museum
in Vienna. Could the person whose hands are portrayed have worked several
years in a mine? What does that say about the story of the origin of the picture
circulating on the web?
4
Displaying Categorical Data

It’s not what you look at that matters, it’s what you see.

Thoreau

Summary
Chapter 4 looks at visualising single categorical variables, nominal, ordinal, or dis-
crete.

4.1 Introduction
Compared to the range of plots proposed for visualising single continuous variable
data, there have been few suggestions for visualising single categorical variable data.
Barcharts (horizontally or vertically) and piecharts are the commonest options. You
could argue that both are area plots where cases are represented by an area propor-
tional to group size rather than being plotted individually. However, since the bars
in a barchart always have the same width, you compare lengths, not areas, which is
much easier.
Occasionally there are suggestions for using individual points for cases and jit-
tering to keep them apart. This does not work well for high-frequency groups, as it
is hard to assess their densities, and the displays for low-frequency groups may ex-
hibit non-existent patterns due to the random jittering. Nevertheless, as always with
exploratory graphics, if a graphic helps to uncover information, it is worth using.
Whether the graphic that is used to find information is the best way of present-
ing it to others is another matter. Single categorical variables usually contain little
information, but it is still worth looking at some examples before moving on to con-
sider multivariate categorical data and categorical variables together with continuous
variables. There is often information in the simplest plots that is obscured in more
complicated displays.

53
54 Graphical Data Analysis with R

There are sixteen states (Bundesländer) in the German Federal Republic. Fig-
ure 4.1 shows three displays of the states using the number of eligible voters in the
2009 election as a proxy for size. The first display uses alphabetical order. Some
states are much bigger than others, with Nordrhein-Westfalen being the biggest and
Bremen (HB is Hansestadt Bremen) the smallest. The second display orders by size,
and it is easier to judge the full ordering and see that Schleswig-Holstein (SH) is
bigger than Brandenburg (BB). The final display orders by size within East/West
groupings, where Berlin (BE) has been classed as ‘East’, as it is geographically. In
fact, part of it, West Berlin, used to be part of West Germany. It is now clearer that
the states in the East are all roughly the same size, with Sachsen being a bit bigger.
Both the biggest and smallest states are in the West.
The data need some preparation: a new variable is created with abbreviated names
for the Bundesländer; numbers of eligible voters are aggregated by Bundesland (the
data are provided for the 299 constituencies); and a new variable is created according
to whether the Bundesländer were part of the old West or East Germany.

data(btw2009, package = "flexclust")

btw2009 <- within (btw2009, stateA <- state)
btw2009 <- within (btw2009,
levels(stateA) <- c("BW", "BY", "BE",
"BB", "HB", "HH", "HE", "MV", "NI", "NW",
"RP", "SL", "SN", "ST", "SH", "TH"))
Voters <- with(btw2009, size <- tapply(eligible, stateA, sum))
Bundesland <- rownames(Voters)
btw9s <- data.frame(Bundesland, Voters)
btw9s$EW <- c("West")
btw9s[c("BB", "BE", "MV","SN","ST","TH"), "EW"] <- "East"
ls <- with(btw9s, Bundesland[order(EW, -Voters)])
btw9s <- within(btw9s, State1 <- factor(Bundesland, levels=ls))

The first version is best if you are used to that default order and know that, for
instance, Berlin is always in second position. Except that it is not. With the full names
of the states, rather than the standard abbreviations, you get a different ordering and
Berlin is in third position. Consistent orderings are helpful for looking at displays of
several different variables, you just have to agree on the ordering. The other versions
are better for specific information relating to a single variable. Whichever one you
use, they all give a quick overview of the relative sizes of the sixteen states. Barcharts
are good for that. For those who would like an older example, take a look at the top
plot on page 11 of [Gannett, 1898]. It shows a horizontal barchart of the “population
of each state and territory” of the United States in 1890.
The German election data are an example of categorical data, counts by category.
The categories may be nominal with no standard order (like a group of instant coffee
brands) or ordinal (for instance when age is recorded as ‘young’, ‘middle-aged’,
or ‘old’) or discrete (such as the number of children in a family). When the data
are ordinal or discrete, the order must be preserved. When the data are nominal,
reordering the categories becomes an important tool for gaining insights from the
data.
Displaying Categorical Data 55

b1 <- ggplot(btw9s, aes(Bundesland, Voters/1000000)) +

geom_bar(stat="identity") +
ylab("Voters (millions)")
b2 <- ggplot(btw9s, aes(reorder(Bundesland, -Voters),
Voters/1000000)) + geom_bar(stat="identity") +
xlab("Bundesland") + ylab("Voters (millions)")
b3 <- ggplot(btw9s, aes(State1, Voters/1000000)) +
geom_bar(stat="identity") + xlab("Bundesland") +
ylab("Voters (millions)")
grid.arrange(b1, b2, b3)

10
Voters (millions)

0
BB BE BW BY HB HE HH MV NI NW RP SH SL SN ST TH
Bundesland

10
Voters (millions)

0
NW BY BW NI HE SN RP BE SH BB ST TH MV HH SL HB
Bundesland

10
Voters (millions)

0
SN BE BB ST TH MV NW BY BW NI HE RP SH HH SL HB
Bundesland

FIGURE 4.1: Three alternative barcharts of the numbers of eligible voters in the
German Bundesländer. The top one is the default alphabetical order, the middle one
is ordered by decreasing number of voters, and the bottom one is ordered by voter
numbers within the former East/West Germanies. There is great variation in size, and
the biggest Bundesländer are in the area that was formerly West Germany.
56 Graphical Data Analysis with R

4.2 What features might categorical variables have?

There might be
Unexpected patterns of results There may be many more of some categories than
others. Some categories may be missing completely.
Uneven distributions Observational studies may exhibit some form of bias, per-
haps too many males. In medical meta analyses many trials are analysed together,
although it can turn out that most of the trials were small and that the results are
dominated by the one or two major trials.
Extra categories Gender may be recorded as ‘M’ and ‘F’, but also as ‘m’ and ‘f’,
‘male’ and ‘female’. In a study of patients with two medical conditions, there
may be some patients diagnosed with a third condition.
Unbalanced experiments Although experiments are usually carefully designed
and carried out, there is always the chance that some data are missing or un-
usable. It is important to know if this occurs and leads to unequal group sizes.
Large numbers of categories In studies including open-ended questions (e.g.,
“Who is your favourite politician?”) there may be far more names mentioned
than you expected.
Don’t knows, refusals, errors, missings, ... Data may not be available for a wide
variety of reasons, and plots summarising how many cases of each type have
arisen can be helpful both in deciding how to handle the data and in properly
qualifying the results from the data that are available. Opinion polls are an obvi-
ous example.
Displaying Categorical Data 57

4.3 Nominal data—no fixed category order

Meta analyses—how big was each study?
The package meta includes three meta analysis datasets. The dataset Fleiss93 has
details of seven studies on the use of aspirin after myocardial infarction. Figure 4.2
plots a barchart of the study sizes, with the studies ordered by the total numbers of
patients (control group and experimental group together). The differences in size are
striking, with one study being much bigger than all the others.

data(Fleiss93, package="meta")
Fleiss93 <- within(Fleiss93, {
total <- n.e + n.c
st <- reorder(study, -(total)) })
ggplot(Fleiss93, aes(st, total)) + geom_bar(stat="identity") +
xlab("") + ylab("") + ylim(0,20000)

20000

15000

10000

5000

ISIS!2 AMIS MRC!2 CDP MRC!1 PARIS GASP

FIGURE 4.2: A barchart of the sizes of the seven studies in the Fleiss93 meta analysis
dataset. The ISIS-2 study had more patients than all the others put together.

If there were many more than seven studies, and that is the usual situation with meta
analyses, it could make sense to plot individual bars for the biggest studies and a
combined bar for the total of the smaller studies. The following code would draw the
corresponding plot for Fleiss93, combining all studies with fewer than 2000 cases:

Fleiss93 <- within(Fleiss93, st1 <- as.character(study))

Fleiss93$st1[Fleiss93$total < 2000] <- "REST"
ggplot(Fleiss93, aes(st1, total)) + geom_bar(stat="identity") +
xlab("") + ylab("") + ylim(0,20000)
58 Graphical Data Analysis with R

Anorexia
In the anorexia dataset in the MASS package, two treatment groups are compared
with a control group. There are 72 cases and you might assume that they were split
equally into three groups of 24 each. Figure 4.3 shows the curiously uneven distribu-
tion (29 and 17 for the two treatment groups and 26 for the control group). Needless
to say, there were probably very good reasons for the split, but it could be useful to
know. What if the groups were initially of the same size and there were dropouts?

data(anorexia, package="MASS")
ggplot(anorexia, aes(Treat)) + geom_bar() + xlab("Treatment")

30

20
count

10

0
CBT Cont FT
Treatment

FIGURE 4.3: A barchart of the group sizes in the anorexia dataset drawn with
barplot(table(anorexia$Treat)). The groups are not of equal size.

A table would show this just as well in principle

with(anorexia, table(Treat))

but a graphic is more convincing.

Who sailed on the Titanic?

The tragic sinking of the Titanic has been discussed endlessly and there are many
films, books, and webpages about the disaster. Despite all this attention the infor-
mation on those who sailed on the ship is incomplete. A lot is known about some
passengers and members of the crew, especially about the first-class passengers and
the officers; less is known about some of the others on the ship.
Although there is not full agreement on all details, the dataset Titanic is generally
thought to be an accurate summary. It contains four pieces of information on 2201
people who were on board: Class (1st, 2nd, 3rd, or crew), Sex, Age (a binary
variable, Child or Adult), and Survived (whether they survived or died). Interest
centres on issues such as survival by Class and Sex (which will be discussed in
§7.2), but it is sensible to have a look at the simple univariate barcharts first, as in
Figure 4.4.
Displaying Categorical Data 59

Titanic1 <- data.frame(Titanic)

p <- ggplot(Titanic1, aes(weight=Freq)) +
ylab("") + ylim(0,2250)
cs <- p + aes(Class) + geom_bar(fill="blue")
sx <- p + aes(Sex) + geom_bar(fill="green")
ag <- p + aes(Age) + geom_bar(fill="tan2")
su <- p + aes(Survived) + geom_bar(fill="red")
grid.arrange(cs, sx, ag, su, nrow=1, widths=c(3, 2, 2, 2))

2000 2000 2000 2000

1500 1500 1500 1500

1000 1000 1000 1000

500 500 500 500

0 0 0 0
1st 2nd 3rd Crew MaleFemale Child Adult No Yes
Class Sex Age Survived

FIGURE 4.4: Barcharts of the four variables in the Titanic dataset. That the majority
of passengers and crew died and that there were far more men than women on board
is well known, the numbers in the different classes less so.

You should always think about what you expect graphics to show, before you
draw them. That way you can be surprised by what you see and value more the infor-
mation presented. Once you have seen Figure 4.4, it seems obvious that there were
more people in the crew than in any of the other classes and that there were fewest
in the second class. That there were more males than females will have surprised no
one, but that there were over three times as many does seem high, probably because
the crew were almost all men. The small number of children may raise questions
about the quality of the data for the Age variable. The fact that overall twice as many
died as survived is well known.
There are technical points worth mentioning. The dataset is supplied as four con-
ditional contingency tables, so it has been converted to a dataframe. A common ver-
tical scale has been used for all four plots to make comparisons across plots possible.
This is fine for the three binary variables, but not so good for the variable Class, a
typical issue when categorical variables have different numbers of categories. (The
upper limit was chosen after drawing default plots. More formally, the maximum
should be determined by calculation.)
The Class barchart has been drawn wider than the others, as it has more cate-
gories. This makes the code more complicated, but it is good to have the flexibility
to do it. For presentation purposes it would have been more elegant to make all bars
the same width. For exploratory purposes this would be gilding the lily.
60 Graphical Data Analysis with R

Opinion polls
Nowadays we are bombarded with results of opinion polls on all kinds of subjects,
and political opinion polls get special emphasis for obvious reasons. Although there
are always people who fail to understand the question, who do not want to choose
any of the options offered, or who just refuse to answer pollsters’ questions, the
numbers of ‘Don’t knows’, or ‘Undecideds’ or whatever you choose to call them, are
seldom reported. For political polls this group can be important, as they could sway
the election if they did indeed decide to vote.
The referendum on Scottish independence held in 2014 looked like it would be
very close just before voting took place. In the end the margin was fairly substantial
(55.3% to 44.7%). Afterwards some commentators suggested that this should have
been predictable, as the ‘Don’t knows’ in opinion polls in referenda have tended to
vote for the status quo [FiveThirtyEight, 2014].
A poll taken in Ireland in August 2013 shows how much difference neglecting
the ‘Don’t knows’ can make. Unusually, perhaps because of the relatively high pro-
portion of ‘Don’t knows’, the Sunday Independent newspaper presented the results
both with and without the ‘Don’t knows’ [Sunday Independent, 2013]. They used
piecharts, and Figure 4.5 shows something similar drawn with R using the reported
percentages. Altogether, 985 people were questioned, 4 of whom said they would
vote for the Green party. Rounding the percentages, as was done by the newspaper,
meant that the Green party was omitted completely from the published piechart with
the ‘Don’t knows’, but was shown in the second piechart.
The top plot, including the ‘Don’t knows’, emphasizes how open the election
was likely to be, whereas the lower plot suggests that Fianna Fail and Sinn Fein
might have been able to form a government together (should such a coalition have
been in the realms of possibility). Piecharts are good for emphasizing when data are
shares of some whole, and can be helpful, as here, for helping to estimate possible
coalition shares. You have to know what the shares are shares of. In the upper plot
it is shares of all respondents, in the lower one it is shares of respondents who said
whom they would support. Piecharts are poor for comparing values. Can you tell
whether Fine Gael or Fianna Fail had more support in the survey? As the two major
parties historically, it is very important to them which one is ahead. In this case, Fine
Gael edged it by around 1%.
The piecharts could probably be improved with a better choice of colour shades
and something close to the standard party colours should be chosen. It could be help-
ful to also report the values, either on the plots or in accompanying tables. Piecharts
are not generally recommended (see, for example, the R help page for piecharts,
where Bill Cleveland is quoted on the comparatively poor accuracy of judgements of
angles) and should mostly be avoided. Piecharts in 3D should most definitely always
be avoided.
Displaying Categorical Data 61

Party <- c("Fine Gael", "Labour", "Fianna Fail",

"Sinn Fein", "Indeps", "Green", "Don’t know")
nos <- c(181, 51, 171, 119, 91, 4, 368)
IrOP <- data.frame(Party, nos)
IrOP <- within(IrOP, {
percwith <- nos/sum(nos)
percnot <- nos/sum(nos[-7])})
par(mfrow=c(2,1), mar = c(2.1, 2.1, 2.1, 2.1))
with(IrOP, pie(percwith, labels=Party, clockwise=TRUE,
col=c("blue", "red", "darkgreen", "black",
"grey", "lightgreen", "white"), radius=1))
with(IrOP, pie(percnot[-7], labels=Party, clockwise=TRUE,
col=c("blue", "red", "darkgreen", "black",
"grey", "lightgreen"), radius=1))

Fine Gael

Don't know
Labour

Fianna Fail

Green
Indeps
Sinn Fein

Green
Indeps

Fine Gael

Sinn Fein

Labour

Fianna Fail

FIGURE 4.5: Piecharts of an Irish political opinion poll from August 2013, one in-
cluding the ‘Don’t knows’ and one not. The top plot emphasizes the large number
of voters who have not decided how they might vote, while the second plot suggests
that a Fianna Fail and Sinn Fein coalition, were such an agreement possible, could
almost have a majority.
62 Graphical Data Analysis with R

4.4 Ordinal data—fixed category order

Surveys
In many surveys there is a string of questions to which respondents give answers on
integer scales from, say, 1 to 5. Sometimes descriptive terms are attached to each
number on the scale, although not always.
The BEPS dataset in the effects package includes seven such questions put to
1525 voters in the British Election Panel Study for 1997-2001. One question was on
a scale from 1 to 11, one from 0 to 3, and the rest were from 1 to 5. Figure 4.6 shows
the barcharts for the voters’ assessment of the leaders of the three main parties at the
time, where a higher value means a better assessment. Few voters used the extremes
of 1 or 5. The leaders of the two main parties, Tony Blair and William Hague, were
either liked (a value of 4 or 5) or disliked (a value of 1 or 2). The middle value of
3 was hardly used for them, although it was used for Charles Kennedy, the leader
of the Liberals. Once you have seen the patterns, there is no difficulty in describing
them and suggesting possible explanations, but would you have expected them?

data("BEPS", package="effects")
a1 <- ggplot(BEPS, aes(factor(Hague))) +
geom_bar(fill="blue") + ylab("") +
xlab("Hague (Conservative)") + ylim(0, 900)
a2 <- ggplot(BEPS, aes(factor(Blair))) +
geom_bar(fill="red") + ylab("") +
xlab("Blair (Labour)") + ylim(0, 900)
a3 <- ggplot(BEPS, aes(factor(Kennedy))) +
geom_bar(fill="yellow") + ylab("") +
xlab("Kennedy (Liberal)") + ylim(0, 900)
grid.arrange(a1, a2, a3, nrow=1)

750 750 750

500 500 500

250 250 250

0 0 0
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5
Hague (Conservative) Blair (Labour) Kennedy (Liberal)

FIGURE 4.6: Assessment of UK party leaders by voters in 2001. Higher values mean
a higher rating. The bars have been filled with the party colours. Opinion seems to
be divided on the leaders of the two main parties, Hague and Blair.
Displaying Categorical Data 63

Two of the variables concerned Europe, one the respondents’ knowledge of the par-
ties’ policies on European integration and one the respondents’ attitudes towards
European integration. What patterns of results might be expected here? Figure 4.7
displays the two barcharts.
Well over half the respondents thought they had good or very good knowledge
of party policies (options 2 and 3), while a sizeable minority thought they had low
knowledge (option 0). Very few chose the option 1. The second barchart plots a scale
based on combining answers to several questions, so it is reasonable to find that
there is a spread of scale values, although that they are represented fairly equally
seems a little surprising. The biggest group was the over 20% who were strongly
against European integration. Given the tenor of the discussions about membership
of the European Union going on in the UK in 2014, the proportion strongly against
European integration now is probably considerably higher than it was at the turn of
the century.

b1 <- ggplot(BEPS, aes(factor(political.knowledge))) +

geom_bar(fill="tan2") + coord_flip() + ylab("") +
xlab("Knowledge of policies on Europe")
b2 <- ggplot(BEPS, aes(factor(Europe))) +
geom_bar(fill="lightgreen") + ylab("") +
xlab("Attitudes to European integration")
grid.arrange(b1, b2, nrow=1, widths=c(4, 8))
Knowledge of policies on Europe

3 300

2
200

1
100

0
0
0 200 400 600 800 1 2 3 4 5 6 7 8 9 10 11
Attitudes to European integration

FIGURE 4.7: UK voters’ attitudes to European integration are shown in the plot on
the right. There is a strongly negative group (those with the value ‘11’), but otherwise
the respondents are roughly equally spread across the scale. The barchart on the left
referring to respondents’ knowledge has been drawn horizontally to differentiate it
from the barchart on the right referring to respondents’ opinions.
64 Graphical Data Analysis with R

And more surveys

The survey dataset from the MASS package gives the results of a survey of 237
Australian statistics students. (It is also used in Exercise 3 of Chapter 3.) Figure 4.8
displays the barcharts for the seven categorical variables and gives an idea of the
range of barcharts that you can find.
For most of the variables a simple line of code suffices, adding informative labels
instead of the variable names. The variables for exercising and smoking have had to
be reordered for the plots and this is something that always has to be watched out for.
Note that the vertical scales vary from about 120 to just over 200.
Interestingly the sexes look exactly equal (a table shows they were), and one per-
son did not give that information. Most students are right-handed (again one person
did not answer the question, not the same one as before, as it turned out). More stu-
dents fold their arms right over left, and a non-negligible minority say that neither
right nor left is on top. With clapping, the majority for the right side is higher and
the ‘neither’ group is bigger. As you might expect, most students do at least some
exercising and few smoke. Many people in Australia still use Imperial units (feet and
inches) to measure height, or rather they did at the time the survey was carried out in
the 1970s.

data(survey, package="MASS")
s1 <- ggplot(survey, aes(Sex)) + geom_bar() + ylab("")
s2 <- ggplot(survey, aes(W.Hnd)) + geom_bar() +
xlab("Writing hand") + ylab("")
s3 <- ggplot(survey, aes(Fold)) + geom_bar() +
xlab("Folding arms: arm on top") + ylab("")
s4 <- ggplot(survey, aes(Clap)) + geom_bar() +
xlab("Clapping: hand on top") + ylab("")
survey <- within(survey,
ExerN <- factor(Exer,
levels=c("None", "Some", "Freq")))
s5 <- ggplot(survey, aes(ExerN)) + geom_bar() +
xlab("Exercise") + ylab("")
s6 <- ggplot(survey, aes(M.I)) + geom_bar() +
xlab("Height units") + ylab("")
survey <- within(survey,
SmokeN <- factor(Smoke,
levels=c("Never", "Occas", "Regul", "Heavy")))
s7 <- ggplot(survey, aes(SmokeN)) + geom_bar() +
xlab("Smoking") + ylab("")
grid.arrange(s1, s2, s3, s4, s5, s6, s7, ncol=3)
Displaying Categorical Data 65

120 125

200
100
90
150
75
60
100
50

30 50 25

0 0 0
Female Male NA Left Right NA L on R Neither R on L
Sex Writing hand Folding arms: arm on top

150 120

90
100 100

60

50 50
30

0 0 0
Left Neither Right NA None Some Freq Imperial Metric NA
Clapping: hand on top Exercise Height units

150

100

50

0
Never Occas Regul Heavy NA
Smoking

FIGURE 4.8: Barcharts of the categorical variables in the survey dataset of 237 statis-
tics students. The individual plots are discussed in the chapter. Barcharts may be of
many different shapes. Each plot has its own vertical axis scale.
66 Graphical Data Analysis with R

4.5 Discrete data—counts and integers

Deaths by horsekicks
Von Bortkiewicz’s dataset about deaths by horsekick in the Prussian army over
twenty years is included in vcd under the name vonBort. Figure 4.9 shows the distri-
bution of the numbers of deaths by corps and year. There is no information provided
on the sizes of the corps over the years, though it is believed that most had four regi-
ments, while three had six and one had eight. There may have been around 750 men
(and horses) in each cavalry regiment.

data(VonBort, package="vcd")
ggplot(VonBort, aes(x=factor(deaths))) + geom_bar() +
xlab("Deaths by horse kick")

150

100
count

50

0
0 1 2 3 4
Deaths by horse kick

FIGURE 4.9: Number of soldiers killed in one of 14 Prussian army corps by year for
twenty years from 1875 to 1894. The distribution is close to a Poisson distribution
with parameter 0.7, suggesting that these might be considered unfortunate random
accidents. The dataset first appeared in [von Bortkiewicz, 1898].

A test of whether a Poisson distribution would be a good fit is not significant.

gf <- goodfit(table(VonBort$deaths))
summary(gf)

#
# Goodness-of-fit test for poisson distribution
#
# X^2 df P(> X^2)
# Likelihood Ratio 2.442786 3 0.4857196
Displaying Categorical Data 67

Looking at the distributions for individual years illustrates a problem that can
arise with data like these. Both plots in Figure 4.10 show the distribution of deaths
by corps for the year 1891. No corps suffered either 2 or 4 deaths and so the plot
on the left ignores those categories. This can be misleading for categories outside
the range of the subset (the value 4 in this case) and is more serious for a missing
category inside the range (the value 2 here). The plot on the right solves the problem
by specifying the levels of the factor deaths.

data(VonBort, package="vcd")
h1 <- ggplot(VonBort[VonBort$year=="1891", ],
aes(x=factor(deaths))) + geom_bar()
h2 <- ggplot(VonBort[VonBort$year=="1891", ],
aes(x=factor(deaths, levels=seq(0, 4)))) +
geom_bar() + scale_x_discrete(drop=FALSE)
grid.arrange(h1, h2, nrow=1)

6 6

4 4
count

count

2 2

0 0

0 1 3 0 1 2 3 4
factor(deaths) factor(deaths, levels = seq(0, 4))

FIGURE 4.10: Numbers of soldiers killed in one of 14 Prussian army corps in 1891.
The plot on the left leaves out values with zero counts, the plot on the right is a more
complete display. With only 14 pieces of data, it is not surprising that there can be
two unusual values (unusual for the year 1891).
68 Graphical Data Analysis with R

Goals in soccer
Scores in many sports lead to discrete data, whether goals in soccer, runs in cricket,
points in rugby, or shots in golf. The dataset UKSoccer in vcd summarizes the results
of season 1995-96 of the English Premier League in a table giving the frequencies
with which specific results occurred. The second line of code converts the table into
a data frame and Figure 4.11 shows the distributions of goals scored by the home and
away teams. If a team scored more than 4 goals, 4 was recorded in the dataset, which
is why the labels have been changed in the code. The vertical scales have been made
the same for comparative purposes.

data(UKSoccer, package="vcd")
PL <- data.frame(UKSoccer)
lx <- c("0","1","2","3","4 or more")
b1 <- ggplot(PL, aes(x=factor(Home), weight=Freq)) +
geom_bar(fill="firebrick1") +
ylab("") + xlab("Home Goals") +
scale_x_discrete(labels=lx) + ylim(0,150)
b2 <- ggplot(PL, aes(x=factor(Away), weight=Freq)) +
geom_bar(fill="cyan1") +
ylab("") + xlab("Away Goals") +
scale_x_discrete(labels=lx) + ylim(0,150)
grid.arrange(b1, b2, nrow=1)

150 150

100 100

50 50

0 0

0 1 2 3 4 or more 0 1 2 3 4 or more
Home Goals Away Goals

FIGURE 4.11: Distributions of the numbers of goals scored by the home and away
teams in the 1995-96 English Premier League season. As you would expect, the
home teams scored more goals. Perhaps it is unexpected that there are relatively few
matches in which either team scores more than two goals.
Displaying Categorical Data 69

Benford’s Law
The observation that the first or leading digits of many sets of numbers do not fol-
low a uniform distribution was known a long time ago. As Simon Newcomb wrote
in his 1881 article [Newcomb, 1881]: “That the ten digits do not occur with equal
frequency must be evident to any one making much use of logarithmic tables, and
noticing how much faster the first pages wear out than the last ones.” However, the
result is known as Benford’s Law, because of his article [Benford, 1938], which de-
scribed the checking of many real datasets he had collected.
There have been several interesting applications in recent years, including studies
of financial fraud and voting irregularities, and references can be found in the relevant
Wikipedia article. The package benford.analysis provides six datasets, including
census.2009, which gives the populations of 19, 509 U.S. towns and cities.
If we only consider the first digit, then there are only nine possible values and the
probability distribution is shown in Figure 4.12:

xx <- 1:9
Ben <- data.frame(xx, pdf=log10(1+1/xx))
ggplot(Ben, aes(factor(xx), weight=pdf)) + geom_bar() +
xlab("") + ylab("") + ylim(0,0.35)

0.3

0.2

0.1

0.0

1 2 3 4 5 6 7 8 9

FIGURE 4.12: The distribution of first digits according to Benford’s law.

70 Graphical Data Analysis with R

4.6 Formats, factors, estimates, and barcharts

Attentive readers will have noticed that drawing graphics of categorical data often
involves preparatory reformatting of the data. Working with categorical data is not
always straightforward and you have to watch both for how the data are stored and
how the categories are coded.

Shape of the dataset

The data may be provided as a collection of individual cases, as a list of the possible
combinations of the categorical variables with their associated frequencies, or as a
set of conditional tables. Depending on which form the data have, some restructuring
may be necessary before drawing particular plots.

Coding of variables
Categorical variable data may be recorded in different ways. Numeric codings were
common in the distant past when computer storage was expensive and limited. Nu-
meric codings have their uses, but there is always the danger of inappropriate calcula-
tions being carried out (e.g., what is the average marital status when ‘1’ = single, ‘2’
= married, ‘3’ = separated, . . . ?), and the implied ordering may need to be changed
too. You can convert the variable to a factor to change the order, but if you are going
to do that, it is probably best to define a new factor variable and replace the codes
with descriptive text. This has the advantage of keeping the original variable: you
never know when you might need it.
If the data are stored as character strings, R assumes the variable is a factor and
the default ordering is alphabetical. Non-alphabetical orderings can be specified us-
ing the levels function. Sometimes this has been done already for datasets avail-
able in R, and you may have to change the ordering, if the default is not what you
need. Reordering should always be carried out with care, since there are too many
ways for things to go wrong.

Estimates shown as bars

In some fields of application estimates are displayed as bars. When confidence inter-
vals are shown as well, the lower part can be hidden by the colouring of the bar and
the upper part looks like the plunger on a dynamite box, which gives rise to the name
‘dynamite plot’. Most statisticians disapprove and recommend using dotplots with
confidence intervals instead [Cleveland, 1993]. In this book there are no dynamite
plots, only barcharts where the bars represent counts or probabilities.
Displaying Categorical Data 71

4.7 Modelling and testing for categorical variables

1. Testing by simulation
χ 2 tests in one form or another are standard. An alternative to relying on the
asymptotic distribution of the test statistic is to simulate many datasets and com-
pare the actual test statistic value with the distribution of simulated values. This
option is available in the R function chisq.test.
2. Evenness of distribution
If numbers are supposed to have been drawn at random, for instance in a lot-
tery, or if a random number generator is to be checked directly, then the null
hypothesis of equally likely probabilities can be carried out with a χ 2 test.

3. Fitting discrete distributions

As always with categorical data, χ 2 tests are important, but it is useful to inspect
any lack of fit visually. Tukey’s hanging rootograms are one approach; plotting
components of χ 2 -statistics is another.

Main points
1. Barcharts are a simple form of display, yet they can provide much information
and often surprise. Errors may become apparent, there may be more categories
than anticipated, category counts may be unexpected (e.g., Figures 4.2 and 4.6).
2. Barcharts can be used for nominal, ordinal, or discrete variables.

3. The order of categories affects the way a barchart of a nominal variable looks, and
different orderings can emphasise different features, as was shown in Figure 4.1.
4. People have strong opinions about piecharts. Don’t let that stop you using them
if you like them, just make sure they show clearly what you want them to. They
are useful for displaying shares, as in Figure 4.5.
72 Graphical Data Analysis with R

Exercises
1. Gastrointestinal damage
The dataset is called Lanza from the package HSAUR2.
(a) Data on four studies are reported. Draw a plot to show whether all four
studies are equally large.
(b) The outcome is measured by the variable classification with scores
of 1 (best) to 5 (worst). How would you describe the distribution?
2. Alzheimer’s
Three groups of patients (one with Alzheimer’s, one with other forms of demen-
tia, and a control group with other diagnoses) were studied. Counts are given in
the dataset alzheimer in the package coin. Prepare a graphical summary of plots
of each of the three variables, smoking, disease, and gender, in a single
window.
(a) Are the disease groups very different in size?
(b) Are there more men or women in the study?
(c) How would you describe the distribution of the smoking variable? Do you
think the smoking data are likely to be reliable?
3. Slot machines
According to the R help entry, this dataset (vlt in package DAAG) was collected
from the three windows of a video lottery terminal while playing the game ‘Dou-
ble Diamond’. There are seven possible symbols that may appear in each win-
dow. Draw equally scaled barcharts to see if the distributions of frequencies are
the same for each window. Describe any important features.
4. Multiple sclerosis
The dataset MSPatients in vcd provides information on the diagnoses of two
neurologists from two different cities on two groups of patients, one from
each city. How do the distributions of the ratings of the neurologists compare?
How would you describe their rating patterns? Draw two barcharts with com-
mon scaling. Before drawing the barcharts, reorder the categories into a sensi-
ble ordinal scale instead of the default alphabetical order. (Hint: initially using
data.frame(as.table(MSPatients)) will put the dataset into a form
that is easier to work with.)
Displaying Categorical Data 73

5. Occupational mobility
According to the R help page, the Yamaguchi87 dataset in vcdExtra has become
a classic for models comparing two-way mobility tables.
(a) How do the distributions of occupations of the sons in the three countries
compare?
(b) How do the distributions of the sons’ and fathers’ occupations in the UK
compare?
(c) Are you surprised by the results or are they what you would have expected?
6. Whisky
The package bayesm includes the dataset Scotch, which reports which brands of
whisky 2218 respondents consumed in the previous year.
(a) Draw a barchart of the number of respondents per brand. What ordering of
the brands do you think is best?
(b) There are 20 named brands and a further category Other.brands. That
entails drawing a lot of bars. If you decided to plot only the biggest brands
individually and group the rest all together in the ‘Other’ group, what cut-
off would you use for defining a big brand?
(c) Another version of the dataset called whiskey is given in the package
flexmix. It is made up of two data frames, whiskey with the basic data, and
whiskey_brands with information on whether the whiskeys are blends or
single malts. How would you incorporate this information in your graphics,
by using colour, by using a different ordering, or by drawing two graphics
rather than one?
(d) Which of the spellings, ‘whisky’ or ‘whiskey’, is more appropriate for this
dataset?

7. Choice of school
The dataset GSOEP9402 in the package AER provides data on 675 14-year-old
children in Germany. The data come from the German Socio-Economic Panel
for the years 1994 to 2002.
(a) Which variables are nominal, ordinal, or discrete?
(b) Draw barcharts for the variables. Are any similar in form, and what expla-
nations would you suggest for these similarities?
(c) The variable meducation refers to the mother’s educational level in
years. Would you describe it as ordinal or discrete, and how should it be
displayed?
(d) Summarise briefly the main information shown by your graphics.
74 Graphical Data Analysis with R

8. Election results
The Bavarian election of Autumn 2013 was a triumph for the CSU party, as they
obtained an absolute majority for the first time for several years. This was partly
because some parties failed to get 5% of the votes or more and could therefore
not win any seats in parliament. A few observers commented that the real winners
were the group who did not vote at all, as they made up a higher proportion of
the electorate than the CSU supporters. The percentages reported were as follows
(where the party percentages refer to their share of the actual vote): ‘didn’t vote’
36.7%, CSU 47.7%, SPD 20.6%, FW 9.0%, Grüne 8.6%, FDP 3.3%, BP 2.1%,
Linke 2.1%, ÖDP 2.0%, Pirates 2.0%, Rest 2.5%.
(a) Draw graphics to present these results both with and without the group of
non-voters. What headlines would you give your graphics if you were to
publish them in a newspaper?
(b) It is, of course, not necessary to list all the parties, especially when several
of them were excluded by the 5% rule. Which ones would you leave out
and why?
(c) Seats are won by the parties with over 5% of the vote, based on their share
of the total votes of qualifying parties. The rules are actually more compli-
cated than that, but these complications can be neglected here. There are
180 seats in all. Draw a graphic showing the seat distribution, using the
available figures. What headline would you suggest for this graphic?
(d) You have probably noticed that the percentages do not add up to 100%. Is
this a problem and what might you do about it?
9. Horsekicks
In his discussion of the deaths due to horsekicks in the Prussian army, von
Bortkiewicz pointed out that four of the corps (G, I, VI, and XI) were bigger
than the others.
(a) Draw two plots, one for the numbers of deaths each year for the four big-
ger corps and one for the other corps. What vertical scale do you think is
appropriate for comparing the two plots?
(b) Estimate the parameter of a Poisson distribution for the other corps. Carry
out a test to see if a Poisson distribution with that λ would be an acceptable
fit for the four bigger corps.

10. Intermission
Edward Hopper’s Nighthawks hangs in the Art Insitute of Chicago. What are the
main features of this painting and is it typical of Hopper’s work?
5
Looking for Structure: Dependency
Relationships and Associations

‘The world is full of obvious things which nobody by any chance ever ob-
serves.’

Sherlock Holmes (in Sir A. Conan Doyle’s The Hound of the Baskervilles)

Summary
Chapter 5 is about examining how pairs of continuous variables are related.

5.1 Introduction
Drawing scatterplots is one of the first things statisticians do when looking at
datasets. Scatterplots can reveal structure that is not readily apparent from summary
statistics or from models, and they are relatively easy to understand and present. They
are the basis of the Gapminder displays [Rosling, 2013] used so effectively by Hans
Rosling to draw attention to patterns of world development.
The major role of scatterplots lies in revealing associations between variables, not
just linear associations, but any kind of association. Scatterplots are also useful for
identifying outliers and for spotting distributional features. However, marginal dis-
tributions cannot always be clearly seen from scatterplots, so this book recommends
you always take a look at the univariate distributions briefly as well.

75
76 Graphical Data Analysis with R

Over ten thousand athletes competed in the London Summer Olympics of 2012.
Figure 5.1 shows a scatterplot of Weight against Height from the dataset oly12
in the package VGAMdata. Note that 1346 of the 10, 384 athletes in the dataset are
not displayed in the plot because one or other of the two measurements is missing.
The expected relationship between weight and height can be clearly seen, although it
is a little obscured by some outliers, which distort the scales, and drawing a second
plot with tighter limits would show the bulk of the data better. There is evidence of
discretisation in the height measurements (notice the parallel vertical lines) and the
same effect would be visible for the weight measurements but for the outliers. Given
the large number of points, there is also a lot of overplotting, and most of the points
in the middle of the plot represent more than one case (there are 57 athletes who are
1.7m in height and weigh 60kg). You can alleviate the problem using alpha-blending,
giving each point a weight equal to the parameter alpha in geom_point, although
outliers are then less easy to see.

data(oly12, package="VGAMdata")
ggplot(oly12, aes(Height, Weight)) + geom_point() +
ggtitle("Athletes at the London Olympics 2012")

Athletes at the London Olympics 2012

200

!
!
! ! !
! ! !! !
150 !
! !
!
! ! ! ! !
! ! ! !
! !! !! !
! !
Weight

!! !!!!! !
! !!
!!
! ! !!! !! !! !!! ! !!
! !!! ! !
! !
!!!!! !! !!
! ! !!! !
!! ! ! !!!
!!! !!!!!!!!!
!! ! !!! !
! !!
! ! !
!! !!! !
!
!!
!!
!!!
! !
! ! ! !
!
!!! !!
! ! !! !
!! !
!!! ! !!!!!
! ! ! !!
!!!!! !! !
!
!!!!
! !! !
!
! !!! !!!
!! !!!!
!!
!
!!!!
!!
!
!!!!!
!!
! !!!! !!
!!!!! ! !
! !
! !!
! !
!!! ! !
! ! !! !
!! ! !!
!!!
!!! !!! !
!! !!!
!!
!! !
!!!!
!!
!!!
!! !
!! !!
!!
!!
! !
!!
!!
!!
!!
!!
!!! !
! !!
100
!
!!
!!! !!! ! !!
! !! !!
! ! ! !!
!!!
!!!
!!
!!!!!!! !
!
!!! !!
!
!!!!!!
!
!!
! !
! !!!!
!!
!!!!!
!! !!!! !
!!!!
! !
!!!! !!
! !!!! !!
!!
!!!!
!!
! !!
!
!!!!
!!
!!!!!
!!
! !
!
!!!!
!
!!
! !
!!!
!
!
!!!
!
!!!
!
!! !
!! !
!!!
!
!! !
!!
! !
!!
!!!
!!
!
!!!
!
!
!!!
!!!
!!!
!
!!
!!
!
!!
!
!!
!!
!
!!
!!
!
!!
!!
! !
!
!!
!
!!
!!
!!
!!! ! !
!! ! ! !!! !
! !!!
!!
! !!
!! !!
!! !!!
!
!
!! !!
!!
!!
!!!!!
!!
!
!!!!!
!
!!!
! !!
!!
!!!
!!!!!!!!
!!!
!!!
! !!
!! !!
!!!
! !!!!!
!!
!! !!!
!!!
!!!
!!
!!!
!!!!
!!!
!! !
!!
!!!!
!!
!!
!!
!!
!!!!
!!! !!!
!!
!!! !!
!!
!! !
!!
! ! !!!
!!
!!!!
!!!!
!!
!!!
!!!
!
!!
!!
!!
!!
!!
!!
!!
!
!!!
! !!
!!
!
! !!
!
!!
! !!!
!!
! ! !
!!! !
!!!
!!!!!!!!
!!!
!! !!
!!
!!
!
!!!
!
!!
!
!!
!!
!!
!
!!
!!
!
!
!
!!!
!
!!!
!
!!
!!
!
!!
!
!
!
!!!
!
!
!
!
!
!
!
!
!
!
!
!!
!!!
!!
!!
!!!
!!!
!!!!
!
!! !! !
! ! !!!! !
!!
!
!
!
!!
!! !!
!
!!
!!
! !
!!
!
!!
!
!!!
!
!!
!
!!
!!!
!
!!
!
!!
!
!
!
!
!
!
!
!
!
!
!!!
!
!
!!
!!
!
!
!!
!
!
!
!!
!!
!!!!!
!
!
!!
!!
!
!!
!
!
!!
!!
!!
!!
!
!
!!
!
!
!!
!!
!
!
!!!
!
!!
!
! !! !
!
! ! ! !
! !
!!
!!
!!
! !
!!
!!!
!!
!
!!!
!
!!
!!
!!
!
!!
!!
!!
!
!!
!!
!!!
!
!!
!!!! ! ! !
! !!
! !!!!!
!!
!!! !!!!
!!!!!
!! !!
!!
!! !!
!! !
!!
! !
!!!
!!
! !!
!
!! !!
!!
! !!
!
!!!!!
!!
! !
!!
!!
! !!!!
!!
!! ! !
!! !
! !
!!!! ! !!! !
! !!! !
!!
!!!!!
!
!
!!!
!!
!!
!!
!
!
!
!!!
!
!!
!
!
!!!
!
!
!!!
!!
!
!!
!
!!
! !
!
!
!
!
!
!
!
!!
!!
!!
!!
!!!
!
!!!!
!!
!!! ! ! !
! !!
! !
!!!
! !!
!
!!
! !
!!
! !!!
! !!
!!! !!
!!!
!!!
!!
!!
! !
!!
!
!!
!!
!! ! !!!
!
!!! !!
!
!
!! !
!!
!!!
!
!!
!!
!! !
!!
!!
!!
!!!
!!
!!
!
!!
!
!!
!
!
!
!
!!
!
!!
!
!!
!
!!
!!!!
!!
!
!!!
!
!!
!
!!
!!
!
!
!!
!!
!!
!
!!
!!!
!!
!
!
!!
!
!!
!
!
!!
!
!!
!! !!!!
!!
!!! !
!!
!
! !
!!
! ! !!
!!
!!!
!
!!
!
!
!
!!
!!!
!
!
!!!
!!!
!!
!
!!
!!
!
!!
!
!! !!
!!
!!
!
!!
!!
!
!
!!
!
!
!
!!
!
!
!!
! !
!
!
!
!
!
!
!
!!!
!
!
!
!
!!
!!
!
!
!
!!!!
!
!!
!!!
!
!!
!
!
! !!! !
!!
!! !
!!!!
!!
!!
!!
!!!
! !!
!!
!!
!!!!
! !
!!
!!
!!!
!! !
!
!!
!!
!
!!
!!
!!
!!
! !
!!!
! !
! ! ! !
!! ! ! !
! ! !
!
!
!!!!
!!
!! !
!
!
!!
!
!
!
!!
!
!
!
!
!!
!!
!
!
!
!
!!!
!
!
!
! !
!
!!
!!
!!
!
!
!
!
!
!
!
!
!!
!!
!!
!!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!!
!!
!!
!
!
!
!
!
!!
!!
!
!
!
!!
!
!
!
!!
!
!
!
!
!
!
!!
!
!
!
!
!!
!
!!
!
!!
!!! !! !
!! ! ! !
50
! ! !!!
!!!!
! !!
!!!!
!!
!!
!!
! !!!
!!!!
!!!!
! !
!!
! !
!!
!!!!!
!! !!
!!
! !!! !
!!
! !
!!!!
!
!!! !
!
!!
!! !
!
!!
!!
! !!
!!!
!!!
!!
!!!!!
! !!
!!!
! !!
! ! ! ! ! !
! ! !!! !!
!!!! !!
!!
!!!
!!!!!
!!!
!
!!
!
!!
!
!!
!!!!
!!!!!
!!!!
! !!
! !
! !
!!!!
! !
!
!!
!!!!
!!
! !!
! !!!!!!
!! !! !
!
!
!!
!!
! !
!!!! !
! !

1.50 1.75 2.00 2.25

Height

FIGURE 5.1: A scatterplot of weight against height for the athletes at the London
Olympics 2012. You can see that weight increases with height. There are several
outliers which affect the scales.
Looking for Structure: Dependency Relationships and Associations 77

5.2 What features might be visible in scatterplots?

Features that might be found in a scatterplot include
Causal relationships (linear and nonlinear) One variable may have a direct influ-
ence on another in some way. For example, people with more experience tend to
get paid more. It is standard to put the dependent variable on the vertical axis.
Sometimes initial ideas are misleading and X is more sensibly seen as depending
on Y and not the other way round.
Associations Variables may be associated with one another without being directly
causally related. Children get good marks in English and in Maths because they
are intelligent, not because ability in one subject is the reason for the ability in
the other.
Outliers or groups of outliers Cases can be outliers in two dimensions without be-
ing outliers in either dimension separately. Taller people are generally heavier,
but there may be people of moderate height who are so heavy or light for their
height that they stand out in comparison with the rest of the population.

Clusters Sometimes there are groups of cases which are separate from the rest of
the data. In a scatterplot of the two petal measurements for Fisher’s iris dataset,
Figure 1.4, you can see that the setosa flowers have much lower values than the
other two varieties.
Gaps Occasionally, particular combinations of values do not occur. Movies which
are rated highly are rated often, but movies which few people like are seldom
rated, so in a plot of average rating against number of ratings the bottom right of
the plot is empty, Figure 5.7.
Barriers Some combinations of values may not be possible. No one can have more
years of experience than their age, and a plot of the two variables will have a
linear boundary, which should not be crossed.

Conditional relationship Sometimes the relationship between two variables is bet-

ter summarised by a conditional description than by a function. A plot of income
against age is likely to differ before and after retirement age.
The following two sections show a range of different scatterplots for a number of
different applications.
78 Graphical Data Analysis with R

5.3 Looking at pairs of continuous variables

The evils of drink?
Karl Pearson investigated the influence of drink on various aspects of family life at
the beginning of the twentieth century. The dataset DrinksWages is in the collection
of historic datasets made available in the package HistData. Figure 5.2 shows a
scatterplot of wage (average weekly wage in shillings) plotted against drinks/n
(the proportion of drinkers) for 70 different trades. There is no apparent relationship.

data(DrinksWages, package="HistData")
ggplot(DrinksWages, aes(drinks/n, wage)) + geom_point() +
xlab("Proportion of drinkers") + xlim(0,1) + ylim(0,40)

40 !
!
!
! ! !
!
! !
!
!
! !
! ! !
! ! ! !
30 ! ! ! !
! !
! ! ! !
!
! ! ! !
! ! !
!
! ! !
! ! !
!
! ! !
!
wage

! ! !
20 !
!
!
!
!

! ! !
! !
!

!
!

10

0.00 0.25 0.50 0.75 1.00

Proportion of drinkers

FIGURE 5.2: A scatterplot of average wages against proportion of drinkers for all
70 trades in Pearson’s DrinksWages dataset in the HistData package. There are a
surprising number of trades where either all workers are drinkers or none are.

It is perhaps initially surprising that some trades have 100% drinkers and some
100% non-drinkers. A look at the distribution of numbers in the study by trade (or
indeed a table) explains why, as you can check for yourself using one of these alter-
natives:

with(DrinksWages, hist(n, breaks=0:max(n)))

with(DrinksWages, table(n))

Of the 70 trades, over a third (26) have only 1 or 2 members in the survey. The
biggest group with 100% drinkers (7 drinkers) turns out to be the chimney sweeps
and the biggest temperance group the gasworkers (5 non-drinkers):
Looking for Structure: Dependency Relationships and Associations 79

with(DrinksWages, max(n[drinks==0]))
with(DrinksWages, trade[drinks==0 & n==max(n[drinks==0])])
with(DrinksWages, max(n[sober==0]))
with(DrinksWages, trade[sober==0 & n==max(n[sober==0])])

Excluding the smaller trades (say, all with less than five) gives Figure 5.3. There
is no particular pattern and certainly no evidence of drink being associated with lower
average wages. (Whether this is the best way to investigate this question is another
matter.) Note that the scales in Figures 5.2 and 5.3 have been explicitly made equal
to keep them comparable.

bigDW <- filter(DrinksWages, n > 4)

ggplot(bigDW, aes(drinks/n, wage)) + geom_point() +
xlab("Proportion of drinkers") + xlim(0,1) + ylim(0,40)

40
!

!
! !
!
!
! !
30 ! !
! !
!
! !
! !
!
!
!
! !
!
wage

! !
20 !

! !
!

10

0.00 0.25 0.50 0.75 1.00

Proportion of drinkers

FIGURE 5.3: A scatterplot of average wages against proportion of drinkers for all
trades with a group size of more than 4. There is no obvious relationship between the
two.
80 Graphical Data Analysis with R

Old Faithful
The Old Faithful geyser in Yellowstone National Park in Wyoming is a famous tourist
attraction. The dataset geyser in MASS provides 299 observations of the duration of
eruptions and the time to the next eruption. Figure 5.4 plots the data.

data(geyser, package="MASS")
ggplot(geyser, aes(duration, waiting)) + geom_point()

100
!
!!
!
!! ! ! ! ! !
! ! !
! !
! !
!
! !! ! ! !
! ! ! !! ! !
! !!! !! ! ! ! ! !
! ! ! ! !
! ! !! ! !
!!!! ! ! ! ! !!
!!!! ! !
! ! ! ! !
80
!!! !! ! ! ! ! ! !
! !!! !! ! ! !!
!
waiting

!! ! ! ! !! !
!!! !!! ! ! ! ! !
!!! ! !! !
!!! ! ! !! !
! ! ! ! ! ! !
!! ! ! ! !!
! ! ! ! !!
! ! !! !
! ! !! !
!
! ! ! !
! !!
!
! ! ! ! !! !
!
!
! ! ! !!
60
! ! !
! ! !! ! !
! !
! ! !
! ! ! ! ! ! !
! !
! ! !
! !!
! !!! ! !
! ! ! ! !
!! ! !
! !!
! !
! ! ! !! !!! ! !
! !
! ! ! !
! ! !
! !
!
!

40
1 2 3 4 5
duration

FIGURE 5.4: A scatterplot of the waiting time to the next eruption vs. the duration of
the current eruption for the Old Faithful geyser in Yellowstone National Park. After
an eruption of short duration, you have to wait longer for the next one.

A short duration implies a long waiting time until the next eruption, while a long
duration can imply a short or long waiting time. There may be 3 clusters and possibly
a couple of outlying values, but there is also a suggestion of rounded values for the
eruption durations (note the numbers of durations of 2 and 4 minutes). To assess the
possibility of clustering, consider a density estimate. Figure 5.5 displays contours of
a bivariate density estimate supporting the idea of there being three clusters.
Contour plots of density estimates show equal levels of the estimated density
function, but are not associated with probabilities. Using the hdrcde package you
can estimate highest density regions, the smallest areas not including specified pro-
portions of the distribution. Figure 5.6 displays the geyser data in this way for the
proportions 1%, 5%, 50%, 75%. This suggests pretty much the same outliers as be-
fore, but does not support the three concentrations conclusion as strongly.
The original source of this dataset is not given on the R Help page. The age of
the dataset and the apparently rounded values suggest that better quality data might
be obtainable now. The Wikipedia page on Old Faithful is not consistent with these
data at all [Wikipedia, 2013]:
With a margin of error of 10 minutes, Old Faithful will erupt 65 minutes after
an eruption lasting less than 2.5 minutes or 91 minutes after an eruption
lasting more than 2.5 minutes.
Looking for Structure: Dependency Relationships and Associations 81

ggplot(geyser, aes(duration, waiting)) + geom_point() +

geom_density2d()

100
!
!!
!
!! ! ! ! ! !
! ! !
! !
! !
!
! !! ! ! !
! ! ! !! ! !
! !!! !! ! ! ! ! !
! ! ! ! !
! ! !! ! !
!!!! ! ! ! ! !!
!!!! ! !
! ! ! ! !
80
!!! !! ! ! ! ! ! !
! !!! !! ! ! !!
!
waiting

!
! ! ! ! !! !
!!! !! ! ! ! ! ! !
!!! ! !! !
!! ! ! ! !! !
! ! ! ! ! ! !
!! ! ! ! !!
! ! ! ! !!
! ! !! !
! ! !! !
!
! ! ! !
! !!
!
! ! ! ! !! !
!
!
! ! ! !!
60
! ! !
! ! !! ! !
! !
! ! !
! ! ! ! ! ! !
! !
! ! !
! !!
! !!! ! !
! ! ! ! !
!! ! !
! !
!! !
! ! ! !! !!! ! !
! !
! ! ! !
! ! !
! !
!
!

40
1 2 3 4 5
duration

FIGURE 5.5: The same scatterplot with bivariate density estimate contours. There is
evidence of three concentrations of data, two univariate outliers (one eruption with
low duration and one with a high waiting time until the next eruption), and one bi-
variate outlier.

library(hdrcde)
par(mar=c(3.1, 4.1, 1.1, 2.1))
with(geyser, hdr.boxplot.2d(duration, waiting,
show.points=TRUE, prob=c(0.01,0.05,0.5,0.75)))

!
100

!
!!
!
!! ! ! ! ! !
! ! !
! !
! !
!
! !! ! ! !
! ! !!! ! !
! !!! !! ! ! ! ! !
! ! ! ! !
! !!! ! !
!!!! ! ! ! ! !!
!!!!! !
!! !!
! ! ! ! !
80

! ! ! ! ! ! ! !
! !! ! !! ! !
! !!
!
! ! ! ! !! !
!!! !! ! ! ! ! ! !
!!! ! !! !
!!! ! ! !! !
! ! ! ! ! ! !
!
! ! ! ! !!
! ! ! ! !!
! ! !! !
! ! !! !
!
! ! ! !
! !!
!
! ! ! ! !! !
!
!
! ! ! !!
60

! ! !
! ! !! ! !
! !
! ! !
! ! ! ! ! ! !
! !
! ! !
! !!
! !!! ! !
! ! ! ! !
!! ! !
! !
!! !
! ! ! !! !!! ! !
! !
! ! ! !
! ! !
! !
!
!
40

1 2 3 4 5

FIGURE 5.6: Another version of the scatterplot, but now with highest density regions
based on a bivariate density estimate. There is less evidence of three data concentra-
tions than in the previous plot and there is a slightly different set of possible outliers.
82 Graphical Data Analysis with R

Movie ratings
The distribution of movie lengths in the dataset movies from ggplot2 was discussed
in §3.3. Here we are going to look at two other variables, rating (average IMDb
user rating) and votes (number of people who rated the movie), using a scatterplot.

ggplot(movies, aes(votes, rating)) + geom_point() + ylim(1,10)

!
!
!
!
!
!
!
!
!
!
!
! ! !
!
! !
!
! ! !
!
! ! ! ! ! ! !! !
!
!! ! ! ! ! !! ! !!
!
! !! ! ! ! ! ! ! !
!
!
! !!!
! !! ! !! ! ! ! !! !
!
!
!!!! !!! !!! !!!!!!! ! !!! !! ! !! ! !
!!! !!
!
! !!!! ! !! !
!!! ! ! !! !! !!!
! ! ! !
!
!
!!!!!!
!!!
!!!!!
!!!!
! !! !!!! !! ! ! !! ! ! ! ! !
!
!
!!
!!!!! !!!!!! !
! ! !!! !!! !! ! ! !!! !! ! !
!
!
!!
!!
!!!
!!
!!!!!
! !!!!!
! !! !!! !! ! !! ! ! !! ! !! ! !
!
!
!!
!!
!!
! !!!
!!
!!!!
!!! !!! !!! !! ! ! ! ! !! ! ! !! !
!
!
!!
!!
!
!!!
!!! !!
! !
!!!
!!!!
!!!!!
! !! !! !!!!! !!! ! !! !!
! ! !
!
!
!!
!!
!
!!!
!!
!!
!!
!!!!!!
! !!!
!!!!!!!! !!!!! !!
! !!!!!! ! !!
!!
!
!
!!
!
!!
!!
!!!!!
!
!!
!!! !!!!!!!!!!!!
!!
!!! ! !
! !
!!! !
!!!!!!
! !
!
!
!!
!!
!
!!!
!!
!!!!
! !!
!!!!!!
! ! !!!!!!!!!
!!!! ! !! ! !! !
!
!
!
!!
!!
!
!!!
!!
!!!
!
!!
!!!!
!!!! !
!!!!! !
!!!!
!! !!!! ! !! ! ! !
!
!
!!!
!!
!!!
!!
!!
!!
!!
!!!! ! !
!! !!!!!!
!!!! ! !
!!!!! ! !! ! !
!
!
!!
!
!!
!!!
!!!!
!!
!
!!!!
!!! !!!
!!!! ! ! !!!
!! ! ! ! !
!
!!
!!
!!!
!
!!!
!!
!!!
!!
!!!!
!!!!
! !!!! !! !
!!!!!!!!! ! !! !!
!
!
!!
!
!!
!!
!!!
!!
!!
!!!!
!!
!!! !!
! !!
!!! !! ! !!!!!!!! ! ! ! ! !
!
!
!!
!
!!
!!
!!!
!!!
!
!!!!!
!!!!!!! ! ! ! !!
!
!
!!
!!
!
!!!
!!
!!!
!!
!
!!!!
!
!!!!!!! !!
!!!!
! !!!!
!! !!! !! ! ! !
!
!
!!
!!
!
!!!!!
!!
!!
! !!
!!
!!!!!!!
! !!!!!!
!
! !!!
! !! !
!
!
!!
!
!!
!!
!!!
!!
!!!
!!!!
!
!!
!!!!
! !!!!!!! !!! !!!! !
!!
!
!
!!
!!
!!
!
!!!
!!
!!!
!
!!!!!
!!! !
! !!!!! ! ! ! !!
!
!!
!
!!
!!
!
!!!
!!
!!!
!!!!!
!!
!!
!!!!!
! !!! !!!!
!! !! !
!
!
!!
!
!!
!!!
!!!
!!
!!!
!
! !!!
! !!!! ! !!! !! ! ! !
!
!
!!
!!
!
!!
!!!
!!!
!
!!!!!!
!!
!!!!!
! ! ! !!!!!!! !
!
!
!!
!
!!
!!!
!!
!!!
!!!!!!
! !!! !!!!! !! !! ! ! ! !
!
!
!!
!!
!!
!!
!!
!!
!!!
!
!!
!!!!
!!
!! !!! !! !!
!
!
!!
!
!!
!!!!
!!!!! !!
! !!!! !!!! ! ! !!
a ng

!!
!!
!
!!!
!
!!!
!!!!
!!
!!!!!!!
! !
!!!!!! !
! ! !
!
!!
!!
!!
!!
!!!
!!
!!!
!!!!!!! !!! !!! ! ! ! ! !
!
!
!!
!!
!
!!!
!!!!
!!!!
! !!!!!! !! !
!
!
!!
!!
!!
!!!!!!
! !!!! !!!!! !! ! !
!
!
!!
!!
!!
!!!!!
!!
!!
!!!
!!!
! !!! !! !
! ! !
!
!
!!
!!
!
!!!
!!!!
!!!!
! !!!!!!! ! !
!
!
!!
!!
!!
!!!
!!!!!!
! !!!! !
!
!
!!
!!
!
!!!!
!!!
!!
!!!! !
!
!
!!
!!
!!!
!!
!!!!!! !!
!
!
!!
!!
!!
!!!
!!! ! !
!
!
!!
!!
!!
!!!
!!!!
!! !! !
!
!
!!
!!
!!!
!!!!! !!
!! !
!
!
!!
!!
!!!!!
! !!
!!! !!!!
!
!
!!
!!
!
!!!!!!!
!!
!
!
!!
!!
!!!!!!!! !
!
!
!!
!!!
!!!!
!!
!
!
!!
!!
!!!!! !
!
!
!!
!!!! ! !
!
!
!!
!!!! !!!
!
!
!!
!!!
!!!!!
!
!
!!
!!
!!! ! ! !
!
!
!!!
!! !!
!
!
!
!!!
!!!!!
! !
!
!
!!!! ! !!
!
!
!!! !! ! !!
!
!
!!!! !
!
!
!!
!!!! !!
!
!
!!!! !!
!
!
!!!!!!
!
!
!!! !!
!
!
!! !!
!
!!
!
!
!
!!!! !
!
!!!!
!
!
!!!! !
!
!
!!!!! ! !
!
! !
!
!!! !
!
!
!
!!
!
!
!!!
!
!!!!
!! !
!
!!
! !
!
!
!
!
!
!
!
!
!
!

vo es

FIGURE 5 7 A sca erp o of he average ra ngs of fi ms by he number of peo-

p e who vo ed from he da ase mov es F ms ra ed very of en have h gher average
ra ngs bu he h ghes ra ngs are ach eved by fi ms ha are ra ed far ess of en

F gure 5 7 shows he somewha unexpec ed resu The sca erp o a mos akes he
form of a sma e er ‘r’ There are many ns gh s ha can be ga ned from h s p o
1 There are no fi ms w h o s of vo es and a ow average ra ng
2 For fi ms w h more han a sma number of vo es he average ra ng ncreases
w h he number of vo es
3 No fi m w h o s of vo es has an average ra ng c ose o he max mum poss b e
There a mos seems o be a barr er wh ch canno be crossed
4 A few fi ms w h a h gh number of vo es over 50 000 ook ke ou ers They
have a d s nc y ower ra ng han o her fi ms w h s m ar numbers of vo es
5 F ms w h a ow number of vo es may have any average ra ng from he owes
o he h ghes
6 The on y fi ms w h very h gh average ra ngs are fi ms w h re a ve y few vo es
Sca erp o s can revea a o of nforma on
Looking for Structure: Dependency Relationships and Associations 83

5.4 Adding models: lines and smooths

People are good at spotting patterns that are not really there (and are doubtless also
good at missing some patterns that really are there). If you think there is a linear
causal relationship between the two variables in a scatterplot, it makes sense to fit a
model and to add it to the display.

Cars and mpg

The dataset Cars93 from MASS contains 27 pieces of information for 93 cars.
The data were collected around twenty years ago [Lock, 1993]. Plotting MPG.city
against Weight clearly shows a nonlinear relationship, as fuel economy decreases
with weight quite quickly initially and then more slowly (Figure 5.8).

data(Cars93, package="MASS")
ggplot(Cars93, aes(Weight, MPG.city)) + geom_point() +
geom_smooth(colour="green") + ylim(0,50)

50

40 !

!
!
! !
30 !!!!
!
!
MPG.city

!!
! !
! ! ! ! !
! !! !
!
! ! ! !! ! !
! ! !! ! ! !
! ! ! !
20 ! !! !
!
! !! !
! !!
!!!! !
! !! !! !! ! !
! ! !
! ! !! !
! ! !
! !

10

2000 2500 3000 3500 4000

Weight

FIGURE 5.8: A scatterplot of MPG.city against Weight for the Cars93 dataset.
A smooth has been overlaid. Heavier cars get progressively fewer miles to the gallon.
The car weighing about 2350 pounds with a relatively high mpg was the Honda Civic.
The y axis has been drawn from 0 to emphasise the lower limit of the mpg value.
84 Graphical Data Analysis with R

Pearson heights
Figure 5.9 shows the fathers and sons height data of Pearson (already discussed in
Section 3.3) with the best linear fit added, together with a 95% pointwise confidence
interval. The y = x diagonal has been drawn for comparative purposes.

data(father.son, package="UsingR")
ggplot(father.son, aes(fheight, sheight)) + geom_point() +
geom_smooth(method="lm", colour="red") +
geom_abline(slope=1, intercept=0)

! !

! ! !
!
!
!
!! ! !
!
75 ! !
! !
! !
! !! ! !! !
!
! !! ! !! !! ! ! !! ! !
!
! ! ! ! !!!
!
!
! !
! !!! !!! !! ! ! !!! !
!!
! ! !! ! ! !
! ! !! ! ! !! ! !! ! ! ! ! !
! ! ! ! !!!! ! ! !! !
! ! ! ! ! !! !! !! ! !!! !
! ! !!! ! !! ! !! ! !! !!!
! !
!! !!!! ! ! ! !!!!! ! !! ! !!!
!! ! !! ! !!
!!
!
! !
! !! !! ! ! !!! ! !!
! !
!!!! ! !! !! ! ! !
! !! !!!! !! !!!!! !! ! !!! ! !! ! ! ! ! !
!! ! ! !! !!
! !!!!!!! ! ! ! ! ! ! !
! ! ! !! ! !! ! !! !! !! !! !! !!!!! !!
! !
!!! ! ! ! !
70 ! !! ! ! !!!!
!!
!!
!
!!! !!!!! ! ! ! ! !!! !!
!! ! !
! ! !! ! !
! ! ! !!!!! ! !!!!! !!
!!!!! !! ! !
! ! ! !! !!!!!! !!! !! !!! !! ! !!!!!
sheight

! ! ! !
!! ! !!! !!!! !! !! !!!!
!!!! !
!!
!
!!
!!!! ! !! ! !! !!!!
! ! !! !
! ! !! !!!! !!! ! ! !
!
!!!!!!!
!
!! !
!!
!
!!
!!!
!
!!!!!!!
!
!!!!
! ! !
!!! !! ! ! !! ! !!
! ! ! !
!!! !!
!
! !!! !!!!!!
! !
!
! ! !! !
! ! !
! ! ! !! ! !!!
!! !! ! !
!!!
!! ! !!!! !!! ! ! !
! ! !! !!!!!!!
!! !!!! ! !!! !! ! ! ! !!!
! ! !!! !! ! !!! ! !!!!!!!
!
!! ! ! !!
! !
!! !!!!
! !
!!!
!! !! ! ! !
!
!! ! !! ! !! !!
! !! !!
!! !!!
!
! !!!! ! ! ! ! !
!! ! !!! !!
! !!!!!! !!!!!!! ! !!!! !!
!!!!!!! !
! !
!!!
!!!!! ! !
!
!!!!! !
! !!! !!! !!! ! !
! ! !
!
!!
!
!!!
! !
!!
!
! !
!
!
!
!
!!
!!! !!!!! !!
! !
! !!!!! !!
!! !!
!
! !
!! ! !
!
!!! !! ! !
! ! !!! !
! ! !! !
! ! !! !
!! ! !!! !
!! !! !! !!
! ! ! ! ! !
!
! !! !
!!!!! ! !! ! ! ! !! !! !
!!
! ! !
! ! ! ! ! !!!! ! !! ! !! ! !
! ! ! ! !
!
!! ! !! !!!!! ! !!! ! !
!! !! !! ! !!
!! ! !!!! ! !!
65
! ! !! !
! !! !! !! ! !!
!! ! ! ! ! !! ! ! ! ! ! !
!
! ! !! ! ! !
! ! !! !! !! ! ! !! ! !
!
! ! !!! ! ! ! ! !
!! !!! !! !! ! !
! ! ! ! ! ! !!
! ! !!
! ! ! !
!
! !
! ! !
! !
!! !
!

60 !
!
!

!
!

60 65 70 75
fheight

FIGURE 5.9: A scatterplot of sons’ heights against fathers’ heights from the dataset
father.son. The best fit regression line has a slope of just over 0.5, as can be seen by
comparison with the line y = x. The height of a man is influenced by the height of his
father, but there is a lot of unexplained variability (the correlation is almost exactly
0.5).

Figure 5.9 illustrates regression to the mean. Tall fathers have sons who are tall, but
on average not as tall as their fathers. Small fathers have sons who are small, but on
average not as small as their fathers. The fit of the model can be examined with

data(father.son, package="UsingR")
m1 <- lm(sheight~fheight, father.son)
summary(m1)
par(mfrow=c(2,2))
plot(m1)
Looking for Structure: Dependency Relationships and Associations 85

To explore further whether a non-linear model might be warranted, you could fit a
smoother or plot a smoother and the best fit regression line together. For these data,
a linear model is fine, since the two curves are practically identical, as you can see in
Figure 5.10.

data(father.son, package="UsingR")
ggplot(father.son, aes(fheight, sheight)) + geom_point() +
geom_smooth(method="lm", colour="red", se=FALSE) +
stat_smooth()

! !

! ! !
!
!
!
!! ! !
!
75 ! !
! !
!
! !! ! ! !! ! !
! !! ! !! !! ! ! !! ! !
!! !! ! ! !!! !
! !
! !
!! !! !! ! ! !!! !
!!! !! ! !! ! !! ! !
! ! ! !
! ! ! ! !! ! ! ! ! ! !
! ! ! ! !! ! ! !
! ! ! ! ! !! !! ! !
! ! ! ! !!
! ! !!! ! !! ! !! ! !! !!!
! !
!! !!!! ! ! ! !!!!! ! !! ! !!! !
! !
! !!!
! !
! !! !! ! ! !!! !! !! !
!!! !!!
! !!! !!! ! !! !
! !
! !! !!!! !! !!!!! !!!! !!! ! ! !! ! !
!
! ! ! !! !!! !!!! !!! ! ! !! ! !
!
!
! ! ! !!! ! ! !! ! !!! ! !! !! ! !
!
! !
70
! ! !
!!!!!
!!! ! ! ! !
!!!! ! !!!
!!!! ! ! !! !! !!!!
! !
!!
!
!
!
!
!!!
!!!! !!
!! ! !!! !! ! ! !
!!! !! ! !! ! !
! ! ! !!
!! !!!!!! !! ! ! !! !!!! !
sheight

! ! ! !
!! ! !!! !!!! !! !! !!!!
!!!!
!!
!
!!
!
!!!!!! ! !! ! !! !!!!
! ! !! !
! ! !! !!!! !!! ! ! !! !!!!!! ! !
!! !!!!!!
!
!!!!
! ! !
! ! !!! ! ! !!!! !
!!
!!
!!!!
!!
!!!
! !
! !
! !!
! !!!
!!
!! !! !!!
!! !! ! !!!!!! !
!!
!!
!
!!!
!
! ! ! !
! !
!!
!! !! !
!! !!! !
!! !
! !!! !!! !!
!! !! ! !!! !! ! !!!
! ! !!! !! ! !!! !!!!
! !!
!
!!!!!!
! ! ! !!!! !!!! !!
!
!! ! !! ! !! !!
! !! ! !
!!! ! !! !!! ! !! ! ! !!
! !!!!!! !!!!!!! !!! ! !! !!!
!!!!!!
!!!!!!! ! ! !
!! ! !!! !! ! ! !
!! !
! !
!!
!!!
!!! !
!!! !
!
!
!!
!
!!!!!! ! !!! !!
!!!!!!! !!!! !!!
!! !!
! ! ! !
!
!!
!! !!
!
! !! !!
!
!!!!!!!!
!
! !!
! !
!!! ! !!!!! ! ! ! ! !
! ! ! ! !! !!!! !!!! !! !! !
!! ! ! ! ! !
!!! !! ! !!!! ! !
!! ! ! ! !!! !! !!
!! ! ! !
!
! !!! ! ! !! !! ! ! ! ! !
! ! ! ! !!! !! !!
! ! !
!
!! ! ! !!
! !! ! ! !
!! ! ! !!
!! ! !!!! ! !!
65
! ! !! !
! ! !! !! ! !!
!! ! ! ! ! !! !! ! ! ! ! ! !
! !
! ! ! !! !
! !
! ! !! ! ! ! !
! ! !
! !!!! ! ! ! ! ! !
! !! !!! !! !! !
!
! ! ! ! !
! !!
! ! !
! ! ! !
!
! !
! ! !
! !
!! !
!

60 !
!
!

!
!

60 65 70 75
fheight

FIGURE 5.10: A scatterplot of sons’ heights against fathers’ heights with both a
linear regression line and a smoother overlaid. The two models are in close agreement
and the line lies inside the confidence interval of the smooth. A lot of unexplained
variability remains.

Adding a line or a smooth (or both) makes modelling explicit. When you look
at a graphic you employ implicit models in judging what you see and it is good to
formalise these where possible. With something like heights of fathers and sons you
might expect some kind of positive association and judge how the plot looks with
that image in mind. In other situations, perhaps when the same plot includes both
males and females, you might expect to see signs of there being two groups. Prior
expectations can be based on context and existing knowledge. Once you have seen a
plot, the data themselves influence what features you study and how you judge them.
Subjective impressions can supply valuable insights, but need to be confirmed by
checking with other views of the data and with objective testing, where possible.
86 Graphical Data Analysis with R

5.5 Comparing groups within scatterplots

The Olympic athletes’ dataset shown in Figure 5.1 suffered from overplotting, which
is one good reason for adjusting the scatterplot when displaying the data. Either
overlaid density estimates or alpha-blending might be used, but given the structure
of the data, where we can expect height and weight to differ by the sex of the athletes
and their sport, a better approach would be to split up the data by possible explanatory
variables. For instance, the following code plots a scatterplot for the females above
one for the males:

ggplot(oly12, aes(Height, Weight)) +

geom_point(size = 1) + facet_wrap(~Sex, ncol=1)

There are 42 sports listed in the dataset and plotting all the scatterplots together
gives Figure 5.11. Although each plot is quite small, you can still easily identify
a number of features. For some sports, all athletes are missing at least one of the
values (e.g., boxing, gymnastics) and for others, there are very few athletes. Where
there are enough data points, the association between height and weight generally
holds well, apart from some outliers. It is noticeable that the relationship is less good
for athletics, where a large number of very different events have all been grouped
together. With judo and wrestling, the relationship is also weaker, possibly because
of the range of different weight classes in these sports.
Whenever you split up a dataset in this way it is important to have all the plots
the same size with the same scales for comparative purposes. This is handled au-
tomatically with facetting in ggplot2 or within lattice. However, it is also useful to
organise the plots, so that the comparisons of most interest are easiest to make. Some-
times having certain plots in the same row is best, sometimes having them in the same
column. And often it will make sense to pick out particular subsets and make a plot
just with them. If you want to compare the height and weight scatterplots for judo,
weightlifting, and wrestling, then create the subset first and plot accordingly, e.g.

oly12JWW <- filter(oly12, Sport %in%

c("Judo", "Weightlifting", "Wrestling"))
ggplot(oly12JWW, aes(Height, Weight)) +
geom_point(size = 1) + facet_wrap(~Sport) +
ggtitle("Weight and Height by Sport")
Looking for Structure: Dependency Relationships and Associations 87

oly12S <- within(oly12, Sport <- abbreviate(Sport, 12))

ggplot(oly12S, aes(Height, Weight)) +
geom_point(size = 1) + facet_wrap(~Sport) +
ggtitle("Weight and Height by Sport")

Weight and Height by Sport

Archery Athletics Athltcs,Trth Badminton Basketball BeachVllybll Boxing
200
150
! !
! !
! ! ! !!
!!!!
! !! !! !
!!! !!
!!!!! !!!!
! ! !
!! !!!!!!!!
! !
! !
! !!!
!!!
! !!! !
!! ! !! !

100
! !! !!
! !
!!!
!!
!!! !!!! ! !
! !!!!!
!!
!!
!!!!!
!!!! ! ! !!!
!
!!
!
!
!
!!
!
!
!!
!
!
!!!
!
! ! !!
!!! !!! !
!!!
! ! ! !!!!
!! !
!!
! !!! !! !!!
!
!!!
! !
!
!
!!!
!! ! !
!! ! !! !
!!!
!
!!!
!!
! !
!!!
!
!!!
!
!! !!
!
!! ! !!!
!! !
!!!!
!!!
! !!!
!
!!
!!
! !! !!!
! !!
!!
!!
!
!!!
!!
!!!
!
!
!!!
!
!!
!!
!
!
!
!!!
!
!
! ! ! ! !!!!! !
!!!
!!!
!!!
! !!! !! !
!!
!
!
!!!
! ! !
!!
!
!!!!! !
!!!!!!
!!!!!!
!!!
!!! !
!!
!!
! ! ! !!!!!!!!
!
!!
!!
!
!
!!
!!
!!!! !
! !
!!!
!!!
!! !
!! !!! !! !! ! !!
! !
!!!!
!
!
!
!!!
!!
!!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!!
!
!!
!
!
!
!
!
!!
!
!
!
!
!
!
!!
!!
! !!
!!
!
!
!! !
!!!
! !!
!!!!
! !
! !
!!!
!
!!
!
!!
!
!
!!
!!
! !
!! ! ! !
! !
!! !!
!
!!
!
!
! !
!!!
!!! ! ! ! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!! !
! !!!!!
! !
!
!
!
!
!!
!
!
!
!
!
!
!
!!
!!
! !!
! !
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!!
!
!! ! !!!!!!
!! ! !
!
!
!! ! ! !! ! !!!! !!! !!
!! !!
!

50
!
!!!!!!!
!
!! !!
!!!!!!
!!
!!!
!!!!
!!!!!
!
!! !
!!! ! !!
!!
! ! !
!!
!!! !!!!
!!
!!
!!!
!! !!
! !
!!
! !
!!!! ! ! !
!!!!
!
!!
!
!
!!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!!
!
!
!!
!
!
!
!!
!
!
!
!
!
!!
!!
!
!!
!
!
!
!
!
!
!!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!!
!
!
!!
!
! ! ! !!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!!
!
!!! ! !
!!!
!!
!!
!
!
!
!!!!!
! ! !
!!
!! !
!
!!!! !!!!
! ! !! !! !!
!!
!
!!
!
!
!!!
!
!
!!
! !
!
!!
!
!
!!
!
!!
!
!!!
!!
!! ! ! !!
!
!!
!!
!
!!!!
!!!
!! ! !
!
! !! ! !
!!
!!
!!
!!
!
!
!!
!
!
!
!
!
!
!
!!
!
!
!!
!
!
!!
!
!
!
!
!!
!
!
!
!!
!
!!
!
!
!!!
!! !
! !
! !!! ! !
!!!!
!
!!
!
!
!!
!!
!
!
!!
!
!
!
!
!
!
!
!!
!
!
!
!!
!
!
!!
!
!!!
!
! !
! !!!!!

C!MB,C!R,C!T Canoe Slalom Canoe Sprint Cycl!BMX,C!T Cycl!MBk,C!R Cycl!MBk,C!T Cycling!BMX

200
150
100
! !
!!!!!
!!!!
!!
!!
!!
!
!!
!!
!!
!
!!
!!
!
!
!!!!!
! !!!
! !!
!!
! !
!!
! !! !!
!!
!
!!
!!
!
!!
!
!!
!!!
! ! ! !! !!!
!
!!
!!! !!
!
!!
!!
!!
!
!
!
!!
!
!
!!
! ! !!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!!!
!
! ! ! !! !
!
!!!!
!
!!!
!!
!!
!!
!!!
! !
!!
!!
!
!!
!!
!
!!
!
!!
!
! ! ! ! !

50
! !! !! !!!!!
!!!!
!!
!!
! !!!
! !!!!
!
!
!!!!!
!
!!!
!! ! !
! ! !
!!
!!
!!
!
!!
!
!!
!
!
!
!
!! !! !
!! !
! ! !
! ! !!!!
! !!! !

Cycling!Road Cycling!Trck Cycln!Rd,C!T Cyclng!MntnB Diving Equestrian Fencing

200
150
100 ! !!
!
!
! !
!!!!
!!!!!
! !
!!
!
!
!!
!
!!
!!
!!
!
!!
!
!
!!
!
!
!
!!!!
!!!
!

!!! ! !!
!
!
!
! !
!
!!
!
!
! ! !
!!
!
!
!
!
!
!!
!!
!
!
! !
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!! !
!
!!
!
!!!! ! ! !!!
!! !!!
!
!!
!
!
!!
!!!!! !
! !! ! !! !
!
! ! !! !
!! !!!!!
!
!!
!!
!!
!!
! !
! !
!!!!
!
!!
!
!
!!!!!
!
!!!
!!
!!
!
!!
!
!!
!
!!
!
!!
!!! !
!!!! !
!!
!!!
!!
!!
!
! ! !!!! ! ! !
!!
!!
!!! !!
! !!
!!
!!
!!
!
! !!! !
!
!!
!
!!
! !
!
!!
!!!
!
!!
! !!!
!!!!
!!
!
!
!
!
!!!
!!!
!

50
!!
! !!!! !!!
! ! ! !! !! ! ! ! !
! ! !!!!!
! ! ! !!!!
!
!!!!
! !!
!
!!!
! !
!!
!
!!
!!
!
!
!!!
!!
!
! !! !
!
!!
!
!
!
!!!
!!
!
!!!!
!!!!
!! ! !
!!!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!!
!!
!
! !
!
!
!!
!
!
!
!
!! !!!!
!
!
!
!
!
!
!
!
!
!
!!
!!
!
!
!!
!!
!
! !!!
!
!!
!
!
!!
!
!
!!
!
!
!
!!
!
!!
!
! !!
!!!! !
!!!! !!
! !
!
!!!!!
!!
! ! ! ! !!!
!!!
!!
!!
!
!!
!
!
!!!!
! !!!
!!
! !!
!!
!!
!
!!
!
!!!
!
!!! !! !
!
!!
!
!!!!
! !
!!!!
! !
! !! !!! !! ! !!
!!!! !!!!!
! ! !!
!
!!
!!!
! !! !! !
!!!!!
!!
!!!
! !!
!!!! !
!! ! !!
! !
!!! !
Weight

!! ! !!
!!!
! !! !!
! !!
!
!!!! !
! ! !
! ! !!! !!! ! !
!

Football Gymnstcs!Art Gymnstcs!Rhy Handball Hockey Judo ModrnPntthln

200
!

150 !
!!
!! !
!
! ! !
!
!

!! !
!
! ! ! !!!

100
! !!!! ! !
!! ! !!!!
! ! !!
!
!!!!!!!
!
!
!
!
!
!
!! ! !!
! ! ! !
!! !!!!
!!
!!!
! ! !! ! !!!!!!!
!!
!!
!!! !
!! ! !!! !
!
!
!
!
!
!
!
!
!
!
!!
!
!!
!
!!
!
! ! !!!
!!!
!!
!!!!
! !
!!
! ! !!!!
! !!
!!!
!!
! ! !!
!!! !!!!!
!!
!!
!!
!!! !
!!! !!!
!!
!!
!!
!!
!!
!
!!
!
!
!
!
!!
!
!
!
!! !
!!
! !
!
!
!
!
!!
!
!
!!
!!!
! ! !
!
!!
!
!!
!
!!
!
!
!!
!
!!
!
!!
!
!
!
! ! !!!!!
!
!! !
!
!
!! !
! !!
!
! ! ! !
!! !
! ! !!!
! !! ! ! !
! !
!
!!
!!
!
!
!
!
!
!!
!
!
!
!
!
!!
!!
!
!
!
!
!
!
!
!
!!
!
!!
!
!!
!!
!! ! !!
! !
!!
!
!
!
!!
!!
!
!
!!
!!
!
!
!!
!
!!
!
!
!
!
!
!
!!!
!!
!
!!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!!
!
!!!! !! ! !!
!!
!!!!!
!!! ! !!!!
!!
!!
!!
!!
!
!!!!
!
!!
!
!!
!
!
!!
!!
!!
!!
!!!
!
!!
!!
!
!
!
!!!!
! !
! ! !!
!!
!
!!
!!
!
!
!!
!!
!
! ! !!!!
!!
!
!!
!
!
!!
!
!
!!
!
!
!!
!!
!!!
! !
!!!
!!!
!
!
!!!
!
!
!!
!!!
!
!
!!
!!
! ! !!
!!
!
!
!

50
! !
!!!
!
!
!
!!
!
!
!
!
!
!!
!!
!
!
!!
!!!
!
!
!
!!
!
!!
!
!
!
!
!
!
!!
! !! !!
!!
!
!
!
!
!
!
!
!
!
!!
!!
!
!
!!
!
!
! !! !!
!
!!
!!
!
!
!
!
!
!!
!
!
!
!
!!
!
!!
!!
!
!!
! !!!!
!
!!
!
!
!!
!
!
!
!
!
!
!!
!!!
! !
! ! !!!
! !
!
!!!
!!
!!
!
!!
!
!!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!!
!
!
!
!!
!
!
!!
! !
! ! !!!!!
!!!!
! !
!! !
!
!
!!
!
!
!!
!
!!
!
!!
!!
!
!!
!!
!
!
!!
!!!
!
! ! !!!!
!!
!!
!!
!!!!! !!! !
!
!!
!! !
!
!
!!!
!
!
!!
!
!!
!
!
!
!
!
!
!!!
!
!!
!!! !! ! !!
!
!
!
!
!!
!
!!
!
!!
!!
!
!
!
!!
!!
!!
!
!
!!!
!
!! ! !
!!
!
!
!
!
!!
!
!
!
!!!
!
! ! !! !!
!
!!
!!
!
!!
!! ! ! ! !!! !!
!!!! ! !
! !

Rowing Sailing Shooting Swimming SynchrnsdSwm Table Tennis Taekwondo

200
150 !
!

100
! ! !
!!
! !!!! !! ! ! !
!!!
!!
!!
!!
!!
! !! !!!
!
! ! !
!! !!
!!!
!!
!!
!!
!! !!!
!!
!!! ! !!!! ! !!!
!!!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!!
! ! !!
!!
!
!
!!
!
!!
!!
! ! !! !
!!!!
!!!! ! ! ! ! !!!
!
! !
!!
!!
! !
!
!!
!!
!
!
!!
!!
!
!!
!
!
!!
!!
! !! ! !!! !! !!!!
!
!!!
!
!
!
!
!!! ! !
!! !
!
!
!!!
!!!
!
!
!
!
!
!
!
!
!
!
!
!
!!!!!! !!
!
!!
!!
!
!!!!!
! !!! !!
! !!!! !! ! ! !
!!
!! !!!! !!!
!!!
!
!!
!!
!!
!
!!
! !
!
!!
!!!
!!!
! !!!!!!
! !!! !
! !!!! !!
!
!
!
!!
!
!
!!
! ! !!!
!!
!!!
!
!!
!
!
!!
!
!!
!
!
!!
!!
!
!!
! !
!!
!! !
!!
!!!
!!!
!
!!
!
!
!!
!!
!
!! ! !! !!
!!
!
!
!
!
!
!
!
!!
!!
!
!
!!
!!!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
!!
!!!!!
!
! !!!
!
!!
!!
!
!
!!
!!
!!
!
!
!!
!
!
!!
!
!
!
!!
!
!
!!
!
!
!!
!!!
! !! !
!!
!!!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!!
!!
! ! !! ! !
!!
!
!
!
!
!
!!
!
!
!!
!
!
!!
!
!
!
!!
!!
!
!!!!! !!!
!
!
!
!
!
!
!!
!
!!!
!
!!
!
!
!
!!
!
!
!
!!
!
!!
!
!
!!
!
!
!
!!
! !!
!!!!!
! ! !! !
!!
!!!
!
! !!
!!
!!! !
!!
!!
!!
!!!
!!
!
!!
!!
!!
!!
!!! !
!!
!!
!!!!!
!!!! ! !!! !
!!!!
!!!!!!!! !
!!
!!
!!
!!!
!
!
!!
!!
!
!!
!!
! !
! !!! !

50
!!
! !! !
! !
!!
!
! !
!!!!
!!!!
!! !! ! !
! !
!! ! ! ! !!
!!
!!
!!
!!!!!
!!
!!!! !!! !!!
!! !! !!
!
! !! !!
!!
!
!
!
!
!
!
!!
!
!!
!
!
!!!
!
! ! !!!!!
!
!
!!!
!!
!!
!
!
!
!
!!
!
! !!! ! !
!
!
!
!
!!
!
!
!!
!
!
!
!
!!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!!
!!
!!!
! ! !!
! !!
!!!!
!!
!
!!
!
!!!!
!!
!
!
!!!
!!
!! ! !!!!!!
!!!
!
!!
!
!
! !
! !! !!
!!!
!!
!
!!!
!!
!!! ! !!!!
!
!!
!
!!
!!
!
!!
!
!
!
!!
!
!!!
!
!!
!
!! !!
! !!
!
!
!!!
!!!
!!!!!!
! !!!!!!
!!
! !
!!!
!!!
!
!!!
!!
!
!
!
!!
!!
!! !!
!!
!!
!
!!
!
!
!!
!
!
!
!!
!
!
!!
!!
!!
!!!
! !!
!
!
!
!
!
!!
!
!!!!
!
!! !
!
!! !
! ! ! !!!
!!!!! !
!!
!!
! !
!!!!!
!!!

Tennis Trampoline Triathlon Volleyball Water Polo Weightliftng Wrestling

200
!

150
!
! ! !
! !
!
! !!!!
! !
! !!! !!
!!
!!
!!!!!!

100
! ! ! !!
!
! !
!!
!!!
! !!!
!!!
!
!
!
!
! !! !
! !!
!! !
! !
!!
! ! !! ! !! !!!!!
! !
!! !!
! !!!!!!
!!!! !
!! ! !! !!!
!
!
!!
!
!
!!!
!
!!!
!!
!! ! ! ! !
!! !!!
!!
!!
!!!!
!
! !
!!!! ! !!
!
!!
! !
!
!
!!
!
!
!
!!!
! !!!! !! ! ! !! !
!!
!!!
!!!!!
!! !
!!!!
! ! !!!
!!
!!
!!!
!!
!
!
!!!
!
!
!!
!
!! !
!!
!
!! !!!
!!
!!
!
!!!!! !!!!!!!
! ! !!
!!!
!!
!!
! !
!! !!!
! !
!
!!
!
!!
! !
!
! !! !!
! !!!
!
!
!!
!!
!
!!!! ! !!!
!!!!!! !!!
!
!
!
!!
!!
!
!
!
!!
!
!
!
! !!
!! ! !! !
!
!!!
!
!
!
!
!
!!!
! !
!!!!!!
!! !!!! !
!!
! !!!!!
!
!
!
!
! !
!!
!!!
!!
!!!!! !
!! !!!
!!!!!! !! !!
!!!!
!! !! ! !
!!
!!
!
!!!
!
!!
!!
! ! !
!! !
! !
!! !!!!!!!! !!!
!
!!!!!!
! !
! !! !! ! !!
!
!!!!
!!
!!
!!!
!!!!!!
!!
! !
!!!!!!
! ! ! !
!!!
!
!!!
!!
!
! !! !
!!
!
! ! !
!!
! !!
!!!!!!!! ! !! ! ! !

50
!!
!!
! !! !
!!!
!!
! !!!
!!
!! !
! !!
!!
! !
!! !
!!
!!
! !!
! ! !!!!
! !
!!!!
! !!! ! !
!!
!
!!
!!!!!!
!
! !
!!
!!
!!!
!
!
!!
! ! !!! !
!
!
!!! ! !
!
!
! !!
! !
!!
!!
!
!! !
!
!
!!!
! !
! ! !
!!!
!!
!
!!
!
!
!!
!!
!! ! ! !! !!!
!!!!
!! ! !
!
!!
!!
!!
!
!!
! !!!!! ! !!!!
!
!
!!!
!!!
!!
! !!!
!
!
!!!
!
!!! ! !!!! !!!!!
!!!!!
!
!
!!!
!
!
! ! !!!!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!!
! !
!
! !!
!
!
!
!
!!!!!
!
!!!
!
! ! ! ! !
! !!
!!
!
!
!
!
! !!
!
!
!
!
!
! !!
! !
!!
!!!!
!

1.501.752.002.251.501.752.002.251.501.752.002.251.501.752.002.251.501.752.002.251.501.752.002.251.501.752.002.25
Height

FIGURE 5.11: Scatterplots of weight by height for the different types of sport at the
London Olympics. Data are missing for some sports. The association between weight
and height looks linear for most sports. The names have been limited to a maximum
of 12 characters using the abbreviate function.
88 Graphical Data Analysis with R

5.6 Scatterplot matrices for looking at many pairs of variables

Scatterplot matrices (sploms) are tables of scatterplots with each variable plotted
against all of the others. They give excellent initial overviews of the relationships
between continuous variables in datasets with small numbers of variables.
There are many different ways of drawing sploms. You can have different options
of what is plotted on the diagonal or of what is plotted above or below the diagonal.
Since the plots above are just flipped versions of the ones below, some analysts prefer
to provide statistical information or other displays in the other half of a splom. You
can restrict sploms to continuous variables or provide additional types of plot for
including categorical variables. If you plot histograms or density estimates down the
diagonal, then the scatterplot matrix gives you an overview of the one-dimensional
distributions as well.

Crime in the U.S.

The dataset crime.us from the VGAMdata package includes the absolute crime fig-
ures and the crime rates by population for the fifty U.S. states in 2009. Figure 5.12
shows a splom of the rates for seven kinds of crime (the dataset also includes rates
for the four crimes of violence together and the three property crimes together). It
also shows density estimates for the variables down the diagonal. Whatever precon-
ceived notions we might have of how the rates might be related to one another, the
splom provides a first, quick summary of the associations between the variables, both
through the graphics and through the correlation coefficients.
The scales have not been drawn for two reasons. They overlap with the variable
names and they are too small to read easily on a graphic this size (larger versions
work better if you want to see the scales). It is important to remember that the rate
levels are quite different for different crimes. State murder rates in this dataset go up
to almost 12 per 100, 000 population, whereas larceny rates go up to just over 2700
per 100, 000 population.
The rates for most crimes seem positively associated, some more strongly than
others (e.g., murder and burglary), and the rate of rape is not associated much with
any of the other crime rates. The larceny rate is not closely associated with the four
violent crime rates. There are a few outliers, which stand out more in some plots than
others, such as the two states with high rates of motor vehicle theft in the scatterplot
with larceny (California and Nevada). As the dataset only has 50 cases, you cannot
read too much into the shape of the graphics or the correlation coefficients. It is still
instructive to compare each plot with its correlation to see how they compare. The
different sizes of the State populations should also be taken into account. Alaska,
North Dakota, Vermont, and Wyoming are each treated as having equal weight with
California, even though their populations are less than 2% of California’s.
Looking for Structure: Dependency Relationships and Associations 89

data(crime.us, package="VGAMdata")
crime.usR <- crime.us
names(crime.usR) <- gsub("*Rate", "", names(crime.usR))
names(crime.usR)[19:20] <- c("Larceny", "MotorVTheft")
ggpairs(crime.usR[, c(13:16, 18:20)],
title="Crime rates in the USA",
diag=list(continuous=’density’), axisLabels=’none’)

Crime rates in the USA

Corr: Corr: Corr: Corr: Corr: Corr:

Murder

0.0716 0.657 0.706 0.717 0.424 0.487

!
Corr: Corr: Corr: Corr: Corr:
Rape

!
!!
!
!

!0.156 0.379 0.196 0.178 0.147

!! ! !! !
! ! !! ! !
!
!! !!
!! ! !!! !
!!!
!! ! !!!!
! !! ! !
!!
! !
! !
! !
Robbery

Corr: Corr: Corr: Corr:

! ! !!!
!
!! ! !!!
! !! ! !
! ! !
!
! !! ! !! !
!! !! !
!! ! !
0.584 0.488 0.319 0.629
!
!! ! ! !! !
! ! !!
! ! !! ! ! !
!
! !!!! !!
!! ! !
! ! ! !
!! ! !!
!!
!! !!! ! !
!
! !
! ! !
! ! ! ! ! ! ! !
! ! ! ! !
!! ! ! !! !

Corr: Corr: Corr:

Assault

!
! !! !
!
! ! !
!
! !
! ! ! !
!!
!
! !! ! ! ! ! !
! ! !! ! !

0.607 0.499 0.542

!
!! ! !
! ! !
!
! !!!!! ! !! !
! !!
! !
!
!!!!!!! ! !!! !
!
!! ! !! !!
! !
! !! ! !
! !! ! ! ! !! ! !
!
! !! ! !
!!! !!!
! ! ! ! ! ! !
!! ! ! !!
! ! !
! ! ! !
! ! ! !
! ! ! !
!!! ! !! ! ! !!!! !!! ! !! !!
Burglary

!!! !! ! !!

Corr: Corr:
! ! !!! ! ! !
! ! ! !
! ! !! !! !
!!! !
!
! ! ! !
!
! ! ! !!! ! !
! ! !
0.688 0.524
! !! ! !! ! ! !
!
!! ! !!! !! !! ! ! ! !!
! !
!!! ! !! !
!! ! ! !! !! ! ! ! !!
!
!!
!
!!!
!
! !!! ! ! ! !! !! ! !!!!
!
! !
! !!!! ! ! !
! ! ! ! ! !! ! !! !
!! ! !!
! ! !!! !!! !
! ! ! ! ! ! ! !
! ! ! ! !
! !! !! !! !! ! ! ! ! ! !! !
!
! ! !! ! !
!! ! !!
! ! ! ! !! ! !
!
Larceny

Corr:
!!! !!
!! ! !! ! !
! ! !! ! !! !
! ! ! ! ! !!!! !
! !!!! ! !!!
!! ! ! ! ! !! ! ! ! !!! ! ! !
! ! ! ! !
! ! !
!
! ! ! !! ! ! ! !! ! !
! !! ! !! !! ! !
0.532
! ! ! ! ! !
!! ! ! ! !! ! !!
! ! !! ! ! ! ! !
!! !! !! !! ! ! !! ! ! ! ! !! ! ! ! !!! !! !
! !! ! ! ! !! ! ! !! !
! !! ! ! !!
!!
!! ! ! ! ! !
! ! ! !!! !!
!
! !! !
! !
! !
! !! !!!
! ! !
!
! ! ! ! !
! ! ! ! ! !
MotorVTheft

! ! ! ! ! !
! ! ! ! ! !
!! ! ! !! !! ! !
! !! !! ! ! ! ! ! ! ! !! !
!! ! ! ! !! ! !
! ! !! !! !
! ! !!! ! !! ! !! ! !!
!! ! !
!! !
!
! ! !!! !!!
!
! ! !! ! !!! ! ! ! ! ! !
!! ! ! !
! !
! !!!!!!
!!
!!
!
!!! !! !
!
!!! !! ! !! !!! !!! !! !! !
!! !!
!
!!
!! ! ! !
! !!
!
!! ! !! ! !! ! ! ! ! !! ! !!
! ! !! !!! ! !!! ! ! !
! !! ! !! !!! ! ! !!
! !! !! !! ! !
! ! ! !
!
!! ! !! ! ! !!!
! ! ! !
! ! !!!!!! !!! ! ! ! !!! ! !
! ! ! ! ! !
!!!!! ! !!
! ! !
!!
!! !! !
!!
! !! !! ! ! !!!

Murder Rape Robbery Assault Burglary Larceny MotorVTheft

FIGURE 5.12: A scatterplot matrix of the 7 crime rates in the crime.us dataset. Some
rates are positively associated, others not. The crime of rape is least associated with
the others.
90 Graphical Data Analysis with R

Swiss banknotes
The dataset bank from the gclus package includes six measurements on each of 100
genuine Swiss banknotes and 100 forged notes. The dataset was used extensively in
a fine multivariate statistics textbook [Flury and Riedwyl, 1988].

library(car)
data(bank, package="gclus")
par(mar=c(1.1, 1.1, 1.1, 1.1))
spm(select(bank, Length:Diagonal), pch=c(16, 16),
diagonal="histogram", smoother=FALSE,
reg.line=FALSE, groups=bank$Status)

129.0 130.5 7 9 11 138 140 142

Length
! ! ! ! !
! 0

214.0 215.5
! ! ! ! !
! ! !! ! ! ! !!!! ! !!
!!! ! !!! ! ! !! ! ! !!!!
! !!!!! ! !!!! ! ! !!! !!
! 1 !!
!!
!
!! !!!
!
!!! !
!!
!!!
!!
!
!!! ! !!
! !
!!!!!!
!
!!
!!! ! !
!!!!
! ! ! !
!!
!!
!
!
!!
!
!
!! !
!!
!
!!!
!!
!
!
!!!!
!! ! ! ! !!
!
!
!
!
!!
!! !!!!!
!
!!
!
!
!
!
!
! !
!!!!
!!
!
!! ! ! !
!!
!
!!
!!!
! !
!
!
!
!
!
!
! !
!
!!
!
!!!
!
! ! !
!!!!!!!
!
!! !!
!!!
!!!!
!!
!! !!! !
!!
!!!
!!!
!!
!
!! !!
!!
!!! !!!!
!! ! !
!!!!!
!
!!!!
!!
!!
!
!!
! !!!!
! !!!
! !
!!!
!
!!
! !!!
!!! ! ! !
!!!
! !
!!
!!
!!
!
!
!!!
!!
! !!
! !!
!!!!!
!!
! !!!
! !!
!
! !!!!!
!
!!!
!
!!!
!
!!!
!!!! !!
!!
!! !!!!! ! !!!! ! !!!
!!!!! ! ! !!!! !!!!
!!
! !!!!
!!!!
!
! !!
!!!
!
!!! !
!
!! !
!!
!
!!
!!
!!!
!
!!
!!
!!
!
!!!
!!
!
!
!!!!!
! ! !
!!
!!
!
!!
!!
!!!
!
!!! !!
!! !
!!!
!!!
! !!
!! !!
!!!!!!
!!
!
!
!
!!
!! !!
!
!!
!! !!
!!!! !!!
!! !
!!!!
! !
! !!!
!! !!!
!!!
!!!
!
!!!!! ! !!!!
! !
!!
!!!
!
!!!
!
!!
!!
!
!
! ! !! !
! ! !!!
!
!!
!
!!!
!!
!
!! !
! !
!!
! !!
!
!!
!
!! !!
!!!!
!! ! !!!!! !!
! !
!!
!!
!!
!!!! ! !! !!!
! !
!
! ! !
!
!
!!!
!
!!!
! !
!
! !!
! !
!!! !! ! !!
!
!! ! !
!!
!
!!! !
! !!
!
! !!
!!!
!! ! ! ! ! ! ! !!! !! ! !
!!
!
! ! ! ! ! ! !!! ! ! ! ! ! !

! ! ! ! !
Left
130.5

! ! !! !!!!! ! ! !
! ! ! ! !! ! !!! ! !!! !!! !!
!!!!!
!!!!!
!!!
!! ! !!
!!!!
!!
!!
! !!!! !!!!!!!! !
!! !! ! !!! !
!
!!!!
! !!!!
!! ! !!
! !! !!!! !
!!
!
!!!
! !! !!!!
!!!!!
! !!!!
!
!
!! !!!!!
!!
!!!!
!!
!!!
!!!
!
!!!!
!!!!
!!!!!!
!!!!!
!! !
!!!!
! !
!!!!!
! !!!
!!!
!
!!!
!!
!!
!!
!
!!!
!
!!
! !
!
!! !
!
! !!
!
!!
!
!
!!
!
!
!!!
!
!!
! !
!!!!!
!! !! !
!!!!!
!!!
!!!!
!!!!!
!!! !
!!!!
!!
!!
!
!!!
!!
!
!
!!!
!!
!
!!
!
!! !
!
!
!!!!!
!!!
! !! !
!
!!
!
!!!
!
!! !
!
!
!
!!
!!!
!!!
!!
!!
!!!!!
!!!
!!
!!!
!
!
!
!
!
!
!
!!
!!!
!!
!!
!
!
!! ! ! ! !
!
!!
!
!!
!
!
!
!
!
!!
!
!
!
!!
!
!! !! !!!
!
!
!
!!!
!! ! !!
!!!
!!! !!!
!!!!!!! !
!!!!
!!
!! !! ! ! ! !
! ! !
! !!!!! !!!
!! ! !! ! !!
!!!
!!!!!!! !! !
!! !
!! ! !!!!
! !! ! !! !!!!!
!!! ! !
!
!!!!!
! !!
! !!
!!!
!! ! !!
!! !!!!!!!
! !!
! !!!! !
!! !!!
!!!
!
!
!!!
!
!!!
!
!!!!!!!!
!!
!!!!!!
!!! !! !!
!! !!
!! !!
!! ! !
! !!!
!!
!!
! !!!
!!!
!
!
!!!
!! !
! !! ! !!
! !!
!!!
!!!
!!!!!!!!! !!! ! ! !!
! !
!
!
!!
!
!!
!!
!!
!
!!!!! ! !!!!!!! !!
!!! !! !
!!!!!! !!!!!
!
! !!
129.0

! ! ! ! !

Right
! !! !! ! ! ! ! ! !
! !!!

130.5
! ! !! ! ! ! ! !
!!! !!!
!!!! ! !
!!
!!!!!
!! ! !!
!!!
!!
!!
!! !!
!!
!!!
!! !!
!! !
! !
!!
!
!
!!
!
! ! !
!
! !
! !!
!!!
!
!!
!
!!!!!
!!
!
!
!!
! !
!!!
!!!
!
!
! ! !
! !!
!!!!!! !!!
!
!!! !
!
!
!
!!! !!
!!
!!!!!!!!! !!!! !! !
! !
!!!
!!!
!!
! ! !
!!! !! ! !
!!!
!!!!
!! !!
!!!
!
!!!! !! !
!!!
!! !! !!!!!! !!!
! !! !!!!!!!
!! !!!
!! !
!!
!
!!!
! !!!!
!! !
!!!!!
!! !! !
!!
!
!!
!!
!!
!!!!
! !
!!
!
!!
!
!
!!
!!!!! !!! !!!
!
!!
!
!!
!
!!
!!
!!
! ! !! !! ! !!
!
!!
!! !!!
!!
!!! !
!!!!
!!!!!!!
!!
! ! !!!
!!
!! !
!!
!!!
!!
!!!!!
!!
!
! !
!!!
!
!
!!
!
! !!
!
!
!
! !
!!!!!
!
!!! !!
!
!
! !! ! !!
!!!!
!
!
!
!!
!
!
!
!
!
!!
!!
!!
!
!!
!
!!!!!
!
!
! !
!!
!!
!!
!
!!
!
!
!! ! !!
!!!
!!
!
!!
!!
!!
!! !! !!
!!!! !! !!!!!!!! ! ! !
!!!!
!!
!!!!! ! ! ! ! !!
! !!
! !
!!!!!!
!! ! ! ! !!
!!!
!
! !
!
!! !!
!
!
!!
!
!!
!!
!!
! !!!
!! ! !
!!!! !
!!!!! !
! !!
!
!
!!!
!!!!
!
!!! !!
! !
!
!
!!
!!!
! !! ! !!
!
!
!!
!!
!
!!! ! !!
! !
!!!!
!!!
!!
!
!!! ! !!
!!! ! !! ! ! !! !!!! ! ! !!
!!!!!
129.0

! !! !!!
!! !! ! !!! !
!!!! ! ! !!!! !! ! !!
!!!
! !!! !!!!! !
!!
! !
!! !
!!
!! !!!!!!!

!! ! ! ! !
!! ! ! !
Bottom
! !! ! !!
!! !!!!!! !! !!
!!!!
!
!!! !!!! !
!!!
!!!!!
!
!!!!
! !
! ! !! !! ! ! !
!!
!!!!
!!!! !! !!!!
!!!!
!! !!!!!!! !
!! !
! !! !!
!
!!
!
!!! !
!!
!!!
!!
!!!
!
!
11

! !! !!!! ! !!
!!! !!! ! ! ! !!!! !!!! !
!! ! ! ! !!!!! !!!! ! !!!!!
!!!! !!!!!
!!!
! ! !!
!! !! ! !!!
! ! !! !!!!! ! ! ! !
!!!!!! !
! !!
!
!!! !
!!!
! ! ! !!!!!! !!!!!!!! !
! !!!!
! !!
!
!! !
! !!!
!
!!!! !
!!!!!!!
!!!!! ! !!!! ! !! ! ! ! ! !
!!!!
!!
! !! !!
! !!! ! ! !
! !
!!!!! !!!!! !!!
!!! ! !!! !! !
! !!!
! !!!! ! !! !
! ! !! !
!!
!!
!
! !! ! !
!!
!!!
!! ! !!! !! !!!!! ! ! ! !
!! ! !! !!!! !! ! !
!!! ! !
! ! !
! !!!
!!!!
9

! ! ! !!
!!! !!!!! !! ! !
!! !! ! ! !!!!!! !
!! !
!!! !
!!!!
!!!!! ! ! !!!!!! !!
! !!! !
!!!!!!!!!! !
!!!
!!!! !! !!
!!! !!!! !! !!
!!
! !!
!
!
!!
! !
! !
! !!!
!
!!
!!!!
!
! !!! ! !!!!!!!!! !! !!!
!! ! !!
!! ! !! ! !!!!!! !!!! !!
!!!!
!!! !!
!!!!!!! !
! !! !!! !! !
!!
!!!
! !!
!
! !!! ! ! !!! !!! !! ! !! !
!!! ! !
!
! !!!
!! !!
!!
!
!!!!!!!! !!!
!!!
!! !!! !
!! !
!!
!
!!!!
!!! ! !! !!
!! ! ! !!
!!! ! !! ! ! !!!
! ! !!!!! !!!
7

Top
!! ! ! ! !!! !!!! !! ! !!
12

!!! ! !! !
!!!! ! !! ! !! !! ! !
!!
!
!!!!! !
!!!!!! ! !!
!!! !!!!!
!
!!!
! ! ! !!
! ! !!
!
!
!!
!
!
!
!
!!! ! !!!
!
!!!! !
!
! !!!!!!!!
!!!
!! !! !!!
! !!!!
!!!! ! !!
! !!
!!
!! !!
!!! !! ! ! ! !
!!!!!
! ! !! ! !! !
!!!
!!!
!
! !!!
!
! !! !!!!
!!!!!
!!
!!!!
!!!!!!!!
!! !!
!! !!!
!!
!!
!!!
!!! !! !!!!!
!! !!
!! !!
!!!!! !! !
!! !!
!!
!
!!! !
!! !!!
!!! !
!!
!!
!!
!!!
! !!
!!!
!!
!!
!
!! ! !
!!! !!
!!
!!
!! !!
!!
!!
!! ! !!!!!!
! !
!!!!!
!!
!!!!!
!!
!
!!
!!
!
!
! !
!
!
!!!
!
!
!
!
!
!
!
! !!! !!!!!!
!
!
! !
!
!
!!!
!
!!
!
!
!!
!
!
!!
!
!!
!
!
!! !!!!!
!!!
!!
!
!!
!!
!!
!!!!!
!! ! !
!! !! ! !!
! !!
!!!
!!
!!!!
!!
!!!
!! !
! !
!!!!
! !!
!! !! ! !
!
!!!
! !! !! ! !!
!
!!!
!!
!!
!! ! !
!!!!
!
! !!! !
!
!
! !
!!!!
!!!! ! ! ! ! !!! !! ! !!!! !! ! !!!!
!!!! ! ! !!!!!! !!
! !!
!!
!!!!
10

!!! !!!
!! !!
!!
!!!!!!!! ! !
!!! !!
!
! !!!!! !
!! !! !!!! !!!!!
!!!! !!
!!!!
!!! !!!!! !
!!
! ! !!!
! !! !
! !
! !! !! !! !!! !! !!! !! ! !
!!!!! ! !!!!!!
!!
!!!
!
!!!!!!!
! !!
!! !!
!! ! !!!
! !!
!!! !
!
!!
!
! ! ! !! ! !
!!! !
! !!
!!! !!! ! !!!
!! !!
! !
!
!
!!
!! !
!!
! ! !!!!
!!
!
!
!
!
!!!!
!
!
!
!! ! ! !!
!!! !! !
!! !! !! ! !!!!! ! ! ! !! !!
! ! ! ! !
8

! ! ! ! !

Diagonal
138 140 142

! ! ! ! !!! !!! ! !! ! ! ! !!!!!!

!!!!!
!!!
!!
!! !!
!! ! ! ! !! ! !!!!!
!!
!! !! !
!!!! !
!!!
!!!
!!
!!!!
!!!!! ! !!!
!!
!!!!!!
! !!
!!! !!!!
!!
!!!
!!
!!! !!
!!! !!
!
!!
!
!!!
!
! !
!!
!! !!!! ! !
! !! !
!!
!!
!!!!
!
! !
!!
!
! ! !
! !! !!! !!!
!!
!! !!!
!!! !!
!! !!
!!!!
!!
!
!!!
!!
!!
!!!!!
!! ! ! !
!
! !!
!!!
! !
!!!
!!
!!
!!
!!
!!!! !
!
!!
!!
!!
!
!
!
!
!
!
!
!!
!
!
!!
!!
!
!
!
!!
! ! !
!
!!!
!
!!!!!!
!!
!!
! !!
!! !!
!!!!!!! !!
!!!!!! !!!!! ! ! !
!
!! !! !
!!
!
! !!!
!!! ! !
!
!
!!!
!!
!
! !
! !
!
! !! !!!!
!! ! !! ! ! !!! !!
!!!
!! !
!
!! ! ! !
!!
! !
!!!
! !!
!
! !!! !
! !
!!
! !
!!
!!!!!
!!! ! ! ! !! ! ! !!!! !
!!
!!!!!! !!!
!!! !!!!! ! ! !!!!!! ! ! !!
!
! !!
!!!!
!!!! !!
!!!
! !!!!
!! !!
!
!!
!
! !!
!!!!
!
!!
!!!!!!
! ! ! !
!
!!
!!
!!!!!
!!!
!
!
!
!
!!!
!
!!
! !
!
!!
!
!!
!!! !!
!! !!!!!
!
!!!!
!!
! !!
!! !
!!! !!
!!!
!!!!!!! !
! ! !
!!!
!!!
! !
!
!!
!!
!!!
!
!
!!! ! !
!
!
!!
!
!! !
!!!
!
!!
!!
!
!!
!
!!
!!!!
!!
!!! !! !!!
!
!!!
!!
!!
!
!!
! !! !!!
!!!
!!!
!!!
!!! ! !!
!
!!
!!!
!!!!!
! !
!!
!!!!! ! !
!
!!!!
!
!
!!!
!
! !
!!!
! !!!! ! !
!!
!! ! !
!
! !
! !! !!
!! !!
!
!!
!
! !
!
! !!
!
!
!!
! !!! !
!
! !!! !! ! !
! !!!! !
! ! ! ! !
! !
!! ! ! ! !
!!! !
!
! ! !!
! !
! !!! ! ! ! ! ! !
! !! ! ! !
!! !!

214.0 215.5 129.0 130.5 8 10 12

FIGURE 5.13: A scatterplot matrix of the Swiss banknotes dataset with the forged
notes coloured in red. In some of the scatterplots the groups of notes are well sepa-
rated. Some variables are associated, some not. There are a few possible outliers, not
all of them forgeries.
Looking for Structure: Dependency Relationships and Associations 91

A splom of the variables is shown in Figure 5.13. Left and Right are strongly
positively correlated. There is evidence of negative association amongst the last three
variables (Bottom, Top, and Diagonal) and there are also suggestions that the
genuine and counterfeit notes can be distinguished using those variables. The scat-
terplot for Bottom and Top is particularly interesting, as the overall association is
slightly positive, while the two possible subgroups each have a negative association
between the two variables. The default display for the spm function from car in-
cludes smoothers with confidence bands, linear regression lines in the scatterplots,
and density estimates on the diagonal. It also uses open circles for the points. All this
makes for a fairly cluttered display, so other options have been chosen here.

Functions for drawing sploms

In R you can choose between the functions listed in Table 5.1 and probably there are
other alternatives too. Some of these are faster than others.

function package
plot
pairs
spm car
cpairs gclus
splom lattice
ggpairs GGally
gpairs gpairs
pairs.mod SMPracticals
pairscor.fnc languageR
... ...

TABLE 5.1: Splom functions in R

Each of the functions offers different versions. In particular, some can cope with
categorical variables and some cannot. In this chapter we have looked at some simple
scatterplot matrices involving just continuous variables (and only a few of them at
that). If you want to look at more elaborate sploms with more variables, take a look
at the examples on the help pages of some of the functions, for instance:
• cpairs with panels coloured by level of correlation

library(gclus)
judge.cor <- cor(USJudgeRatings)
judge.color <- dmat.color(judge.cor)
cpairs(USJudgeRatings, panel.colors=judge.color,
pch=".", gap=.5)
92 Graphical Data Analysis with R

• gpairs with loess smooths

library(gpairs)
data(Leaves)
gpairs(Leaves[1:10], lower.pars=list(scatter=’loess’))

• pairs.mod with standard scatterplots as well as scatterplots of partial residuals

library(SMPracticals)
data(mathmarks)
pairs.mod(mathmarks)

Sploms take up a lot of space. A splom for m variables includes m(m − 1) scatterplots
and m diagonal elements. Even if only the upper triangle of plots is drawn, you
still need m rows and m columns for m variables. This makes sploms ineffective for
large numbers of variables, especially when there are many cases, as each individual
plot can only be drawn fairly small. Nevertheless, for a quick overview, perhaps for
deciding which of a number of scatterplots to look into in more detail, they can be
very useful.

5.7 Scatterplot options

• Point size
Very small points can hardly be seen and very large points overlap each other and
make the plot look like a collection of clumps. Making points a little larger can
be useful for emphasising outliers; making points a little smaller can be useful
for distinguishing groups of points. Occasionally, particular point sizes can lead
to undesirable visual effects, especially if the data are gridded with parallel strips
of points close together.
Point size can also be used to represent the value of a continuous non-negative
variable. When the basic symbol is a circle, then this gives a bubble chart. It can
be quite useful for smallish datasets without too much overlapping and is used in
Gapminder’s displays of global development data [Rosling, 2013].
• Symbols for points
The current default in R for scatterplots is to use open circles. As attentive readers
will have gathered, I prefer small filled circles. You should use what you think
conveys information best for you.
When a dataset is small and made up of different groups, it is helpful to be able
to tell the groups apart in a scatterplot. An old solution is to draw the members
of each group with a different symbol. This can work if the points do not overlap
Looking for Structure: Dependency Relationships and Associations 93

and if the symbols are easy to identify. Mostly it does not, and colour is more
effective anyway. With large datasets, using different symbols looks cluttered
and messy, so a set of scatterplots, one for each group, as in a trellis display,
works better.
• Alpha-blending
A partial solution to overplotting problems is to use alpha-blending. Each point
is given a weight between 0 and 1 and where several points overlap, the resulting
area is drawn correspondingly darker (in fact more opaque). For instance, if α =
0.1, then any area with ten or more points has the maximum darkness. The effect
is to emphasise areas of higher density and downplay areas of lesser density,
so that outliers cannot be seen so easily. Using no alpha-blending (equivalent
to α = 1) can leave the bulk of the data for a large dataset looking like a solid
indistinguishable mass. Alpha-blending works better interactively, when you can
explore a range of alpha values quickly to see what information is shown at
different levels.
• Colouring points
Colour is often used to distinguish points by groups. If the data are fairly well
separated (whatever that means in practice), this is helpful. You always have to
bear in mind that when there are points on top of one another, the visible colour
is the colour of the last point drawn. So if there are three colours, red, green, and
blue, the plot may look different if drawn in that order (maybe lots of blue), than
if drawn in the reverse order (maybe lots of red).
If you think an interesting structure is evident in a coloured scatterplot, then it
is worthwhile drawing a trellis plot of scatterplots by group. Although coloured
scatterplots may have disadvantages, it is often easier to identify a possible group
structure there than in a trellis plot and it is always good to check.
When colour is used to highlight particular points, it is usually drawn last and
hence on top. The same cautionary advice applies here as for colouring by
groups.
• Splom options
The many options available for sploms were referred to in §5.6. You can display
statistics instead of half of the scatterplots, add models to scatterplots, or show
marginal information on the diagonal.
94 Graphical Data Analysis with R

5.8 Modelling and testing for relationships between variables

1. Correlation
If you calculate a correlation coefficient, you should always draw a scatterplot
to learn what the coefficient might mean. Correlation coefficients measure linear
association and it is a rare scatterplot where that is all you can see. By the same
token, if you draw a scatterplot and see a linear association, it is a good idea to
calculate the correlation coefficient to measure just what level of correlation you
have in front of you (cf. [Cleveland et al., 1982]).
Correlation coefficients are sometimes accompanied by p-values in publications
and the null hypothesis tested always seems to be H0 : ρ = 0. Fisher showed
long ago in 1915 that there is a good approximation for testing null hypotheses
of correlation coefficients being equal to more interesting and informative values
than 0, but this is seldom done. Pity.
2. Regression
Given a presumed causal relationship between the y-variable and the x-variable
of a scatterplot, regression may be used to fit a model. It can be helpful to overlay
the model on a scatterplot of the data, to add confidence bands for the fit, or to
add predictive intervals for possible new points.
3. Smoothing
If no analytic model is proposed for Y as a function of X, then a nonlinear
smoother can be tried. loess (local weighted regression) is an interesting ap-
proach and is used in some situations in R as a default, for instance in plots of
model objects. Recently spline functions have been used more, partly because
they have better statistical properties and because it is easier to draw confidence
bands for them.
4. Bivariate density estimation
kde2d from MASS, kde from ks, or bkde2d from KernSmooth may be used.
The highest density region package hdrcde is also an interesting possibility. As
with univariate density estimation, there is no testing of the estimates produced,
a curious gap in theory.
5. Outliers
Points which are outliers in scatterplots may be outliers on one of the two di-
mensions individually or purely bivariate outliers. Deciding whether a point is a
bivariate outlier or whether several points in a group are outliers is tricky. Den-
sity estimators are one approach, but they can be problematic because of either
masking or swamping.
Looking for Structure: Dependency Relationships and Associations 95

Main points
1. Scatterplots can take many different forms and provide a lot of information about
the relationship between two variables (cf. §5.2). Look again at the movies scat-
terplot in Figure 5.7.

2. Adding lines or smooths to scatterplots is easy and often provides valuable guid-
ance, as can be seen in Figure 5.9, where the relationship between sons’ and
fathers’ heights is difficult to assess from the scatterplot alone.
3. Trellis displays are very effective for comparing scatterplots by subgroup, espe-
cially when the groups overlap (cf. Figure 5.11).
4. Sploms are excellent for giving a quick overview of a few variables. For example,
Figure 5.12 summarises the U.S. crime rates dataset well.

Exercises
1. Movie ratings
In Figure 5.7, there were a number of films with very high average ratings but
few votes.
(a) How does the scatterplot look if you exclude all films with fewer than 100
votes?
(b) What about excluding all films with an average rating greater than 9? In the
latter case, you would exclude at least one film with a lot of votes. What
limit would you choose and why?
2. Meta analysis
The results of 70 studies on thrombolytic therapy after acute myocardial infarc-
tion are reported in the Olkin95 datatset from the meta package.
(a) Draw a scatterplot of the number of observations in each experimental
group (n.e) against the corresponding number of observations in each con-
trol group (n.c). How would you summarise the plot?
(b) Over half of the studies involve fewer than 100 patients in each group. If
you restrict the scatterplot to this range, do you gain additional information?
96 Graphical Data Analysis with R

3. Zuni
The zuni dataset from the package lawstat was introduced in Exercise 5 in Chap-
ter 3.
(a) Plot a scatterplot of average revenue per pupil (Revenue) against the cor-
responding number of pupils (Mem). What information can you gather from
the plot?
(b) The distribution of the number of pupils is skew with one large outlier. How
does the scatterplot look if you plot Revenue against the log of the number
of pupils? What additional information have you discovered, if any? Is there
any point in logging average revenue per pupil as well?
4. Pearson heights
Pearsons’ height data for fathers and sons were considered in Sections 3.3 and 5.4.
(a) Draw a scatterplot of the heights. Are there any cases you would regard as
outliers?
(b) Draw a plot including both points and highest density regions. Which cases
would be regarded as outliers under this model, do you think?
(c) Fit a linear model to the data and a loess smooth and plot your results. Is a
nonlinear model necessary?
5. Bank discrimination
A subset of Roberts’ bank sex discrimination dataset from 1979 is available in
the package Sleuth2 under the name case1202. Consider the three variables mea-
sured in months, Senior (seniority), Age, and Exper (work experience prior
to joining the bank).
(a) Are there any notable features in the scatterplot matrix of the three vari-
ables? Can you explain them?
(b) Why do the scatterplots involving seniority not have the structure of the
scatterplot of experience against age?

6. Cars
In Figure 5.8, MPG.city was plotted against Weight. In many countries, fuel
performance is measured in litres per 100 km rather than in miles per gallon, in
effect the inverse criterion. If you plot 1/MPG.City against Horsepower, do
you get a linear relationship? Which cars would you describe as outliers now?
7. Leaves
The leafshape dataset in the DAAG package includes three measurements on
each leaf (length, width, petiole) and the logarithms of the three measurements.
(a) Draw sploms for the two sets of three variables. What conclusions would
you draw from each set? Which do you find more useful?
(b) Redraw the sploms, colouring the cases by the variable arch, describing
the type of leaf architecture. What additional structure can you see?
Looking for Structure: Dependency Relationships and Associations 97

8. Olive oils from Italy

The olive oils dataset is well known and can be found in several packages,
for instance as olives in extracat. The original source for the data is the paper
[Forina et al., 1983].
(a) Draw a scatterplot matrix of the eight continuous variables. Which of the
fatty acids are strongly positively associated and which strongly negatively
associated?
(b) Are there outliers or other features worth mentioning?
9. Boston housing
The Boston dataset was introduced in Chapter 3.
(a) Draw a splom of all the continuous variables (i.e., all except the variable
chas). Which variables are positively associated with medv, the median
home value?
(b) Several of the scatterplots involving the variable crim, the per capita crime
rate, have an unusual form, where higher values of crim only occur for one
particular value of the other variable. How would you explain this?
(c) There are many different scatterplot forms in the display. Pick out five and
describe how you would interpret them.
10. Hertzsprung-Russell
The Hertzsprung-Russell diagram is a famous scatterplot of the relationship be-
tween the absolute magnitudes of stars and their effective temperatures and is
over one hundred years old. Although examples of the plot can be found all over
the place, it is surprisingly difficult to find the data underlying them. There is a
dataset of 47 cases, starsCYG, in the package robustbase, but that is really too
small. The dataset HRstars with 6220 stars in package GDAdata is from the Yale
Trigonometric Parallax Dataset and was downloaded from [Mihos, 2005].
(a) Plot Y against X. How does your plot differ from the plots you find on
the web, for instance from a Google search for images of the Hertzsprung-
Russell diagram?
(b) The plots seem to use different numbers of stars. Are some more likely to
be used than others?
(c) You can colour and annotate your plot using techniques described in Chap-
ter 13. What would you suggest?
11. Intermission
The painting La Grande Jatte by Georges Seurat hangs in the Art Insitute of
Chicago, a classic of putting dots together to form an overall impression. Do you
think the artist intended his painting only to be viewed from a distance?
6
Investigating Multivariate Continuous Data

Numerical quantities focus on expected values, graphical summaries on un-

expected values.

John Tukey

Summary
Chapter 6 discusses parallel coordinate plots for studying many continuous variables
simultaneously.

6.1 Introduction
Scatterplots are superb for looking at bivariate data, but they are not so effective for
exploring in higher dimensions. Matrices of scatterplots are pretty good, but do not
really convey what multivariate structure there might be in the data, even with inter-
active linking. Dimension reduction methods (principal component analysis, factor
analysis, or multidimensional scaling) have their supporters, and like all graphics-
related approaches can provide insights when they are appropriate for the dataset
in hand. Their disadvantage is that they approximate the data, and it is difficult to
assess how good the approximations are. Rotating plots, which dynamically rotate
through two-dimensional projections, are a particularly flexible dimension reduction
approach. They are attractive and can uncover interesting information when com-
bined with projection pursuit indices, but are really a specialist tool.
In recent years, parallel coordinate plots have become popular for multivariate
continuous data. They were discussed in depth by Inselberg, and their geometry is
covered in detail in his book [Inselberg, 2009]. Wegman was the first to suggest their
use for data analysis [Wegman, 1990]. Chapter 1 of [Hartigan, 1975] talks about pro-
file plots, a very similar idea, which was inhibited in practice at the time by the com-
puting power available. It is interesting to look at Hartigan’s plots and be impressed
both by what he achieved then and by how far graphics have come since.

99
100 Graphical Data Analysis with R

6.2 What is a parallel coordinate plot (pcp)?

data(food, package="MMST")
names(food) <- c("Fat", "Food.energy", "Carbohyd", "Protein",
"Cholest", "Wt", "Satur.Fat")
ggparcoord(data = food, columns = c(1:7), scale="uniminmax") +
xlab("") + ylab("")

FIGURE 6.1: A parallel coordinate plot of the seven variables in the food dataset
from the MMST package. There are outliers and all the distributions look skewed
to the right, because most of the lines are packed into the foot of the plot. Setting
scale="uniminmax" means that each variable is scaled individually from its
minimum to its maximum.

With a scatterplot, the x and y axes are perpendicular to one another. In a parallel
coordinate plot all axes are parallel to one another. Each variable has its own individ-
ual vertical axis (or alternatively all the axes are horizontal) and the axis is usually
scaled from the minimum to the maximum case values for the variable, so that the
full range of each axis is used. The values of each case on adjacent axes are joined
by lines, so that a polygonal line across all axes defines a case. Hartigan called these
lines ‘profiles’, and that is a good way of thinking of them. Figure 6.1 shows a pcp
of the seven variables of the food dataset from the MMST package.
The food dataset has 961 cases. The variables all look skewed to the right, as
there are more lines at the bottom of the plot than at the top. One case is an extreme
value on several of the variables (the line starting top left). There appears to be a
small subgroup with similar sets of highish values (the group of lines close together
across the first five variables).
Izenman’s textbook [Izenman and Sommer, 1988] suggests dividing each food
Investigating Multivariate Continuous Data 101

value by the serving weight of the food and Figure 6.2 shows the corresponding
parallel coordinate plot (without the weight variable, of course). There are now a
couple of distinct groups at high levels of the fat by weight variable. The protein by
weight and the cholesterol by weight variables each have an extreme outlier.

food1 <- food/food$Wt

ggparcoord(data = food1, columns=c(1:5, 7),
scale="uniminmax", alphaLines=0.2) +
xlab("") + ylab("")

FIGURE 6.2: A parallel coordinate plot of the six variables in the food dataset di-
vided by the variable weight.grams, the weight of a serving. The transformed
carbohydrates variable is no longer skew and there are some distinct groups with
high fat levels. Individual outliers can be seen on cholesterol and protein and a small
group on saturated fat. The alpha-blending parameter alphaLines (§6.7) has been
used to make the plot easier to read. Each case is plotted with a weight of 0.2, so that
fully black lines represent at least 5 cases.

Parallel coordinate plots can include axes for categorical variables as well, where all
cases for a particular category are assigned the same numerical value. It is advisable
to avoid drawing the axes for categorical variables beside one another, as then many
lines are drawn on top of one another and it is difficult to see structure.

Functions for drawing pcp’s

There are several functions for pcp’s in R and ggparcoord is the one primarily
used in this chapter. It has a lot of options, both for what is displayed and how it is
displayed, which is a good thing, because there are a lot of possible ways of drawing
pcp’s and these can have a big impact on what you see. As always, different people
have different preferences and it is sensible to experiment a little to find out what
works best for you. Just settling for the defaults is a flawed strategy, particularly
102 Graphical Data Analysis with R

function package
ggparcoord GGally
parcoord MASS
cparcoord gclus
pcp PairViz
parallelplot lattice
parallel.ade epade
ipcp iplots
... ...

TABLE 6.1: pcp functions in R

when there are so many alternatives offered, but you do have to put in the work to
find out what the options do and how the resulting displays should be interpreted.
Other functions listed in Table 6.1 include parcoord, which offers basic func-
tionality including colouring, and cparcoord in gclus is a modified version of it,
as is pcp. The function parallel.ade offers a scaling alternative and so does
ipcp in iplots, which has the additional advantage of being interactive.
In fact, parallel coordinate plots really need to be interactive to be fully effective.
Highlighting cases which are of interest on one axis and seeing where they appear
on the other axes instantaneously is a simple and intuitive process. Specifying an ap-
propriate condition to define the group of interest and redrawing the plot each time
is laborious. Reordering axes on the fly is more effective than writing out the neces-
sary function and drawing the plot again. A good strategy is to use a tool like ipcp
initially to ascertain what information is of interest and how you want to display it,
and then draw the chosen plot using your favourite static pcp function.

6.3 Features you can see with parallel coordinate plots

[Wegman, 1990] suggested that it is possible to identify many different multivariate
features in parallel coordinate plots. This can be true in particular applications, but
in general the claim is too optimistic. What is definitely true is that having identified
outliers, correlations, or clusters with analytic approaches, parallel coordinate plots
are very useful for checking the results.
Parallel coordinate plots give quick overviews of univariate distributions for sev-
eral variables at once: whether they are skew, whether there are outliers, whether
there are gaps or concentrations of data (all of these can be seen in Figures 6.1 and
6.2). Bivariate associations between adjacent variables can sometimes be seen and
occasionally some multivariate structures, such as groups of cases with very similar
values across a number of variables or cases which are outliers on more than one
variable (as in Figure 6.1). These patterns are best checked and investigated further
by colouring the group of interest. Figure 6.3 shows an example for the food dataset,
Investigating Multivariate Continuous Data 103

drawn horizontally this time using coord_flip. Note that the cases have been re-
ordered to ensure that the selected cases are drawn last, so that those lines are on
top.

food1 <- within(food1,

fatX <- factor(ifelse(Fat > 0.75, 1, 0)))
ggparcoord(data = food1[order(food1$fatX),],
columns=c(1:5, 7), groupColumn="fatX",
scale="uniminmax") + xlab("") + ylab("") +
theme(legend.position = "none") + coord_flip()

FIGURE 6.3: A parallel coordinate plot of the six variables in the food dataset di-
vided by serving weight. The two subgroups with high values of fat per serving have
been selected using the derived variable fatX. As expected, some of these cases
have the highest values for saturated fat. This is another version of Figure 6.2.
104 Graphical Data Analysis with R

Wegman and others have suggested riffling through sufficient reorder-

ings of the axes to observe all possible adjacencies. In a similar vein,
[Hurley and Oldford, 2010] proposed adding extra copies of axes to include all the
adjacencies. If your main aim is to look for bivariate associations, then you are better
off with scatterplots. There is an example in Figure 6.2 where it looks as if the trans-
formed carbohydrates and protein variables may be negatively correlated because of
the lines crossing to form an ‘X’ shape. In fact the correlation is only -0.087, al-
though the relationship between the variables is worth looking at for other reasons,
as Figure 6.4 shows.

ggplot(food1, aes(Protein, Carbohyd)) + geom_point()

1.00 !
!
!
!
!

! ! !
!!
! !
!
!!
! !
!
! ! !! ! !
! ! !! !! ! !
! ! ! !
!!
0.75
! !
! !!! !
!!! ! ! !! !
! !
! ! !
!!
! !
!
! !! !
!! !
!!
! !! !! !
!
! ! !
! !
!!!
! ! !
!! !! ! ! !!!! !!
Carbohyd

!!
!!! !
! !!! ! !! !
!
! ! !! ! !
!!!! !!
! !!! !! !! !
0.50 !
! !! !! ! !!
! !! !
! ! ! ! !!!
!!
! !!!
! !! !
! !
!!
!! !
!
!!
!!! !!!
!
!!
! !
! ! !
! !! ! !
! ! ! !!
!!!
! ! !
!
! !
! !! !!
! !! !
0.25
!!!!! ! !!
!! !!
! ! ! ! ! !
!! ! !
!!!
!!!
!!!
!
!
!! ! !! ! ! !
! !
!
!!!!
! !!!!! ! ! ! ! !
! !!
!!!
! !!!! ! ! !
!!!!!
! !!! ! !!
!! !! ! ! !
!
!!!!! ! !!! !! ! ! ! !
!! ! ! !!
! !
!!
!!!!!!
! ! ! !!
! !
!
!
!
!
!!!!!!!
!
!! ! ! !
!
!
! ! !! !! !
!
!!
!
!
!
!
!
!
!!! ! !
! !! !!! ! !!
!
! !!
!!!
! !!!
!
! ! ! !
!!! !!
! !!!! ! !
!!
!!
! !!!
!! ! !
!
! !!!! ! ! !! ! !
!
! !
!
!
!!
!!!!!! !!!
! ! ! ! !
!!!!!!! !
!!!
!!!
!!
!!!
! !
!!
!
!!! !!! !
!!!
!
!!!
!!
!!
!!! !! !
!
!!!
!!
!!!!!! ! ! !! !!! ! ! ! ! !
!!!
!!
! ! !! !!
! ! ! ! ! ! !
! ! ! !
0.00
!! !!!!!! !
!!
!!!
!! !! !! !!!!!!
!!!! !
!!!
!!
!!
!!!!!
!!!
!!!! ! ! !

0.00 0.25 0.50 0.75

Protein

FIGURE 6.4: A scatterplot of carbohydrate and protein values standardised by serv-

ing weight. Most of the points have low protein values, and there are quite a few
foods with either no protein or no carbohydrates. The apparent diagonal boundary is
determined by foods composed of almost only protein and carbohydrates.
Investigating Multivariate Continuous Data 105

Fisher’s iris dataset has been used so often as an example for so many statis-
tical approaches, including in this book, because it makes for such an excellent il-
lustration. In a pcp of the data in Figure 6.5 you can see how clearly the setosa
species is separated from the other two species on the petal measurements, how it is
lower on three of the measurements, but higher on Sepal.Width, where there is
also a setosa outlier, how the two petal measurements almost separate the other
two species, and how additionally there are two unusual virginica values on
Sepal.Width.

ggparcoord(iris, columns=1:4, groupColumn="Species")

1 Species
setosa
value

versicolor
0
virginica

!1

!2

Sepal.Length Sepal.Width Petal.Length Petal.Width

variable

FIGURE 6.5: A parallel coordinate plot of the four variables in the iris dataset
coloured by species and using the ggparcoord default scaling, standardising by
the mean and standard deviation. The petal measurements separate the groups well
and there is evidence of cases which are a little different from the rest of their species.

There is no guarantee that a pcp will reveal interesting information about a dataset
or indeed any information. That applies to all graphics. What can be said is that when
graphics can be drawn so quickly and easily it is worth checking them out to see what
information they do reveal.
106 Graphical Data Analysis with R

6.4 Interpreting clustering results

Cluster analysis is a popular data analysis tool, although most of the methods have
little statistical basis and results are judged more by what users want to do with them
than by any statistical approach. Even if some well-separated clusters are obtained,
you still need to find a way of describing them and that means working out for which
variables or groups of variables they differ from one another and how. A productive
first step is to compare the clusters graphically with a parallel coordinate plot in
which the clusters are given different colours.
Figure 6.6 is a pcp of the USArrests dataset from 1973, coloured by the three
clusters found with an average link clustering, the first example on the hclust help
page. The Assault variable separates the clusters, and the reason becomes obvious
if you draw the same plot with the option scale="globalminmax", which puts
all the variables on a common scale: Assault has a much higher range of values
than the other three variables, so it dominates the multivariate distance calculations.

hcav <- hclust(dist(USArrests), method="ave")

clu3 <- cutree(hcav, k=3)
clus <- factor(clu3)
usa1 <- cbind(USArrests, clus)
ggparcoord(usa1, columns=1:4, groupColumn="clus",
scale="uniminmax", mapping = aes(size = 1)) +
xlab("") + ylab("") +
theme(legend.position = "none")

1.00

0.75

0.50

0.25

0.00

Murder Assault UrbanPop Rape

FIGURE 6.6: A parallel coordinate plot of the USArrests dataset. The three clusters
found with an average linkage clustering have been coloured. The clusters are defined
by the Assault variable, as can be seen by the clear separation of the three colours
on that variable. The size parameter has been used to make the lines thicker.
Investigating Multivariate Continuous Data 107

A clustering using the variables more equally could be carried out using
dist(scale(USArrests)) in hclust, as Figure 6.7 shows. The Assault
variable still separates the clusters fairly clearly on its own, but now there are only
two main clusters, not three, as one cluster is a state on its own. Because the dataset is
small, you can list the cluster members or draw a dendrogram with plot(hcav2)
to identify the outlying state as Alaska.

hcav2 <- hclust(dist(scale(USArrests)), method="ave")

clu32 <- cutree(hcav2, k=3)
clus2 <- factor(clu32)
usa2 <- cbind(USArrests, clus2)
ggparcoord(usa2, columns=1:4, groupColumn="clus2",
scale="uniminmax", mapping = aes(size = 1)) +
xlab("") + ylab("") +
theme(legend.position = "none")

1.00

0.75

0.50

0.25

0.00

Murder Assault UrbanPop Rape

FIGURE 6.7: A parallel coordinate plot of the USArrests dataset drawn using scaled
variables and average linkage clustering. The three cluster solution is now more of a
two cluster solution with one outlying case (which turned out to be Alaska).

Most applications in practice are not this simple and there is an extensive range of
additional clustering approaches as well as alternative distance measures that might
be used instead. The key point to remember is that whatever clustering method you
use it is valuable to explore the results you obtain graphically rather than just accept-
ing them without any further investigation.
108 Graphical Data Analysis with R

6.5 Parallel coordinate plots and time series

Time series are a special kind of data and demand special treatment. There is plenty
of support in R for analysing and displaying time series, outlined in the relevant
Task View, TimeSeries [Hyndman, 2013]. Interestingly, some time series can also
be plotted as pcp’s. Since this sheds light on both pcp’s and time series, it is worth
looking at briefly.
If data are recorded at regular intervals, multiple series can be displayed simul-
taneously in a pcp with each axis being one of the time points. Even if the data
are recorded irregularly, a pcp can still be used, provided that all series have been
recorded at the same times, although you have to bear in mind the distortion of the
time axis when interpreting any resulting patterns. As data are usually recorded every
day, every month, or every year, this is not often an issue. What is an issue is that
the time points have to be in the columns and the series in the rows, which either
requires a transposition of the data, or, if there are different sets of series, some data
reorganisation. This can be seen in the code needed for Figure 6.8, where reshape’s
melting and casting have been used. Note that it is essential in plotting time series
to use a common scaling for all axes (hence the choice of the scale option) and
that you cannot change the order of the axes because the time ordering is given. In
the parcoord default implementation any series with any missing values are com-
pletely excluded.
Figure 6.8 shows an example of acres of corn planted in states of the United
States over the past 150 years or so. The sharp rise for five states in the 19th century,
the dips at the time of the Depression in the 1930s and in the early 1980s, and various
unusual individual patterns are all visible. Identifying the individual states is not so
easy, using an interactive tool like ipcp in iplots is better for that. It turns out that
Kansas was one of the early ‘big’ states but declined erratically; Nebraska had the
steepest fall, but also some large increases; Iowa and Illinois were always ‘big’.
Colours were assigned by default alphabetically, which is why Iowa and Illinois
look alike. No legend was included to allow more space for the display. This is typical
for initial exploratory graphics, first check to see if there is anything worth studying
in more detail, then invest time in improving the plot.
Plotting many time series together gives a rapid overview, highlighting general
trends (if there are any) and picking out series which follow different patterns to the
rest. As always there is no guarantee that a particular graphic will provide immediate
information, perhaps there are no notable features to be found. What can be guaran-
teed is that being able to draw a range of plots quickly is an effective way of gaining
first insights into a dataset.
 Investigating Multivariate Continuous Data

FIGURE 6.8: A pcp of the acres of corn planted by US states from 1866 to 2011 from the nass.corn dataset in the agridat package.
Only the states with more than 250, 000 acres planted in 2011 and no missing values are included. Setting scale="globalminmax"
means the same scaling is used for all variables. Several interesting patterns can be seen, as discussed in the text.
library(reshape2); data(nass.corn, package="agridat")
c1 <- melt(nass.corn, id=c("year", "state"))
c1 <- within(c1, StateV <- interaction(state, variable)); c2 <- dcast(c1, StateV~year)
ggparcoord(subset(c2[1:48,], c2[1:48,147]> 250000), columns=2:147, groupColumn="StateV",
scale="globalminmax") + xlab("Year") + ylab("Acres") +
scale_x_discrete(breaks=seq(1865, 2015, 10)) + theme(legend.position = "none")
109
110 Graphical Data Analysis with R

A possible advantage of the pcp view is that it is trivial to switch to a cross-sectional

view. Have a look at Figure 6.9, where the only changes to the ggparcoord func-
tion for Figure 6.8 are the addition of boxplots and the use of alpha-blending to
downplay the lines. It is now possible to see that two of the states were individual
high outliers almost every year. The long-term trend for the states as a group, rep-
resented by the median, can be seen, which is not readily possible in Figure 6.8.
The years where many states planted lower numbers of acres are also easier to pick
out. Cross-sectional views are useful when we are interested in the distributions of
values at particular time points and enable us to make rough judgements as to when
individual series may be considered as outliers compared to the rest or not.
The cross-sectional view is also useful for aligning series in different ways (cf.
§6.7). Parallel coordinate plots may not be the best option for displaying time series,
but they are often a pretty good one. The time series examples illustrate how the
number of axes can be very high and yet pcp’s can still convey useful information.
 Investigating Multivariate Continuous Data

FIGURE 6.9: A parallel boxplot of the nass.corn data as specified Figure 6.8. The median trend of the number of acres planted by state
can be seen. It is also observable that two states (Iowa and Illinois) are outliers almost every year.
ggparcoord(subset(c2[1:48,], c2[1:48,147]> 250000), columns=2:147, groupColumn="StateV",
scale="globalminmax", boxplot=TRUE, alphaLines=0.5) + xlab("Year") + ylab("Acres") +
scale_x_discrete(breaks=seq(1865, 2015, 10)) + theme(legend.position = "none")
111
112 Graphical Data Analysis with R

6.6 Parallel coordinate plots for indices

Index values are weighted combinations of the values of their components. Stock
market indices are used to represent performances of groups of shares and the con-
sumer price index summarises prices across a wide range of products. Both of these
indices are time series, and an obvious graphic display is to overlay the index series
in bold on top of the series for the separate components. If a pcp is used for this, then
the axes are time points and the cases are the components. Alternatively, indices can
be constructed in the same way for different individuals and these individuals can
be compared by the single index value and additionally by the component values for
each one. In this situation a pcp can be drawn with one axis for the index itself and
separate axes for each of the components. The cases are then the individuals and the
polygonal lines are their profiles.
The dataset uniranks in the package GDAdata is an example of this kind of
data. 120 universities in the UK were ranked using a combination of eight cri-
teria [Guardian, 2013]. The AverageTeachingScore is the overall index and
the criteria are the components. Figure 6.10 is a pcp of the data with the variable
StudentStaffRatio inverted to a staff:student ratio, so that higher values mean
a higher overall score for the university. The vertical axis scale has been removed,
since it just goes from the minimum to the maximum for each variable, and there
might be a temptation to interpret a (0, 1) scale.
A number of features can be seen. High values on the first axis, the index, are
mostly associated with high values on the other variables, although not necessarily
strongly, and there are clear exceptions (for instance, the third-ranked university has
a relatively low value for NSSTeaching). The gaps for the three NSS variables re-
flect limited numbers of possible values for those variables. The up and down pattern
for the last few variables is probably due to their being skewed in different directions
rather than to any negative association. There are several cases which seem extreme
compared to the others on single variables, and there appear to be some bivariate
outliers as well, where individual lines go against the trend between two axes.
Plots of this kind are most effective when some individuals or groups are iden-
tified by colour. This works best interactively, when selections can be made directly
on the graphic and redrawing occurs immediately. With static graphics you need to
define a variable to specify the selection you want. If the selection is data-driven, for
instance depending on what you see in a first plot, then coding is necessary. If you are
interested in predefined groups it is easier and one option could be a trellis display
of pcp’s. Figure 6.11 shows an example where one of the groups of universities has
been picked out.
 Investigating Multivariate Continuous Data

FIGURE 6.10: A pcp of the uniranks dataset from the GDAdata package. Three poorly performing universities had at least one missing
value on one of the variables and have been excluded. There are a number of outliers and some of the measures are skew.
data(uniranks, package="GDAdata")
names(uniranks)[c(5, 6, 8, 10, 11, 13)] <- c("AvTeach", "NSSTeach", "SpendperSt", "Careers",
"VAddScore", "NSSFeedb")
uniranks1 <- within(uniranks, StaffStu <- 1/(StudentStaffRatio))
ggparcoord(uniranks1, columns=c(5:8, 10:14), scale="uniminmax", alphaLines=1/3) +
xlab("") + ylab("") + theme(axis.ticks.y = element_blank(), axis.text.y = element_blank())
113
114 Graphical Data Analysis with R

The Russell group is made up of 24 of the top UK universities. In Figure 6.11

they have been coloured red to highlight them. The order of the axes has also
been changed, based on what the variables represent, so that it now starts with
EntryTarif and ends with CareerProspects. On all variables bar one,
NSSFeedback, the Russell group have good scores. There is one high-ranking uni-
versity (St. Andrews, 4th), which is not a member of the Russell group and one,
Exeter, with a relatively poor staff:student ratio.

uniranks2 <- within(uniranks1,

Rus <- ifelse(UniGroup=="Russell", "Russell", "not"))
ggparcoord(uniranks2[order(uniranks2$Rus, decreasing=TRUE),],
columns=c(5:8, 10:14),
order=c(5,12,8,9,14,6,13,7,11,10),
groupColumn="Rus", scale="uniminmax") +
xlab("") + ylab("") +
theme(legend.position = "none",
axis.ticks.y = element_blank(),
axis.text.y = element_blank()) +
scale_colour_manual(values = c("red","grey"))

FIGURE 6.11: The uniranks dataset with the Russell group selected in red. These
universities scored well in general, although not on NSSFeedback. The code shows
that a new variable specifying the Russell group universities was derived and then
used to sort the cases (the first order). The order of the variables (columns) was
specified by the second order.
Investigating Multivariate Continuous Data 115

6.7 Options for parallel coordinate plots

When the pcp is interactive, so that cases can be selected and highlighted, then more
information can be uncovered. With static graphics it is necessary to use drawing
options in various combinations to get at the information.

Alignment
Instead of aligning the variables by their minimum and maximum values, they can be
aligned by their mean or median with a corresponding adjustment to the minimum
and maximum limits. This permits comparing the variability across variables better.
A particularly instructive example arises in plotting the detailed results from the Tour
de France cycle race. A selection of parallel coordinate plots has been used for this
on the ‘Statistical graphics and more’ blog ([Theus, 2013]) for a number of years to
present the Tour de France results.
The median time for any stage will represent the time taken by the main bunch
of riders, the peloton. This gives a sensible standard against which to judge each
rider’s performance. On the other hand, it is also informative to align each stage to
the time taken by the overall winner of the Tour, as it is immediately apparent on
which stages other riders gained time on him. ggparcoord provides an option for
specifying alignment, although only in conjunction with individual scaling of the
axes (scale="center" or "centerObs"). This means that for a comparison
of variabilities on a common scale, you have to transform the data first by subtracting
the centering statistic.
The following code shows an example for the nass.corn dataset from §6.5 align-
ing to the mean:

mz <- as.data.frame(apply(c2[1:48,2:147], 2,
function(x) x - mean(x,na.rm=TRUE)))
StateV <- c2[1:48,1]
mzA <- as.data.frame(cbind(StateV, mz))
ggparcoord(mzA, columns=2:147, scale="globalminmax",
groupColumn="StateV") +
xlab("Year") + ylab("Acres") +
scale_x_discrete(breaks=seq(1865, 2015, 10)) +
theme(legend.position = "none")
116 Graphical Data Analysis with R

Scaling
Whenever you want to compare measurements of different kinds you have to find
some way of putting them on an equivalent scale. This also arises with other mul-
tivariate displays (as will be discussed in §8.3) and with time series (§11.3). There
is a surprising number of alternative standardisations you can use and it is as well
to think carefully about which one is most appropriate. At any rate, you should al-
ways make clear which one you have used, if you present displays to others, and you
should equally always check what has been used when interpreting graphics prepared
by someone else.
The traditional default scaling for pcp axes depends on the respective minimum
and maximum values for each variable. In effect, the vertical scale is drawn from
0 (minimum) to 1 (maximum). (Note that this is not the default scaling used by
ggparcoord, which uses the standardised scaling mentioned below.) The basic
transformation for case i of variable j is
xi j − mini xi j
yi j =
maxi xi j − mini xi j

It is not obligatory to scale the variables individually, and if all variables have equiv-
alent measurement scales it may make sense to put them on a common scale. This
would apply if you had exam marks for students for several different subjects. You
could then scale all the axes from 0 to 100 (if those were the possible minima and
maxima) or from a suitable rounded figure below the minimum across all exams to a
rounded figure above the maximum across all exams.
Another approach is to standardise each variable, using either the standard devi-
ation or the IQR (interquartile range), a robust measure of variability. For instance,
xi j − x¯j
zi j =
sd(x j )

Figure 6.12 displays four different scalings for the body dataset from the gclus pack-
age. (The variable names have been abbreviated to prevent text overlapping.) Each
plot emphasises different aspects and each has a correspondingly different vertical
scale.
In the top left panel the outliers on individual variables are picked out. In the top
right panel the outliers are still visible and you can see different distributional shapes.
In the lower left panel the different value ranges of the variables are emphasised and
you get some idea of what scales there are and which variables have similar scales.
The display in the lower right panel suggests some possible clusters and offers a
different view of the outliers.
Investigating Multivariate Continuous Data 117

data(body, package="gclus")
body1 <- body
names(body1) <- abbreviate(names(body), 2)
names(body1)[c(4:5, 11:13, 19:21)] <-
c("CDp", "CD", "Ch", "Ws", "Ab", "Cl", "An", "Wr")
a1 <- ggparcoord(body1, columns=1:24, alphaLines=0.1) +
xlab("") + ylab("")
a2 <- ggparcoord(body1, columns=1:24, scale="uniminmax",
alphaLines=0.1) + xlab("") + ylab("")
a3 <- ggparcoord(body1, columns=1:24,
scale="globalminmax", alphaLines=0.1) +
xlab("") + ylab("")
a4 <- ggparcoord(body1, columns=1:24, scale="center",
scaleSummary="median", alphaLines=0.1) +
xlab("") + ylab("")
grid.arrange(a1, a2, a3, a4)

FIGURE 6.12: Four pcp’s of the 24 measurements for the 507 men and women in the
body dataset, each with a different scaling. Top left, the data for each variable have
been standardised; top right, each variable has been transformed to a min-max scale
individually; lower left, the variables have all been transformed to the same min-max
scale; and in the lower right panel the variables have been standardised individually
and centred on their medians. Interpretations are discussed in the text.
118 Graphical Data Analysis with R

Outliers
If a variable has outliers, then the rest of the data has to be squeezed into a small
section of the variable’s axis. Graphics are not robust against univariate outliers. If a
form of common scaling is used for a pcp, then this effect is even more restrictive,
as one variable with outliers affects the scale for all. While a pcp of the raw data
emphasises the outliers, you may want to redraw the plot without them. The code for
three alternatives is given below: removing the cases with any outliers, trimming all
outlier values to the chosen limits, restricting the plot to the chosen limits.
This code uses the outer fences of boxplots to define outliers, which means they
have to be further away from the hinges than three times the IQR. You could use any
limits you prefer (and can justify). In each case a function is written to transform the
data and applied to the variables to be plotted. Only the first display is shown.
• Remove the cases with outliers (Figure 6.13)
With 7 variables from the food dataset, this resulted in 260 of the 961 cases being
excluded, reflecting just how skew the data are. Without the outliers, you can ob-
serve regular gaps on the axes for both the fat and protein variables, suggesting
a limited discrete range of possible values. Given the large number of cases ex-
cluded, it would be unwise to draw too many conclusions. This approach works
better when there are only a few outliers to be excluded.

fc <- function(xv) {
bu <- boxplot(xv, plot=FALSE)$stats[5]
cxv <- ifelse(xv > bu, NA, xv)
bl <- boxplot(xv, plot=FALSE)$stats[1]
cxv <- ifelse(cxv < bl, NA, cxv)}
data(food, package="MMST")
rxfood <- as.data.frame(apply(food,2,fc))
ggparcoord(data = rxfood, columns = c(1:7),
scale="uniminmax", missing="exclude",
alphaLines=0.3) + xlab("") + ylab("")

FIGURE 6.13: The food dataset with outliers on individual variables removed.
Investigating Multivariate Continuous Data 119

• Trim all outliers to the chosen limits

fb <- function(xv) {
bu <- boxplot(xv, plot=FALSE)$stats[5]
rxv <- ifelse(xv > bu, bu, xv)
bl <- boxplot(xv, plot=FALSE)$stats[1]
rxv <- ifelse(rxv < bl, bl, rxv)}
data(food, package="MMST")
rfood <- as.data.frame(apply(food,2,fb))
ggparcoord(data = rfood, columns = c(1:7),
scale="uniminmax", alphaLines=0.3)

This plot includes all the cases, but 260 of them have been trimmed to the outer
hinges of at least one of the variables. It is top-heavy at the outer limits when
there are lots of outliers and all the outliers at each end of a variable are plotted
together.
• Restrict the plot to the chosen limits
fd <- function(xv) {
bu <- boxplot(xv, plot=FALSE)$stats[5]
bl <- boxplot(xv, plot=FALSE)$stats[1]
dxv <- (xv - bl)/(bu - bl)}
data(food, package="MMST")
rofood <- as.data.frame(apply(food,2,fd))
ggparcoord(data = rofood, columns = c(1:7)) +
coord_cartesian(ylim=c(0,1))

Lines going off the top and bottom of the plot are connections to outliers outside
the range of the plot. For this dataset, with so many outliers, it is not an effective
display.
Each of the three approaches defines cases as outliers, which are outlying on at
least one of the variables. The more variables there are, the larger the number of
cases which may be regarded as outliers. Cases which are multivariate outliers, but
not univariate outliers, are not excluded, because they do not affect the plot scales.
As always when dealing with outliers, these are subjective decisions that need
to be considered with care. A further alternative would be to transform the variables
with outliers. Sometimes a log transformation can be helpful, although you have to
watch for negative values and there may still be outliers even after a transformation.
You can see this in Figure 6.2, where an attempt was made to standardise the variables
using the serving weight variable. Note that a linear transformation of a variable has
no effect on its display in a default pcp, but could affect its display if outliers are
taken account of in some way.
120 Graphical Data Analysis with R

Variable order
Any multivariate display is affected by the order of the variables, and default al-
phabetic orders can be less than informative. In Figure 6.11, the order was man-
ually specified to better match the progression of students through the university.
ggparcoord has an option, order, offering a variety of data-driven alternatives,
all of which use scale-free statistics. One approach is to order the axes by the F
statistics from analyses of variance based on a grouping variable.

body1$Gn <- factor(body1$Gn)

ggparcoord(body1, columns=1:24, scale="uniminmax",
alphaLines=0.4, groupColumn="Gn",
order="allClass") + xlab("") + ylab("") +
theme(legend.position = "none",
axis.ticks.y = element_blank(),
axis.text.y = element_blank())

FIGURE 6.14: The body dataset with variable axes ordered by differences between
the genders as measured by F statistics. Women are in red. Forearm and shoulder
measurements have the highest F values.

In Figure 6.14 the cases of the body dataset have been ordered and coloured
by gender, while the variables have been ordered by F statistics. This static image
shows that the variables on the left discriminate clearly between men and women
(the men always having higher values) and that there are some outliers, such as the
woman with very high values on four variables in a row or the man with the smallest
measurement on An (ankle girth). It suffers badly from overplotting for the variables
where the genders overlap. Two graphics would be better, one for women with men
in grey in the background and one for men with women in grey in the background.
A trellis plot would draw only the men in one plot and only the women in the other.
Plotting the other gender in the background maintains context and simplifies keeping
the scales the same. Figure 6.15 shows the results. It is necessary to order the cases to
have the selected group plotted on top. The colours have been chosen to agree with
the colours used in Figure 6.14.
Investigating Multivariate Continuous Data 121

a <- ggparcoord(body1[order(body1$Gn),], columns=c(1:24),

groupColumn="Gn", order="allClass",
scale="uniminmax") + xlab("") + ylab("") +
theme(legend.position = "none",
axis.ticks.y = element_blank(),
axis.text.y = element_blank()) +
scale_colour_manual(values = c("grey","#00BFC4"))
b <- ggparcoord(body1[order(body1$Gn, decreasing=TRUE),],
columns=c(1:24), groupColumn="Gn", order="allClass",
scale="uniminmax") + xlab("") + ylab("") +
theme(legend.position = "none",
axis.ticks.y = element_blank(),
axis.text.y = element_blank()) +
scale_colour_manual(values = c("#F8766D","grey"))
grid.arrange(a,b)

FIGURE 6.15: pcp’s of the body dataset for men (above) and women (below) with
the cases for the other gender drawn in grey in the background. Compare with Fig-
ure 6.14, where overplotting is an issue, particularly for the variables to the right.
Outliers by sex are now much more apparent.
122 Graphical Data Analysis with R

FIGURE 6.16: A pcp of the body dataset with the variables sorted by their median
values. All variables are on the same scale. In general the variability increases with
higher values, especially for the top few variables. The variable in the middle with
relatively high variability is Age (which is a different kind of measurement).
m2 <- apply(body[, 1:24], 2, median, na.rm=TRUE)
m2a <- order(m2)
ggparcoord(data = select(body, -Gender), alphaLines=0.3,
scale="globalminmax", order=m2a) + coord_flip()

If you want to sort the variables by a statistic, such as the mean, the IQR, or the
maximum, which is not scale-free, you have to specify the order yourself. Figure 6.16
shows an example using the body dataset again, sorting by the median of the raw data
and presenting all variables on the same scale.
If the sorting should only be done on a statistic for a particular subset of the cases,
use the appropriate subset in the calculation of the statistic. The following code has
been modified to use only the subset for which Age < 30 in specifying the order:

m3 <- apply(body1[body1$Ag < 30, 1:24], 2, median, na.rm=TRUE)

You will then probably want to emphasise the subset used for sorting using colour
(cf. §6.7 below).
Investigating Multivariate Continuous Data 123

These sortings are not scale-free and have been carried out on the original data.
This is fine if the original variables have comparable scales, but not always advis-
able otherwise. To sort on comparable scales, you have to convert the data yourself
first as in §6.7 or, if the transformation you want is one of the options available in
ggparcoord, use that function to do the work.
For instance, in Figure 6.17 the code orders the pcp axes by the maximum value
of the variables standardised by subtracting the mean and dividing by the standard
deviation. First of all a pcp of the variables in default order is prepared with that trans-
formation. Then the transformed data are extracted from the plot using a reshape2
function and ordered. Finally, the pcp with the desired ordering is produced.

FIGURE 6.17: A pcp of the body dataset with the variables sorted by their maximum
values after standardisation by mean and standard deviation. Possible outliers are
easier to see, particularly the case that is outlying on the two rightmost variables.
B1 <- ggparcoord(data = body1, columns=c(1:24), scale="std")
B2 <- acast(B1$data[ ,c(1,3,4)], .ID ~ variable)
m4 <- apply(B2, 2, max, na.rm=TRUE)
m4r <- order(m4)
ggparcoord(data = body1, alphaLines=0.3,
columns=c(1:24), scale="std", order=m4r)

Sorting is a surprisingly powerful tool for data analysis, whether you sort vari-
ables or cases or both. Ideally, it should be easier to do for parallel coordinate plots.
That would mean presenting a limited range of alternatives in a simple structure and
might not cover all the options we would like to have. This is a topic that deserves
further thought.
124 Graphical Data Analysis with R

Formatting
Formatting includes another raft of choices and only some are listed here. Mundane
decisions like the choice of window size, labelling the axes, or where to put a legend,
can all affect what you can see in a display. While this is true for any graphic, it is
more important for pcp’s, which sometimes need all the window width they can get.

Type of display The standard pcp represents cases by polygonal lines. The individ-
ual points can be plotted on each axis with the option showPoints=TRUE.
This is useful for datasets with few cases; otherwise it only increases the clut-
ter in an already densely packed display. Another option is to plot boxplots for
each variable with boxplot=TRUE. This can give a feel for the distributions of
the variables, especially when there is a large number of cases. Alpha-blending,
discussed below, can be used to hide the lines completely or to downplay the
lines enough for the boxplots to stand out, while still enabling the lines to convey
useful information (cf. Figure 6.9). If the lines are hidden completely, the plot
looks like a collection of individual boxplots, just as you might get from drawing
boxplots of one variable for different subsets of a dataset. In a pcp boxplot, each
case appears in each boxplot, while in a boxplot by subset, each case generally
appears only once in the whole plot.
Missings The default in ggparcoord is to exclude cases with any missings. That
can lead to too many cases being dropped. Other options are to impute missing
values. Omitting profile sections for which one end point is missing is not cur-
rently possible. As ggparcoord decides which cases to exclude after scales
are calculated, not before, a case not shown may have determined the scale used.
Aspect ratio Parallel coordinate plots are best drawn wide and moderately high.
The appropriate window size depends on the number of variables plotted and the
structure of the data. As usual, it is worthwhile experimenting interactively with
a few different sizes to see what information is revealed.
Orientation Drawing pcp’s horizontally rather than vertically offers the advantage
of writing variable names horizontally one above the other, so that no problems
arise with overlapping text. An example was shown in Figure 6.3. It is more
common to draw pcp’s vertically.
Lines Lines can be reformatted in several ways. In principle, their thickness can be
varied. In practice, given the large number of lines, they should always be very
thin. The most important formatting effects are colour and alpha-blending.
Colour Lines can be coloured in groups by a factor variable or on a continuous
shading scale by a numeric variable. Figure 6.18 shows examples of both. In
the upper plot, a new binary variable was created to label the 16 Boston areas
having the maximum medv value of 50 as ‘High’, and the lines are coloured
accordingly. In the lower plot, the lines are coloured according to the case values
of medv, the median housing value for each area. The cases in the upper plot are
ordered to ensure that the group of interest is drawn last.
Investigating Multivariate Continuous Data 125

data(Boston, package="MASS")
Boston1 <- within(Boston,
hmedv <- factor(ifelse(medv == 50,"Top", "Rest")))
Boston1 <- within(Boston1, mlevel <- ifelse(medv==50,1,0.1))
Boston1 <- within(Boston1, medv1 <- medv)
a <- ggparcoord(data = Boston1[order(Boston1$hmedv),],
columns=c(1:14), groupColumn="hmedv",
scale="uniminmax", alphaLines="mlevel",
mapping = aes(size = 1)) + xlab("") + ylab("") +
theme(axis.ticks.y = element_blank(),
axis.text.y = element_blank())
b <- ggparcoord(data = Boston1, columns=c(1:14),
groupColumn="medv1", scale="uniminmax") +
xlab("") + ylab("") +
theme(axis.ticks.y = element_blank(),
axis.text.y = element_blank())
grid.arrange(a,b)

FIGURE 6.18: Plots of the Boston dataset with cases coloured by the variable medv,
the median home value. In the upper plot, the 16 cases with a value of 50 have been
coloured differently and drawn last, so that none of the other 490 cases in the dataset
are drawn on top of them. The other cases have been downplayed by using alpha-
blending, and all cases have been drawn with a thicker line to help the selected group
stand out better. The group differs a little from the rest, but perhaps not as clearly
as might have been thought beforehand. In the lower plot, the cases are shaded by
their value on the variable medv, an interesting idea technically, although not one
that seems to work well for this application.
126 Graphical Data Analysis with R

Alpha-blending Parallel coordinate plots suffer from overplotting. If you display

ten variables for 5, 000 cases, then you need 50, 000 points and 45, 000 line seg-
ments. Alpha-blending can be used to lessen the problem, although how much
it succeeds in solving it depends on the dataset and on what information is con-
tained in it. The option alphaLines in ggparcoord can be used, as we have
seen already in Figure 6.2 and Figure 6.10.
To make some groups stand out more than others, ggparcoord offers the
possibility of applying a user-specified variable in which the level of alpha-
blending can be set for each case individually. Figure 6.19 shows a pcp of the
Boston dataset in which the 132 cases with the maximum value of rad have
been coloured differently and the other cases have been drawn with an alpha-
blending value of 0.1. The order of the variables has also been changed using the
ggparcoord option order.

Boston1 <- Boston1 %>% mutate(

arad = factor(ifelse(rad < max(rad), 0, 1)),
aLevel = ifelse(rad < max(rad), 0.1, 1))
ggparcoord(data = Boston1, columns=c(1:14),
scale="uniminmax", groupColumn= "arad",
alphaLines="aLevel", order="allClass") +
xlab("") + ylab("") +
theme(legend.position = "none",
axis.ticks.y = element_blank(),
axis.text.y = element_blank())

FIGURE 6.19: A pcp of the Boston dataset. The cases with the highest value of rad
(index of accessibility to radial highways), have been given a different colour and
the remaining cases have been de-emphasised using alpha-blending. The order of
the variables has been changed to accentuate the differences between the two groups
of cases. Interestingly, the five variables in which the max(rad) cases all have the
same value, rad, tax, ptratio, zn, and indus, are not placed together. They
also have lower values for dis, higher values for crim, tend to have higher values
for age, and, with a few exceptions, have lower values for medv.
Investigating Multivariate Continuous Data 127

6.8 Modelling and testing for multivariate continuous data

1. Outliers
As discussed several times in the book, there are no fully satisfactory ways of
testing for outliers and you need to consider the individual cases or try some
robust approach. If cases are outliers on several variables at once, then that is
some evidence for viewing them with suspicion. Interactive graphics are the best
approach here, as the order of variables in a pcp determines whether you can
recognise such outliers.
2. Regular gaps in the distribution of a variable
Regular gaps can be due to a measurement feature. For instance, if decathlon
performance data are plotted, both the high jump and pole vault events will have
gaps, since only certain performances are possible.
3. Clusters of cases
Occasionally certain groups of cases stand out as having very similar values
across a set of variables. This might be tested by comparing that group’s means
with the rest or by comparing the multivariate distance of the group from the rest.
4. Separated groups
Sometimes there are clearly separated groups on particular variables. Whether
this means anything depends on the values those cases take on other variables,
especially categorical ones. Linear models can be useful in assessing the features.

6.9 Parallel coordinate plots and comparing model results

With the computing power that is available nowadays it is becoming more and more
common to consider large numbers of models in an analysis. If many different mod-
els are fitted to a dataset, it can be a complex matter to compare all the results. Model
fitting statistics such as AIC offer a quick filtering mechanism, but are not informative
enough for in-depth comparisons. A number of visualisation approaches have been
proposed including using a range of parallel coordinate plots to aid the comparisons
[Unwin et al., 2003]. Apart from plotting the raw data variables, they suggest plot-
ting the t statistics for comparable parameter estimates with one axis per parameter,
and plotting the residuals both by model and by case.
128 Graphical Data Analysis with R

Main points
1. Parallel coordinate plots are a powerful multivariate display, showing all contin-
uous variables at once (e.g., Figures 6.1 and 6.10).
2. There are many pcp display options in addition to the usual formatting options
applicable to other graphics: the choice of variables, the order of variables, what
scaling is used, and how the axes are aligned (§6.7). Choosing good options is
important. Figures 6.14 and 6.19 give examples.
3. pcp’s are helpful for exploring and evaluating the results of analyses such as dis-
criminant or cluster analyses (cf. Figure 6.6). They are also useful for checking
multivariate outliers, which will be further discussed in §9.3.
4. pcp’s are an alternative means for presenting multiple regularly spaced time se-
ries and offer the option of a cross-sectional view of the series (§6.5 and Fig-
ure 6.9).
5. pcp’s are informative for displaying indices and their component parts together
(§6.6).
6. You may need to draw several pcp’s to uncover the different features in a dataset
and other graphics will be useful as well (e.g., the various displays of the food
dataset in Figures 6.1 to 6.4).

Exercises
1. Swiss
The dataset swiss contains a standardized fertility measure and various socio-
economic indicators for each of 47 French-speaking provinces of Switzerland in
about 1888. It was already used in Exercise 6 in Chapter 1.
(a) Draw a parallel coordinate plot of all six variables.
(b) Are there any cases that might be outliers on one or more variables?
(c) What can you say about the distribution of the variable Catholic?
(d) Construct a new variable with values ‘High’ for all provinces with more
than 80% Catholics and ‘Lower’ for the rest. Draw a pcp coloured by the
groups of the new variable. How would you describe the provinces with a
high level of Catholics?
Investigating Multivariate Continuous Data 129

2. Pottery
The package HSAUR2 includes a dataset on the chemical composition of
Romano-British pottery, pottery, with 45 cases and 10 variables.
(a) Draw a pcp of the nine composition variables. What features can you see?
(b) Make a new variable with the cases with low values on MgO. How are these
cases different from the rest on the other variables?
(c) Colour your original pcp using the site information, kiln. Which kilns can
be easily distinguished from the others using which variables?
3. Olive oils
The olive oils dataset was introduced in Exercise 8 of Chapter 5.
(a) Draw a default parallel coordinate plot and describe the various features
you can see.
(b) Draw the same plot and additionally colour the oils by the region they come
from. What additional information can you find?
(c) Discuss which features of the dataset are easier to see with a pcp and which
are easier to see with a scatterplot matrix.
4. Cars
The dataset Cars93 was introduced in §5.4. Draw a pcp of the nine variables
Price, MPG.city, MPG.highway, Horsepower, RPM, Length, Width,
Turn.circle, and Weight.
(a) What conclusions would you draw from your plot?
(b) What plot would you draw to compare US cars with non-US cars on these
variables? What does the plot tell you about the differences between US
cars and the others?
(c) Is a pcp with unimax scaling informative? Try colouring it by the factor
variable Cylinder to gain additional insight.
5. Bodyfat
The dataset bodyfat is available in the MMST package. It provides estimates of
the percentage of body fat of 252 men, determined by underwater weighing, and
body circumference measurements. The dataset is used as a multiple regression
example to see if body fat percentage can be predicted using the other measure-
ments. Draw a parallel coordinate plot for the dataset.
(a) Are there any outliers? What can you say about them?
(b) Can you deduce anything about the height variable?
(c) What can you say about the relationship between the first two variables,
density and bodyfat?
(d) Do you think the ordering of the variables is sensible? What alternative
orderings might be informative?
130 Graphical Data Analysis with R

6. Exam marks
In the package SMPracticals, there is a dataset mathmarks with the marks out
of 100 in five subjects for 88 students. The dataset is fairly old, first appearing in
the statistical literature in [Mardia et al., 1979] and it was used in an example at
the end of §5.6. It is interesting to note that all students had marks in all subjects.
Possibly students who missed an exam were excluded.
(a) Explore the dataset using pcp’s. What information can you uncover and
which pcp would you use to present your results to others?
(b) Apparently the first two exams (mechanics and vectors) were closed
book, while the other three were open book. Draw a pcp with boxplots
to see if there is evidence that the students got lower marks on closed-
book exams. Is it useful to superimpose the polygonal lines (possibly using
alpha-blending) or not?
7. Wine
The wine dataset can be found in the packages gclus, MMST, pgmm,
and rattle. They took the data from the UCI Machine Learning Repository
[Bache and Lichman, 2013]. The original source is an Italian software package
[Forina et al., 1988]. The version in pgmm has about twice as many variables as
the others, and the version in MMST includes the names of the three classes of
wine, rather than the numeric coding that the other versions use.
(a) Use pcp’s to investigate how well the variables separate these classes.
(b) Are there any outliers?
(c) Is there evidence of subgroups within the classes?
8. Boston housing
Carry out a cluster analysis of the Boston data using Ward’s methods
(method=ward.D2) on standardised variables and choose the four cluster so-
lution. You could present your results in a number of ways:
(a) with a single pcp of the variables with the case profiles coloured by cluster;
(b) with several pcp’s, one for each set of cluster cases (at the time of writing
this will not work with ggparcoord if all the cases in a cluster have the same
value on a variable when a default or uniminmax scaling is used);
(c) with several pcp’s, one for each set of cluster cases and with the remaining
cases plotted in the background.
What are the advantages and disadvantages of the three alternatives? What plot
or group of plots would you choose for displaying your clustering results?
9. Intermission
Jackson Pollock’s Convergence is in the Albright-Knox Art Gallery in Buffalo,
New York. What can you see in this picture?
7
Studying Multivariate Categorical Data

Most statistical tables are parchingly dry in the reading.

Herman Melville (Moby Dick)

Summary
Chapter 7 discusses ways of displaying combinations of categorical variables using
various types of mosaicplot.

7.1 Introduction
One-dimensional displays for categorical data like barcharts and piecharts were dis-
cussed in Chapter 4. Displays for multivariate categorical data have been discussed
in research publications, but have received less attention in applications. The many
splendid books on the analysis of categorical data generally include few graphics.
The ideal is to display the counts of cases in the multivariate combinations in an
informative and easy to interpret way.
When there are several variables, each with a few categories, there can be an
unexpectedly large number of possible orderings of variables and categories and it
is a challenge to find an effective display. For J nominal variables {X j , j = 1, . . . , J}
with numbers of categories {c j , j = 1, . . . , J}, the variables can be ordered in J! ways
and the categories within the variables in ΠJj=1 c j ! ways giving J!ΠJj=1 c j ! orderings
in all. In the dataset Alligator in the package vcdExtra there are J = 4 variables:
lake (4 categories), sex (2), size (2), and food (5), making 276, 480 orderings
in all, an impressively large number considering that there are only 219 observations.
As examples will show, there are often compelling arguments for choosing particular
orderings, either from the logic of the application or for data-driven reasons.
Now that modern computing power and software have made drawing graphics
much easier, you can explore a range of displays to identify the ones that work
best. Many types of graphics have been suggested for multivariate categorical data,

131
132 Graphical Data Analysis with R

including mosaicplots, doubledecker plots, fluctuation diagrams, treemaps, associ-

ation plots, and parallel sets/categorical parallel coordinate plots ([Friendly, 2000],
[Meyer et al., 2006], [Kosara et al., 2006], and [Pilhoefer and Unwin, 2013]). In this
chapter mosaicplots and their variants will be used. They provide a wide range of
possibilities for the display of multivariate categorical data.

7.2 Data on the sinking of the Titanic

doubledecker(Survived ~ Sex, data = Titanic,

gp = gpar(fill = c("grey90", "red")))
doubledecker(Survived ~ Class, data = Titanic,
gp = gpar(fill = c("grey90", "red")))

Survived Survived
No

No

Yes

Yes

Male Female Sex 1st 2nd 3rd Crew Class

FIGURE 7.1: Survival rates on the Titanic. Females and first class passengers had
the best survival chances.

The dataset Titanic was used in Chapter 1 Exercise 5 and discussed in §4.3. Barcharts
of the dataset’s four variables were displayed in Figure 4.4 to give a simple overview.
The differing survival rates by groups are of more interest and Figure 7.1 shows
spineplots (see §7.3 below) of how they vary by Sex and Class individually. As
we all know, a much higher proportion of females survived than males and it is
perhaps no surprise that the survival rate was higher for the first class than the second
class and that the third class rate was lower again. Does the same class effect apply
separately for males and females? A doubledecker plot (§7.4), a generalisation of
spineplots, and a form of mosaicplot, can be drawn to investigate.
Figure 7.2 reveals what happened. The four bars to the left show that the male
rates do not decline with class and that the males in second class had the lowest
survival rate of all groups. The four bars to the right show that for the females, the
survival rate declined with class, although of the few females in the crew a high
Studying Multivariate Categorical Data 133

doubledecker(Survived ~ Sex + Class, data = Titanic,

gp = gpar(fill = c("grey90", "red")))

Survived
No

Yes

1st 2nd 3rd Crew 1st 2nd 3rd CClass

Male Female Sex

FIGURE 7.2: A doubledecker plot of the Titanic survival rates by class for each sex.
The width of each bar is proportional to the number of cases for that sex and class. All
female survival rates were higher than all male survival rates and the male survival
rate was lowest for second class passengers.

proportion survived. It is fairly easy to see that all female survival rates are higher
than all male survival rates. However, if the main aim was to compare female and
male rates directly by class, another plot would be better. How to choose which
mosaicplot to draw in any given situation is discussed in §7.5.

7.3 What is a mosaicplot?

In a mosaicplot the graphics area is divided up into rectangles proportional in size
to the counts of the combinations they represent. Lengths are easier to judge and
compare than areas, so it is best to use displays where each rectangle has the same
width or the same height.
Mosaicplots are drawn by starting with an empty rectangle to represent the whole
dataset, then taking the first variable and dividing the horizontal axis into sections
proportional to the sizes of its categories. In the second step each of the rectangles is
divided along its vertical axis according to the sizes of the second variable categories,
as in the two examples in Figure 7.1. In theory you can continue to divide up the
rectangles alternately horizontally and vertically for as many variables as you have.
In practice, the display can quickly have too many combinations and it becomes
impossible to see what is going on.
134 Graphical Data Analysis with R

Figure 7.3 shows a sequence of default mosaicplots for the Alligator dataset men-
tioned in §7.1, beginning with the lake variable and adding the variables sex,
size, and food in turn. The category names have been abbreviated to improve the
legibility of the labelling, that is always a problem with high-dimensional mosaic-
plots. The first plot shows that the lakes each provide about the same amount of data;
the second that Lake Hancock has more female alligators than the others; the third
that generally there are more small alligators than large ones, except for the males
in Lakes Oklawaha and Trafford; the fourth plot that the feeding patterns vary. A
different ordering of the variables might reveal other information.
The study concerned the eating preferences of alligators in four different lakes,
determined by examining the contents of their stomachs. The webpage [Brain, 2004]
has this to say about alligators’ feeding habits: “Once a week is a typical feeding
schedule for alligators living in the wild. Excess calories are stored in fat deposits at
the base of the alligator’s tail. Incredibly, by burning fat reserves, it is possible for an
alligator to last more than two years between feedings.” This dataset was first used
in one of Agresti’s excellent textbooks on categorical data analysis [Agresti, 2007].
The final plot with all four variables displays rectangles for each of the 80 possible
combinations and dotted lines for combinations that were not observed. You can see
that some combinations did not arise in the dataset, but not a lot else.
This example exemplifies the difficulty of displaying multivariate categorical data
effectively and also the difficulty of labelling mosaicplots. In general, there are too
many awkwardly aligned combinations to provide a well-defined labelling. The vcd
package usually does a reasonable job under difficult circumstances.
For a limited number of variables with only a few categories, mosaicplots can
work well, although they need careful design. So what is to be done? Given the
number of different orderings calculated in §7.1, the various alternative mosaicplots
defined in §7.4, and the additional options outlined in §7.6, there are many, many
possibilities for finding a better plot and it is necessary to make informed choices
and do more than just use a default.
A reasonable strategy is to begin with barcharts of the individual variables to get
an initial overview of the dataset. Often some of the features you can discern with
difficulty in higher dimensional displays are directly visible in lower dimensional
ones. For the Alligator dataset you could use code such as

data(Alligator, package="vcdExtra")
Alg1a <- aggregate(count~food, data=Alligator, sum)
ggplot(Alg1a, aes(food, count)) + geom_bar(stat="identity")

to draw each of the barcharts. If you did, you would see that there were more
males than females and more small alligators than large ones, that the numbers were
roughly equal for the four lakes, and that fish and invertebrates were clearly the most
frequently found foods.
Studying Multivariate Categorical Data 135

data(Alligator, package="vcdExtra")
Alg1 <- Alligator
levels(Alg1$lake) <- abbreviate(levels(Alg1$lake), 3)
levels(Alg1$size) <- abbreviate(levels(Alg1$size), 3)
levels(Alg1$food) <- abbreviate(levels(Alg1$food), 2)
par(mfrow=c(2,2), mar=c(4 ,4, 0.1, 0.1))
mosaicplot(xtabs(count ~ lake, data=Alg1), main="")
mosaicplot(xtabs(count ~ lake + sex, data=Alg1), main="")
mosaicplot(xtabs(count ~ lake + sex + size, data=Alg1),
main="")
mosaicplot(xtabs(count ~ lake + sex + size + food, data=Alg1),
main="")
Grg Hnc Okl Trf Grg Hnc Okl Trf

female
sex
male

lake lake

Grg Grg
lrg sml lrg Hnc
sml lrg Okl
sml lrg Trf
sml lrg sml lrg Hnc
sml lrg Okl
sml lrg Trf
sml
br
female

rpot in female
fs
sex

sex

br
fs
male

male
rp ot in

lake lake

FIGURE 7.3: A sequence of mosaicplots for the Alligator dataset. The first plot is a
spineplot of the variable lake, the second adds the variable sex, the third size,
and the fourth food. The category labels have already been abbreviated, so only
drawing the plots bigger would alleviate the overplotting of the labels. The first three
plots give information on the relative importance of the four lakes in the study, on
the sex distribution across the lakes, and on the distribution of alligator sizes by lake
and sex. The final plot attempts to present the four-dimensional information.
136 Graphical Data Analysis with R

pairs(xtabs(count ~ ., Alligator))

lake
George Oklawaha

63 55 48 53

Hancock Trafford
sex
female

89 130

male
size
large

95 124

small
food
bird invert reptile

13 94 61 32 19

fish other

FIGURE 7.4: A plot matrix of barcharts and two-dimensional mosaicplots for the
Alligator dataset. All pairs of variables barring sex and food appear to be related,
since the corresponding bivariate plots do not take the form of a grid.

Alternatively you could use the pairs function in vcd, giving Figure 7.4. This
produces a matrix display with the barcharts of the individual variables down the
diagonal as well as all six mosaicplots both above and below the diagonal. Labelling
is kept to a minimum for space reasons and, while efficient, the plot is hard to read.
Nevertheless, it can be seen that apart from the two variables sex and food, each
pair of variables looks related in one way or another. There are more large males,
small alligators eat more invertebrates, the lakes have different patterns of alliga-
tors’ food preferences. It is always worthwhile to check results in more detail, just
to be sure, and so it would be good practice to pick out some of the interesting mo-
Studying Multivariate Categorical Data 137

saicplots in Figure 7.4 and draw them individually. Figure 7.5 shows two examples,
doubledecker plots of sex and lake and of size and food.
At this stage, rather than drawing versions of the four possible three-dimensional
mosaicplots it would be practical to incorporate the aim of the study, to see how
food depends on the other three variables. As food has five categories with no
particular order a multiple barcharts view (§7.4) could be best. Figure 7.6 shows how
food depends on sex and size of the alligators.

doubledecker(xtabs(count ~ lake + sex, data = Alligator),

gp = gpar(fill = c("grey90", "steelblue")))
doubledecker(xtabs(count ~ food + size, data = Alligator),
gp = gpar(fill = c("grey90", "tomato")))

sex size
female

large

male

small

George Hancock Oklawa Trafford lake bi fish invert othe rep food

FIGURE 7.5: Doubledecker plots for the Alligator dataset. From the left-hand plot
you can see that the sex distribution varies by lake and from the right-hand plot
that the variables size and food are associated.
138 Graphical Data Analysis with R

Alg2 <- aggregate(count~sex + food + size, data=Alligator, sum)

ggplot(Alg2, aes(food, count, fill=sex)) +
geom_bar(stat = "identity") +
facet_grid(size ~ sex) + theme(legend.position="none")

female male

30

20

large
10

0
count

30

20

small
10

0
bird fish invert other reptile bird fish invert other reptile
food

FIGURE 7.6: A multiple barcharts view showing how food depends on sex and
size in the Alligator dataset. Males and females of the same size have similar pat-
terns, but large alligators prefer fish while smaller ones like invertebrates too.

A further step would be to include the lake variable and Figure 7.7 shows a
multiple barcharts view with the food distributions for the four combinations of
sex and size for each of the four lakes separately.
Figure 7.7 is obviously a better choice of display than the default mosaicplot for
all four variables in Figure 7.3, as it is much easier to recognise features of the dataset.
While the differences between the lakes are substantial, it is necessary to be careful in
drawing too firm conclusions given the small numbers involved. Loglinear modelling
would be helpful in determining which features of the data have strong support in
the data, while graphical displays provide an appreciation of the raw data and an
understanding of model results. As always, graphics and modelling complement each
other.
To improve the display clarity of mosaicplots, narrow gaps are left between the
category rectangles. With two variables you have a kind of barchart called a spineplot
[Hummel, 1996]; with spineplots there are no gaps between the categories of the
second variable. Without gaps the conditional rates of the second variable, P(B|A),
are directly displayed. As in a barchart, the areas of the bars are proportional to the
counts, but instead of the bars having equal widths and differing heights, they all have
equal heights and differing widths. Spineplots were designed primarily for interactive
graphics to display rates by category for a second linked variable. Gaps are useful if
Studying Multivariate Categorical Data 139

ggplot(Alg1, aes(food, count, fill=sex)) +

geom_bar(stat = "identity") +
facet_grid(lake ~ sex + size) +
theme(legend.position="none")

female female male male

lrg sml lrg sml
15

10

Grg
5

0
15

10

Hnc
5
count

0
15

10

Okl
5

0
15

10
Trf

0
br fs in ot rp br fs in ot rp br fs in ot rp br fs in ot rp
food

FIGURE 7.7: A multiple barcharts view showing how food depends on sex, size
and lake in the Alligator dataset. There are different patterns for the four lakes, but
numbers are quite small.

you are to divide the resulting rectangles according to the category proportions of a
third variable.
The construction of mosaicplots is hierarchical and the order of the variables has
a big impact on the display. There are other options to consider as well. Instead of
the classical alternating horizontal and vertical divisions, mainly horizontal divisions
might be used, as is the case with doubledecker plots (cf. Figure 7.2), or indeed any
sequence of horizontal and vertical divisions might be used.
140 Graphical Data Analysis with R

Classical mosaicplots go back to the paper [Hartigan and Kleiner, 1981] and aim
to make the most efficient use of the space available. This means that the area rep-
resenting each case is maximised, while the ease of interpretation suffers, as the
rectangles representing the combinations are not aligned for direct comparison. Even
with a small number of combinations a classical mosaicplot may prove to be inef-
fective. With a large number of combinations, no mosaicplot is likely to work. For
anything from 4 to 24 combinations, mosaicplots can be very good. Above that, their
usefulness depends on the structure of the data.

7.4 Different mosaicplots for different questions of interest

We have already seen examples of classical mosaicplots, doubledecker plots and
multiple barcharts. There are other alternatives too. Each mosaicplot variant presents
the data in a slightly different way and is useful for different purposes. Sometimes
counts are emphasised, sometimes rates, and sometimes distributions. In all cases it
is essential to remember that we are looking at conditional distributions and that the
order of the conditioning matters.

Which subgroups appear most often? (Fluctuation diagrams, Figure 7.8)

In this display, the axes are divided up equally for each category of a variable irre-
spective of size, and the rectangles representing the combination counts are drawn
at their appropriate grid points. The rectangle for the combination with the highest
count determines the scale, given the variables and the size of the window.
Fluctuation diagrams are good for representing large contingency tables or tran-
sition matrices, where there is no reason to differentiate between the row variable
and the column variable. Another application is displaying confusion matrices (e.g.,
[Pilhoefer et al., 2012]).
If there is a large number of combinations and only a few occur at all, then a
fluctuation diagram is valuable for revealing this information and for identifying cat-
egorical clusters. Figure 7.8 shows a fluctuation diagram for eight binary variables
of the Zoo dataset in the package seriation. Several distinct groupings stand out and
there are many feature combinations for which there are no animals in the dataset.
Fluctuation diagrams can be drawn in R using the fluctile function in the
package extracat. You could compare this fluctuation diagram with a mosaicplot of
the same data using the code

data(Zoo, package="seriation")
mosaic(table(select(Zoo, c(hair, eggs:backbone))))
Studying Multivariate Categorical Data 141

data(Zoo, package="seriation")
fluctile(table(select(Zoo, c(hair, eggs:backbone))),
label=FALSE)

FIGURE 7.8: A fluctuation diagram of eight binary variables (hair, eggs,

milk, airborne, aquatic, predator, toothed, backbone) for
the 101 cases of the Zoo dataset in the package seriation. A few combinations arise
relatively often, most not at all.
142 Graphical Data Analysis with R

Comparing rates across subgroups

(Doubledecker plots, Figure 7.9)
If only horizontal divisions are used for all but the last variable, you get a dou-
bledecker plot, which is a multivariate spineplot. Each bar has the same height and
therefore both its area and its width are proportional to its count. The plot’s name
comes from how it sometimes looks when you colour the bars by a final binary de-
pendent variable. This is shown in Figure 7.2 for the Titanic dataset. Doubledecker
plots are excellent for comparing rates across all groups (the heights in the shaded
sections of the bars), while conveying some information on the relative sizes of the
groups.
Another example, this time with an ordinal dependent variable, can be seen
in Figure 7.9 for the Arthritis dataset in the package vcd. Improvement can be
“marked”, “some”, or “none” and is dependent on Treatment and Sex. Note how
the ordering of the variables, the ordering of the Treatment categories, and the di-
rection of splitting have all been chosen to reflect a clear pattern in the data. Treated
patients did better than the placebo patients and females did better than males within
those groups. Hardly any males in the placebo group showed any improvement at all.
Doubledecker plots can be drawn in R using the doubledecker function in
vcd or the rmb function in extracat.

data(Arthritis)
Arthritis <- within(Arthritis,
Treatment1 <- factor(Treatment,
levels=levels(Treatment)[c(2,1)]))
mosaic(Improved ~ Treatment1 + Sex, data = Arthritis,
direction = c("v", "v", "h"), zero_size = 0)

Treatment1
Treated Placebo
Some None
Improved
Marked

Female Male Female Male

Sex

FIGURE 7.9: A doubledecker plot of Improvement for arthritis patients conditional

on Treatment and Sex. The treated group did better than the placebo group and
females did better than males.
Studying Multivariate Categorical Data 143

How do patterns compare by subgroup?

(Multiple barcharts, Figure 7.10)
Complex contingency tables with several rows and columns can be awkward to work
with. Tests based on χ 2 statistics may be highly significant, but how should the results
be presented? Fluctuation diagrams, treating the row and column variables equally
can be effective, although comparisons of rectangular areas are harder than compar-
isons of lengths.
If one variable can be regarded as dependent and the other explanatory, then
multiple barcharts are good. They show the distributions of the dependent variable
conditional on the explanatory one. Consider the mosaicplots for the two variables
lake and food for the Alligator dataset in the bottom left and top right corners of
Figure 7.4 and compare them with the multiple barcharts version in Figure 7.10. Note
that the barcharts have been aligned vertically to make comparing distributions easier.
Aligning the barcharts horizontally taking up less space would not be as effective.

Allig3 <- aggregate(count~food + lake, data=Alligator, sum)

ggplot(Allig3, aes(food, count)) +
geom_bar(stat = "identity") + facet_grid(lake ~.)

30

20
George

10

30

20
Hancock

10

0
count

30
Oklawaha

20

10

30

20
Trafford

10

0
bird fish invert other reptile
food

FIGURE 7.10: A multiple barcharts view showing how food depends on lake in
the Alligator dataset. The distributions for Lake George and Lake Hancock are quite
different from those for the other two lakes.
144 Graphical Data Analysis with R

If there is a final dependent variable that has several categories, then colour can
be used to fill the rectangles representing the explanatory variable combinations, and
what you get is a stacked barchart display. For certain data patterns this will work
well, cf. Figure 7.9. More commonly, the result is difficult to interpret, as there is no
cumulative pattern, and so one alternative is to use a grid structure for the explanatory
variables and to plot the final dependent variable as a barchart. The resulting display
is a trellis plot as in Figures 7.6 and 7.7. These plots can be drawn using the rmb
function in extracat or with ggplot2 or lattice.

Just showing the rates

(Same binsize plots, lower plot in Figure 7.11)
This form is different from the others in that each combination is represented by a
rectangle of the same size. Combinations with zero counts are drawn empty. This
plot is useful for two reasons, firstly for showing the pattern of combinations which
do not arise in the data, the missings, and secondly for comparing rates of a binary
dependent variable. Rates can be seen by colouring the rectangles up to a height
equivalent to the appropriate values. The disadvantage of the plot is the lack of count
information, but when the conditioning groups are of equal size or all large this is
irrelevant.
The upper plot of Figure 7.11 is a mosaicplot of the rates of high satisfaction for
the 24 combinations of the three variables Infl, Type, and Cont in the housing
dataset. The order of variables and their alignment (whether horizontal or vertical)
has been chosen to emphasise the pattern of generally increasing rates of high sat-
isfaction for increasing influence given contact and type. Because the groups have
different sizes, they are not aligned, which makes it more difficult to see the pattern.
The lower plot shows the identical data in a same binsize plot. Now the rate patterns
are easier to compare, while the differences in group sizes are ignored. Same binsize
plots with highlighting can be drawn in R using the rmb function in the package
extracat, using the freq.trans = ”const” option. The coding and options
for the two plots are rather different, as they are drawn with different packages, vcd
and extracat. The various ways R packages go about achieving their goals can be
frustrating as well as fruitful.
Studying Multivariate Categorical Data 145

data(housing, package="MASS")
mosaic(xtabs(Freq ~ Cont + Type + Infl + Sat, data = housing),
direction = c("h", "v", "v", "h"),
gp = gpar(fill = c("grey", "grey","red")),
spacing = spacing_highlighting)

Type
Tower Apartment Atrium Terrace

Medium Low
Low

High
Cont

Low
Sat
Medium
High

High
Low MediumHigh Low Medium High Low MediumHigh Low Medium
High
Infl

rmb(formula = ~Type+Cont+Infl+Sat, data = housing, cat.ord = 3,

spine = TRUE, freq.trans = "const")

Type
Tower Apartment Atrium Terrace
Infl
Low Medium High Low Medium High Low Medium High Low Medium High
1 1

0.8 0.8

0.6 0.6
Low

0.4 0.4

0.2 0.2

0 0
Cont

1 1

0.8 0.8

0.6 0.6
High

0.4 0.4

0.2 0.2

0 0

FIGURE 7.11: Rates of high satisfaction for the housing dataset in the MASS pack-
age. The upper plot is a mosaicplot, which shows the different group sizes, and the
lower one a same binsize plot, which only shows the rates.
146 Graphical Data Analysis with R

Comparing patterns for subgroups of very different frequencies

(Relative multiple barcharts, rmb, Figure 7.12)
Like trellis plots, all subplots in a multiple barchart have the same vertical scale. If
the underlying groups are of very different sizes, some of the individual plots will
have too small a height to see any pattern. The rmb plots in the package extracat
offer two solutions.
One solution is to adjust the horizontal scale of each subplot, so that the total
area of the bars in the plot is proportional to the sample size and the full range of
the vertical scale is used. This results in barcharts from small groups being very thin.
The distributions represented by the barcharts formerly had comparable horizontal
scales but not comparable vertical scales, with this approach it is vice versa.
The other alternative offered is to draw all the barcharts with individual vertical
scales and use the colour intensity of the plots to represent the group sizes. Plots
from small groups are drawn with lighter shadings than plots from large groups and
this can be very helpful when it is the distributions that you want to compare and the
absolute values are not so relevant.
The two options are shown for the housing dataset in Figure 7.12. There is evi-
dence that satisfaction increases with increasing influence, that within the influence
subgroups satisfaction is slightly higher with more contact, and that the tower group
have higher satisfaction levels than the other three housing types.
Studying Multivariate Categorical Data 147

data(housing, package="MASS")
rmb(formula = ~Type+Infl+Cont+Sat, data = housing,
col.vars = c(FALSE,TRUE,TRUE,FALSE),
label.opt = list(abbrev = 3, yaxis=FALSE))

Infl
Low Mdm Hgh
Cont
Low Hgh Low Hgh Low Hgh
Twr
Apr
Type
Atr
Trr

rmb(formula = ~Type+Infl+Cont+Sat, data = housing,

eqwidth=TRUE, col.vars = c(FALSE,TRUE,TRUE,FALSE),
label.opt = list(abbrev = 3, yaxis=FALSE))

Infl
Low Mdm Hgh
Cont
Low Hgh Low Hgh Low Hgh
Twr
Apr
Type
Atr
Trr

FIGURE 7.12: rmb plots of the housing dataset. In the upper plot each barchart has
the same vertical scale, but its own horizontal scale, and its total area reflects the size
of the group in that cell. In the lower plot each barchart has the same vertical and
horizontal scales and the intensity of colouring reflects the cell group size.
148 Graphical Data Analysis with R

Graphics supporting modelling (Residual plots)

Mosaicplots can be very helpful for displaying raw data and they can also be used
to support modelling. The rectangles may be drawn with sizes proportional to the
expected values under a particular model and then coloured or part-filled according
to the model residuals. The resulting display is a good way to see if the model fits
and, if it doesn’t, where the problems lie. Some recommend drawing the rectangles
with sizes proportional to the raw data values, so that the shape of the plot remains
unchanged, whichever model is fitted. However, this can mean that there are rect-
angles with zero size, making it difficult to plot residuals. Using expected values is
more consistent with other statistical plots. The vcd package includes examples of
applying mosaicplots in modelling and offers some alternative displays.

7.5 Which mosaicplot is the right one?

There are some obvious principles to follow in choosing a suitable mosaicplot form
and structure, but that does not mean that you can find an ‘optimal’ plot. It is best
to look at several different ones and consider using more than one to display all the
information in a dataset.
Choice of plot form Classic mosaicplots are good for ordinal dependent variables
as you can see cumulative patterns if they exist. For dependent variables with
more than two categories multiple barcharts are often a good choice. If there
are very many possible combinations, then a fluctuation diagram will be helpful
for identifying clusters in the data. Doubledecker plots are good for comparing
rates for a binary dependent variable across all possible groupings. Same binsize
plots are useful for identifying missing combinations and for comparing rates
when all groups are much the same size, while rmb plots are best for comparing
conditional distributions for groups of very different sizes.
Choice of ordering of variables A binary dependent variable should be the last
variable in the plot and is best included as colouring in a vertical split, just
like highlighting in interactive graphics, rather than as a separate variable. Vari-
able ordering establishes what conditional relationships are shown. Figure 7.2
shows survival rates for Class within Sex for the Titanic dataset. Another order-
ing would be better for comparing survival rates by Sex within Class:

doubledecker(Survived ~ Class + Sex, data = Titanic)

Sometimes there is a natural hierarchical ordering that should be used, otherwise

it can be sensible to use data-driven orderings, if they lead to recognisable pat-
terns. An example can be seen in Figure 7.11 for the housing dataset, where the
ordering Cont, Type, Infl was chosen because of the rising pattern that is
revealed.
Studying Multivariate Categorical Data 149

Choice of ordering of variable categories The categories of ordinal variables must

be kept in the correct sequence, either increasing or decreasing. If there is no
sensible default ordering for the categories of a nominal variable, then ordering
by frequency is often best.
Choice of display form Mosaicplots are complex and need a lot of space, too large
is better than too small. The aspect ratio chosen can be crucial in determining
what can be seen in a plot and various sizes should be tried. Restrained use of
colour can be valuable for emphasising particular features. Labelling is difficult,
and so captions and additional annotations can be more important for mosaicplots
than for other kinds of plot.

7.6 Additional options

Aspect ratio All graphics displays look different when the window aspect ratio is
varied and mosaicplots are especially sensitive. For fluctuation diagrams and
the same binsize variant, varying the aspect ratio to produce squarish rectan-
gles seems to work best. This could be said to be related to Cleveland’s principle
for line charts [Cleveland, 1993], that the slopes should be around 45°, as the
rectangle diagonals are then at 45°. For the other forms, taller, thinner rectangles
are better. This is in line with another of Cleveland’s recommendations, given we
want to compare lengths. Deciding what aspect ratio is most informative can be
a subjective judgement and so it is a sensible plan to experiment a little, resizing
the window with the mouse, to find the most effective view.
Gaps between rectangles The most efficient use of space would be to leave no gaps
between the rectangles (as treemaps [Johnson and Shneiderman, 1991] do). This
makes comparisons and alignment difficult, so it is usual to leave gaps between
the categories whenever a split is made, and it is possible to vary the gap size
according to the order of the variables. The higher in the hierarchy the variable
appears, the larger the gaps. The choice of gap sizes and how much they are
reduced as you go down the hierarchy influence the way the graphic looks and
how easy it is to recognise features. Unusually, this is not a situation in which it
is easy to experiment. More research is needed.

Rotation Rotating individual variables (i.e. splitting vertically instead of horizon-

tally and vice versa) alters the display layout and can aid interpretation. If there
are three binary variables and one variable with eight categories, it will usually be
better to split the three binary variables in the same direction. Sometimes it can
be helpful to rotate the whole plot at once. The same effect can be achieved by
rotating each variable individually. As far as possible layouts should be chosen
to emphasise the comparisons you want to make.
150 Graphical Data Analysis with R

Censored zooming There are two kinds of censored zooming to aid the understand-
ing of fluctuation diagrams and multiple barcharts. Censored zooming works best
as an interactive tool.
Ceiling-censored When there are many possible combinations, some will have
zero counts, some will have very low counts, and it becomes difficult to
distinguish between them. This is immaterial if interest is only in the large
groupings, but often categorical outliers can convey important information
or it can be useful to be able to examine differences between small groups.
With ceiling-censored zooming, a maximum rectangle size is set and any
combinations with larger counts are drawn to that size. This magnifies the
smaller combinations, while limiting the larger ones. For example, we can
convert the Titanic dataset to a data frame with count variable Freq and
then construct a new count variable with a specified ceiling in the following
way:
t1 <- as.data.frame(Titanic)
ceil <- 100
t1$Fc <- ifelse(t1$Freq > ceil, ceil, t1$Freq)

and then the plot can be drawn with the new count variable.
Floor-censored When interest is concentrated on larger groups, combinations
with small frequencies may distract rather than help. They can be sup-
pressed by only displaying cells with counts larger than a specified floor.
The preparatory code would be
floor <- 5
t1$Ff <- ifelse(t1$Freq < floor, 0, t1$Freq)

Colour Colour is good for displaying rates in subgroups. It is also valuable for dis-
playing residuals, using separate colours for positive and negative residuals, and
for emphasising particular subgroups. All plots can be improved with a judi-
cious use of colour while many plots can be made worse by an injudicious use
of colour. Colour should emphasise information or add to the attractiveness of a
graphic, there has to be a definite reason for using it.
Studying Multivariate Categorical Data 151

7.7 Modelling and testing for multivariate categorical data

1. Contingency tables
The standard for checking the association of two categorical variables is the
χ 2 −test. In some situations Fisher’s exact test is appropriate. There are also tests
for special situations like McNemar’s and statistics for certain kinds of sets of
tables like Mantel-Haenszel.

2. Associations between categorical variables

If there is a small number of variables with only a few categories and with not
too many sparsely occupied combinations, then loglinear models can be used
to assess independence structures amongst the variables. Loglinear models can
be difficult to interpret and there are usually several that might be considered
acceptable. If the necessary conditions hold, and loglinear models can be fitted,
then mosaicplots are helpful in understanding the results, both in displays of the
raw data and in displays of model residuals.
3. Binary dependent variables
For a binary dependent variable, logistic regression is a good approach. For an or-
dinal dependent variable with more than two categories, cumulative link models
can be used, as for instance implemented in the R package ordinal.

Main points
1. It is difficult to display multivariate categorical data (§7.3).
2. Start with low-dimensional plots to get to know the basic data structure in a
dataset (Figures 7.3 and 7.4).

3. Mosaicplots and their variants offer many good alternatives (§7.4).

4. The order of variables in a plot and the orderings of the categories of variables
strongly influence what information can be found (§7.5).
5. There are many different kinds of mosaicplots and many different options for
them (§7.6). Try out several plots and be prepared to use more than one.
152 Graphical Data Analysis with R

Exercises
1. Cancer
The vcdExtra package includes a data table for an old breast cancer study on the
survival or death of 474 patients, Cancer.
(a) Convert the data to a data frame and draw plots to compare the survival
rates firstly by degree of malignancy and secondly by diagnostic center.
(b) Which plot would you draw to compare survival rates by both degree of
malignancy and diagnostic center? Does the order of explanatory variables
matter?
2. Titanic
None of the chapter’s graphics for the Titanic dataset make use of the Age vari-
able. What mosaicplot would you draw to include this variable and what are your
conclusions?
3. Death penalty
A summary table of death penalty verdicts in Florida over a number of years is
reported in [Agresti, 2007]. There are three binary variables, whether the victim
was black or white, whether the defendant was black or white, and whether the
verdict was the death penalty. The dataset can be found in the packages asbio
(death.penalty) and catdata (deathpenalty). What conclusions would you draw
from the data? Choose an appropriate mosaicplot to support your argument.
4. Berkeley admissions
The dataset on graduate admissions to Berkeley was introduced in §1.3 with bar-
charts of the three variables, Dept (Department), Gender, and Admit. Draw
plots to show how Admit varies by Dept and Gender separately and then a
plot showing how it varies by the different combinations of Dept and Gender.
5. Airline arrivals
The dataset airlineArrival in the package fastR includes details of 11000 ar-
rivals at five different airports for two airlines. Compare a spineplot of the rate
of delay by airline with a mosaicplot of the rates of delay by airline for each
airport separately. Which mosaicplot is most effective here? (This is an example
of Simpson’s paradox. The data were originally reported in [Barnett, 1994].)
6. Knowledge of cancer
The vcdExtra package includes the dataset Dyke about how 1729 survey respon-
dents’ knowledge of cancer depended on whether they listened to the radio, read
newspapers, did ‘solid’ reading, or attended lectures.
(a) What plots would you draw to present the information in this dataset of five
binary variables?
(b) If you were restricted to one plot showing all the variables, which one
would you choose and why?
Studying Multivariate Categorical Data 153

7. Punishment
A dataset on attitudes to corporal punishment of children in Denmark, Punish-
ment, is provided in the vcd package. What plot or plots would you recommend
for analysing this dataset and how would you summarise the information from
the dataset?
8. Hair and eye colour
The HairEyeColor dataset in R has often been used to illustrate mosaicplots.
Data on hair and eye colour of 592 students were published by Snee [Snee, 1974].
(a) Compare a classical mosaicplot with a fluctuation diagram and with a mul-
tiple barcharts version for studying the association between hair colour and
eye colour. What information do you find and which plot displays it best?
(b) How important are the orderings of the categories for the two variables?
(c) The data were later extended by Friendly “for didactic purposes” (R Help),
where, as [Friendly, 1994] puts it, “the division by sex is contrived”. Which
plot would you choose for looking at hair and eye colour by sex and does
the variable sex make a difference?
9. Clothing and intelligence
A classic example of a larger contingency table is the Gilby dataset in the package
vcdExtra. In his 1911 paper Gilby [Gilby, 1911] reported on a study of 1725
children’s standard of clothing (5 categories with the worst two combined from
worst/insufficient to very well clad) and teachers’ rating of their intelligence (7
categories combined into 6 from mentally defective/slow dull to very able). The
χ 2 statistic for the table is extremely significant and we can guess why, but you
cannot easily see from the table. Compare a classic mosaicplot, a spineplot, a
fluctuation diagram, and a multiple barcharts view for these data. Which do you
think conveys the information best?

10. Dinosaurs in North Dakota and Montana

The HCD dataset in the MCPAN package contains counts of dinosaur families
in three stratigraphic levels of the Cretaceous period in the Hell Creek formation
in North Dakota and Montana. The table in R is based on a subset of the data
used in [Sheehan et al., 1991], reported in [Rogers and Hsu, 2001].
(a) Draw barcharts of the numbers found at each of the three levels and of
the counts for each family. (You may find it helpful to reformat the dataset
using the package reshape2 or in some other way.)
(b) Choose a suitable mosaicplot to display the data for families by level. The
aim of the original study [Sheehan et al., 1991] was to study whether di-
nosaur family diversity changed across levels.
11. Intermission
Piet Mondrian’s Broadway Boogie Woogie is in the Museum of Modern Art in
New York. Does the title complement the picture?
8
Getting an Overview

Diagrams prove nothing, but bring outstanding features readily to the eye.

R. A. Fisher ([Fisher, 1925])

Summary
Chapter 8 discusses getting an initial overall view of a dataset.

8.1 Introduction
When you first start work on a dataset it is important to learn what variables it in-
cludes and what the data are like. There will usually be some initial analysis goals,
but it is still necessary to look over the dataset to ensure that you know what you
are dealing with. There could be issues of data quality, perhaps missing values and
outliers (discussed in the next chapter), and there could just be some surprising basic
statistics.
There are several different functions in R for showing what is in a dataset. You
can show the whole dataset (inadvisable for even moderately sized datasets), display
only the first few lines (using the function head), or just list the variables, each
with their type and a few values (using the function str). You can summarise the
dataset using the function summary (base R) or the function describe (Hmisc)
or the function whatis (YaleToolkit). There are doubtless other statistical alterna-
tives. Plotting the dataset is the alternative, complementary approach for getting an
overview and the one we will be concentrating on in this chapter.
As an example, consider the dataset HI from the package Ecdat on the ef-
fect of health insurance on women’s working hours. The data were analysed in
[Olsen, 1998]. First information can be gained with str (the results are shown on
the next page) and some simple graphics (Figures 8.1 and 8.2).

155
156 Graphical Data Analysis with R

data(HI, package="Ecdat")
str(HI)

# ’data.frame’: 22272 obs. of 13 variables:

# $ whrswk : int 0 50 40 40 0 ...
# $ hhi : Factor w/ 2 levels "no","yes": 1 1 2 1 2 ...
# $ whi : Factor w/ 2 levels "no","yes": 1 2 1 2 1 ...
# $ hhi2 : Factor w/ 2 levels "no","yes": 1 1 2 2 2 ...
# $ education : Ord.factor w/ 6 levels
# "<9years"<"9-11years"<..: 4 4 3 4 2 ...
# $ race : Factor w/ 3 levels "white","black",..: 1 1 1 1 1
# ...
# $ hispanic : Factor w/ 2 levels "no","yes": 1 1 1 1 1 ...
# $ experience: num 13 24 43 17 44.5 ...
# $ kidslt6 : int 2 0 0 0 0 ...
# $ kids618 : int 1 1 0 1 0 ...
# $ husby : num 12 1.2 ...
# $ region : Factor w/ 4 levels "other","northcentral",..: 2
# 2 2 2 2 ...
# $ wght : int 214986 210119 219955 210317 219955 ...

There are seven factors (one of which, education, is ordered), two numeric vari-
ables and four integer ones. The two kids variables must be discrete and limited, as
drawing up tables or plotting them would confirm.
Figures 8.1 and 8.2 show histograms of the four continuous variables and bar-
charts of the remaining variables respectively. The histograms’ display is drawn by
forming a new dataset, stacking the continuous variables, so that all the plots can be
drawn together with one line of code. (This could be done for the barcharts too, but
then any category ordering information for nominal variables would be lost.)
The hours worked by the women have two distinct modes, which on closer in-
spection of the data turn out to be 0 hours (i.e., not working) and 40 hours. The
odd shape of the histogram of experience is due to the default binwidth, but the
overall shape is clear. The extended axis below zero is surprising and growing the
window vertically would show that there are a very few cases with a value of −1.
(These arise because the variable is defined as years of potential work experience
= age − years of education − 5.) Both the last two variables, husband’s income and
sampling weight, are skewed to the right. Sampling weight distributions are often
skew with a few cases having exceptionally large weights. The mode in the salary his-
togram at 100 turns out to be real, with 478 husbands’ incomes reported as $99, 999
a year! This can be checked with something like

with(HI[HI$husby > 98 & HI$husby < 102 ,], table(husby))

Getting an Overview 157

library(reshape2)
HIvs <- c("whrswk", "experience", "husby", "wght")
HIs <- melt(HI[, HIvs], value.name = "HIx",
variable.name = "HIvars")
ggplot(HIs, aes(HIx)) + geom_histogram() +
facet_wrap(~ HIvars, scales = "free") +
xlab("") + ylab("")

whrswk experience
8000 1500

6000
1000

4000

500
2000

0 0
0 25 50 75 0 20 40
husby wght

4000

4000
3000

2000
2000

1000

0 0
0 50 100 150 0 400000 800000 1200000

FIGURE 8.1: Histograms of the four continuous variables in the HI dataset from
Ecdat. Women mostly work 40 hours per week or do not work at all. Potential years
of experience range from 0 to around 50. A large group of husbands have little or no
reported income. The distribution of case weights is skewed to the right.
158 Graphical Data Analysis with R

The barcharts in Figure 8.2 show the insurance variables, information on educa-
tion and race, the distributions of young and older children, and the distribution by
region. The variables are selected by finding out which ones have a limited number
of unique values. The weighting variable has been converted to a percentage to make
the scales readable.

uniqv <- function(x) length(unique(x)) < 20

vcs <- names(HI)[sapply(HI, uniqv)]
par(mfrow = n2mfrow(length(vcs)))
relativeWeight <- with(HI, wght/sum(as.numeric(wght))*100)
for(v in vcs)
barplot(tapply(relativeWeight, HI[[v]], sum), main = v)

hhi whi hhi2

50

60
60

40
40
30

20
20
10
0

no yes no yes no yes

education race hispanic

20 40 60 80

20 40 60 80
30
20
10
0

<9years 12years 16years white black other no yes

kidslt6 kids618 region

60

30
60

40

20
40

20

10
20
0

0 1 2 3 4 5 0 1 2 3 4 5 6 7 8 other northcentral south west

FIGURE 8.2: Barcharts of the nine other variables in the HI dataset in Ecdat. The
top row shows that about half the women are covered by their husband’s insurance
and a minority by their own, and that an equivalent minority of husbands have no job
health insurance. The middle row shows that most women have twelve or more years
of education and that there are few blacks and Hispanics in the study. The final plot
at the bottom right shows that more participants were from the South.

Sometimes it is obvious which variables should be treated as continuous and

which as categorical or discrete, sometimes it is ambiguous. If an age variable takes
values from 0 to 100, then it makes sense to treat it as continuous, while if there are
only integer values between 20 and 30 it might be better to treat it as discrete. If you
Getting an Overview 159

guess wrongly, then draw another plot. The whole point of initial overviews is to get
a feel for the data, not to draw perfect pictures. They can come later when you have
learned what information in the data is worth presenting.

8.2 Many individual displays

Instead of jumping in and producing lists of summaries (as with summary) or a
large matrix of primarily scatterplots (as with plot), another approach is to begin
with the basics and work up step by step. Knowing what variables of what kinds
there are and how many cases is a pretty good start and can be achieved using str,
as was done in §8.1. The GDA approach suggested here is to split the variables into
two groups, plotting categorical and discrete variables as barcharts, while plotting the
other variables, where possible, as histograms. Any special variable types left over
(e.g., dates) should be dealt with separately.
Plots of individual variables give a quick view of the variable distributions and of
any features that stand out. Plotting variables in groups, all histograms together and
all barcharts together, is quicker than plotting them one by one and organises them
neatly. The Boston housing dataset was already examined in Figure 3.9, treating all
variables as continuous. In fact the dataset’s 14 variables include one binary, one
discrete, and twelve numeric. In Figure 8.3 the binary and discrete variables have
been plotted and in Figure 8.4 all the continuous variables.

data(Boston, package="MASS")
par(mfrow=c(1,2))
for (i in c("chas", "rad")) {
barplot(table(Boston[, i]),
main=(paste("Barchart of", i)))
}

Barchart of chas Barchart of rad

120
400

100
300

80
60
200

40
100

20
0

0 1 1 2 3 4 5 6 7 8 24

FIGURE 8.3: Barcharts of the variables chas (whether the tract bounds the Charles
River or not) and rad (index of accessibility to radial highways) from the Boston
dataset.
160 Graphical Data Analysis with R

Although neither of these is ideal for the individual variables (for instance the
histogram of medv misses the collection of areas with medv= 50, which was iden-
tified in §3.3), they do offer some direct insights (variable definitions can be found
on the dataset’s R help page):

1. There are few areas with chas = 1.

2. Three values, 4, 5, and 24, dominate the variable rad. The value 24 is quite sep-
arate from the others in value (the plot fails to show that clearly because barcharts
treat a discrete variable’s numbers as category names).
3. The variables tax and indus have gaps that could be investigated.
4. Three variables are highly skewed (crim, zn, and black).
5. And there are suggestive features in other variables to be studied more closely.
Getting an Overview 161

vs1 <- !(names(Boston) %in% c("chas","rad"))

grs <- n2mfrow(sum(as.numeric(vs1)))
par(mfrow=grs)
for (i in names(Boston)[vs1]) {
hist(Boston[,i], col="grey70", xlab="", ylab="",
main=(paste("Histogram of", i)))
}

Histogram of crim Histogram of zn Histogram of indus

100 150
400

300
200

50
100
0

0
0 20 40 60 80 0 20 40 60 80 100 0 5 10 15 20 25

Histogram of nox Histogram of rm Histogram of age

120

100 150
80

50 100
40

50
0

0.4 0.5 0.6 0.7 0.8 0.9 4 5 6 7 8 9 0 20 40 60 80 100

Histogram of dis Histogram of tax Histogram of ptratio

120

120
80

80

50 100
40

40
0

2 4 6 8 10 12 200 400 600 12 14 16 18 20 22

Histogram of black Histogram of lstat Histogram of medv

150

100 150
300

100
50

50
0 100

0 100 200 300 400 0 10 20 30 40 10 20 30 40 50

FIGURE 8.4: Histograms of the twelve continuous variables from the Boston dataset.
162 Graphical Data Analysis with R

8.3 Multivariate overviews

Scatterplot matrices have already been mentioned as a way of studying relationships
between variables and they can be very effective. There are other possibilities as
well. Apart from parallel coordinate plots and trellis plots there are tablelens plots,
heatmaps, and glyphs. The following subsections discuss examples of several of
these alternatives for the Boston dataset (variable definitions can be found on the
dataset’s R help page).

Scatterplot matrices
The default scatterplot matrix (well, default except for using points rather than open
circles) shown in Figure 8.5 is surprisingly informative, even with 14 variables. A
number of features stand out:
1. The curious either/or relationship between crim and zn.

2. The dependencies of medv on lstat and rm.

3. The fact that tax, ptratio, zn, and indus have only single values for the
extreme value 24 of rad.
4. The associations between age and dis and nox.
5. And some potential bivariate outliers such as the point in lstat and age or
the set of high values of nox, which have the same value for rad, tax, and
ptratio.

As a next step you could consider adding univariate displays of the variables down
the diagonal or colouring cases by membership of some subgroup. Figure 8.5 is just
intended to provide a first quick look to help you to decide how to proceed. It may
be a complex graphic, but it has a straightforward structure and is easy to draw. We
should take advantage of the power that software can offer us nowadays.
Getting an Overview 163

plot(Boston, pch=16)

0 60 0.0 0.8 4 6 8 2 8 200 600 0 300 10 40

! ! ! ! ! ! ! ! ! ! ! ! !
!
! !
! !
! !
! !! !
! !
! !
! !
! !
! ! ! !! !!

60
crim !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!!
!
!
!
!
!!
!
!!
!
!!
!!!! !
!
!
!! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
! !
! !!
! ! ! !!
!
! !!
!!!
! !!
! !
!
!
!!
!!
!
!
! ! ! ! ! ! !
! ! ! ! !!!
!
!
!
!
!
! !
! !
!!
!
!!!
!
!
!
!!!! !!
!!!
!!!
!!
!!
!
!
!!
! !
!
! ! !!!
!!!
!
!
!
!
!
!
!
! !
!
!
!
!
!
!!
!! ! !
! !
! !!
!!
!
!!!!!! !
! ! !
!!
!!
!
! !
!
!!
!
!
!!!
!!
!
!
!!
!!!!!
!!! !
!
!
!
!
!
!
!
!
!
!!!
!
!!!
!!!
!
! !
!
!!!!
!
!! !!!!!
!!!! !!
!
!!!
!! !
!
!
!!
!
!
!!!
!
!
!!
!
!!
!!
!!
!
!
!!!!!!
!
!
!
!
!
! !
!
!
!!
!
!!
!
!!
!
!!
!
!
!
!
!!
! !
!
!
!
!!
!!!
!
!
!!
!!!
!!! ! ! !!
!
!
!
!!
!
!!
!
!
!
!
!
!
!
!!
!
!
!
!!
!!
!!
!!
!
!
!
!
!!
!
!!
!
!
!
!!!
!
!!
!!
!
!
!
!
!
!
!
!!
!
!!
!!!
! !!
!!
!!
!
!!
!!!
!!
!
!!!!
!
!
!!
!
!!
!!
!
! !
!
!!
!! !
!!!
!!
!
!!
!!!
!!!
!
!!
!!
!!
!!
!
!!!
!!
!!
!!
!!!
!
!
!!!
!!
!
!!!!
!
!!!
!
!!
!!
!
!!
!!!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!!
!!!!
!!
!
!
!
!
!
!
!
!!
!
!!
!!
!!
!!
!!
!
!! !
!
!!
!!
!!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!!
!
!
!!
!
!!
!!
!
!
!
!!!
!
!
!!!
!!
!
!!!
!
!!! ! !
!!
!!
!
!
!!!!
!
!
!
! !
!
!!
!
!!
!
!
!
!!!
!
!
!!!
!
!!
!
!
!!!
!
!
!! ! !
!!!
!
!!
!
!!
!!
!
!
!!
!!
!
!
!!
!
!!
!
!
!
!!!
!!
!
!!
!
! !
!!
! !!!
!!
!!
!
!!
!
!
!
!
!
!!
!
!!!
!
!
!!
!
!
!
!!
!!!
!
!
!
!
!
!
!
!
!
!
! !
!
!
!
!
!!
!
!
!
!
!
!
!!
!!
!!!
!!
!!
!
!!
!
!
!
!
!
!
!!
!!
!
!
!!
!
!
!!
!!
!!
!!
!!
!!!
!!
!
!
!!
!!
!!
!!
! !!
!
!!
!!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!!
!!
!
!
!
!!
!!
!
!
!
!
!
!
!
!
!
!
!!!
!
!
!!
!
!
!
!
!!
!
!!
!
!!
!
!
!
!
!
!
!
!
!!
!
!
!!!
!!
!
!!
!
!!
! !
!!
!
!!
! !
!
!!
!!
!!
!!
!!
! !
!!
!
!!
!!
!!
!
!
! !
!!
!!
!
!!
! !
!!
!
!
!!
!!

0
!
! !
!! !
! !
!
! !
!! ! ! ! ! !! ! ! !
!
! !
!
!!
!! !
! ! !
!
!! !
!!!
!!!
!
! !!!!!
! !!!
!!! !
!! ! !!
!!
! !!
!!!!!!! !!
!!! !!!
!
!
!
! !
!
!!
!!
! !!
!!
! !!!
!
!
! !
!
!!
!
!! !
! !
!
!
!!
!
! !!
!
!!
!!
!!
!!!!
! !!!
!!
!!!
!!
!
!
!! ! !
!!!!!! !!!
! !
!!!!! ! !! !!! !
!! ! !
!!!
! !!
!
!!
!!
!
!!!
! !!!
!
!!!
!
! !!! !
!!
!!! ! ! ! ! ! ! ! ! !! !
! ! !!!
zn
! ! ! ! !!! ! ! ! ! ! !! ! !!
60

!
! !
!!
! !
! !
!!
! !!!
!!
!
! !! !!! ! !!
!! ! !
!!! !! !! !! !! !!
!
! !
!!
!!
! !!!
!!!
!
!
! !!!! !
! !
!!
!!
!
!!
!!
!!
!
!
!!
!
!
!!
!
!!
!!
!
!
!!!! !
!
!!
!!! !!
!! ! !! ! !
!!
! !!
!!! ! !!
!!
!!
! !
!!!! !
! !
!
! !!!!
!
!
!
!!
!!! ! !
!
!! ! !!!!
!!! !! ! !!! !!! !
!!! ! !! !!! !!!!!!!! ! !
!
!
! !
!!
!!
!
!!
!! !!!!!
! !
!!
!!!
!!
!
!
! !!
! ! ! ! ! !
!!
!
!!!! ! ! !
!
!!
!
!!
!!
! !! !
! !! !
!!
! !!
!
! !!!
! !!! !!!
!!
!!
!! !
! !
! !!
!!!! !! !
!
!!
!
!
!
!
!
!
!
!!
!!
!
!
!!
!
!
!
!
!!
!!!
! !!
!!
!
!!
!
!!
!!!
!
!
!!
!! !!
!
! !! !! !
! !
! !!
!
!!!
!!
!!
!
!
!!! !! !!!!!!
!
!! !!
!!
!
! !
!
!!
!
!!
!!
!
!
!
! !! !
!!!
!! !
!
! !!!
!
! !!!! !! ! ! !
!
!
! !
!!
!!
!
!!!
!!
!!!!!!
! ! !
!!!!
!
!
!!!!
!!
!
!! !!
!!
!
!
!!
!
!!
!
!!
!!
!!
!
!!
!!
!!
!!!
!! !! ! !!
!
!!
!!
! !
!!!
!! !!!!! ! ! !!
!!
!!
!
!!
!!
!!!
!
!!!
!
!!!! ! !!!
!!
!!
!!
!
!
!!
!!!
!!!
!!
!
!!!
!!
!!
!
!!
!
!!
! !!!
!!
!!
!
!!
!
!
!!!
!
!
!
!!!
!!
!
!!
!
!!
!!!!
!
!!
!!
!
! !
!
!!
!!!!
!!!!
!!
!
!!
!!!!
!!
!!
!!
!!
!!
!!
!
!!
!!
!!!
!
!
!!
!!!
!!
!!
!
!!
!
!!
!!
! !
!!
!!
!
!!
!!
!
!!
!!
!
!!
!
!
! !
!
!
!!
!!
!
!
!!
!
!!
!!
!
!
!!
!
!!
!
!!
!!
!
! !!
!!
!!! ! !!!
!!!
!!!!
!!
! !! !!!!!!!
!
!!
!!!!!
! !
!!
!
!!
!
!!
!!
!
!!
!
!!!
!!!
!
!!
!!
!
!!
! !!
!
!!
!!!
!!
!!!
!
!
!!
!!
! !
!
!
!
!!
!
!!
!!
!!
!!
!!!!
!!
!!
!!
!
!!!
!
!
!!
!!!
!!
!
!!
!!
!!
!
!
!
!!!!
!!
!!!
!
!
!!
!
!!
!
!
!!
!
!!
!
!
!!!!
! !
!
!!
!!
!!!
!
!!
!!
!
!!!!!
!!
!!
!!
!!!
!
!!
!!
!!!
!
!!
!!
!
!!
!!
!!
!!
!!
!!
!!
!
!!
!!
!!
!
! !!
!!
!!!!
!!
0

!
! !
! !
! !! !!!
!!
! !!!!!
!
!
! !
!!
!
! !! ! ! !! !!!
!
!!
!
! ! !!
!! !!! !!!!
!
!!
!!!
!
!
! !
! !
! ! !
! ! !
!!
!
!!!
! !
!
!! !
!!
!
! ! !! ! !!! !!
! !!
!!
!!
!! !! ! !!!
!
!!
!!!
!
!!! ! !
!!
!
!!
!!!
!
! !!!!
!! !
!!
!!
!!
!
!! !
!
!!
!
! ! ! !!! !!
!!!!
!!
!
!!
! !
!!
!
!!!
!!
!!
!!
!! ! !! !
!!
!!
!!
!
! !!
!! !!
indus
!
!
!!
!!
!!
!!
!!
!
!!
!!
!!
!!!
!! !! ! ! ! ! !! !
!!
!!
!
! !!!
!!
!!
!!
!!!
!!
!
!!
!!
!
!!
!
!
!!
!
!!
!!
!
!!
!!
!
!!! !!!!
!!!
!
!!
!!
!!!
!!
!
!!
!
!!
!
!!
!
!!
!
!
! !
!
!
!!
!!
!
!
!!
!
!
!!
! ! ! ! !
!!
!
!!
!
!!
!!!
!
!! !!!
!!
!!
!!!
!!
!!
!
!!
!
!!
! !
!!!
!
!!
!!
!!
!!
!
!!
!
!!
!
!!
!!
!!
!
!!
!
!!
!!!
!
!!
!
!
!!!!
! !
!
!
!!
!
!!
!
!
!!
!
!
!
!!!
!
!!
!!!!
!
!!
!!
!
!!
!! !

15
!
!
! ! !
! !
! ! !! !
! !
!!!
!!
!!
!!
!
!
!! ! !!!!! !
!!
!!!!
! !!! !
!
!!
!!
!!
!!
! !
!!
! !
! !
! !! !! ! !!
!
! !!
!!
!
!
!!
!!!!
! !
!
!
!!
!
!
!!
!!!
!
! ! ! ! ! !!!! !!!
!!
!!
!
!
!!
!!
!
!! !
!!
!!!!!! ! !!!!
!!!!
!
!!
! !! !
!
!
!!!!! !! !
!! !!! !!
!!
!! !
!!!
! !
!
!!
!
!!
! !!!!! !!!!!
!
!!
!!
!
!
!
!
! !
!!! !!!
!
!
! ! !!! !! !
!
!
!!
!
!
!
!
!!
!
!
!
!!
!
!!
!
!
!
!
!!!
!
!!
!!
!!!
!! !
!!!!!
!! !
!!!
!!!
!
!!
!
!!!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!!
!
!!
!
!!
!!
!
!!
!
!!
!
!!
!
!! !
!!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!!
!!
!
!
!! !!
!
!
!
!! !
!
!
!!! !!!!
!!!!!
! !!!!!!
!
!
!
!
!
!
!
!
!
! !
!
!
!!
!
!
!
!
!
!!
!
!!
!!
!
!
!
!!
!!
!
!
!
!!
!
!
!
!
!!
!
!
!!
!
!!!
!
!!
!
!
!!
!!
!
!
!!
!!
! !!! !
!!
!!
!
!
!!
!
!
!
!
!!
!!
!!
!!!
!
!
!
!
!
!
!
!
!
!
!!!!!
!
!
!
! !!
! !!! !!!
!! !!!!
!
!
! !
!
!!
!
!!!
!!! !
! !
!
!
!
!!
!
!
!
!!
!
!!
!
!!
!
!
!!
!!!
! !!
!! !
!!!
!!
!
!!
!!
!
!!
!!
!!
!!
!!!
! !!
!!!
!!
!!!
!!
!
!
!!
!!!
!!!!
!!
!!! !!
!
!!
!!
!
!!
!!
!
!!
!
!
!!
!
!!
!
!
!!!
!
!!
!
!
!!
!
!!
!
!!
! !
!
!!
!
!!
!
!
!
!
!
!!!!! !! !!
!
!! !
!!
!!!! !
!!
!
!
!
! !!
!
!!!
!!
!
!!
!
!
!!
!!
!!!
!!
!
!!
!
!
!!!!!
! !
!!
!
!!
!
!!
!
!
!!
!
!
!
!
!
!
!!
!
!
!
!
!!
! !
!!
!
!!!
!
!!
!!!!
!!
!!!
!!
!
!!
!
! !!!
! !!!!
! !!
!!!!
!!
!
!
!
!
! !
!
!
!
!
!
!
!
!!
!
!!!! !
!!
!!
!!!
!
!
!
!
!
!
!
!!
!
!
!
!!
!
!!
!!
!
!!
!!!
!! !
!
!!
!
!
!
!!!
!!
!
!!
!!
!!
!
!
!
!
!
!
!
!!
!!
!
!!!! !
!
!!
!!
!
!!
!!
! ! !!
!
!! !!
!
!
!!
!
!!!
!
!!
!
!!
!!
!
!!
!!! !
! !
!
!
!!
!
!!
!!
!! !
!
!
!!
!
!!
!
!!
!!
!!
!
!
!
! !
!! !
!!!!
!
!
!
!!
!!
! !
!
!!
! ! !
!!
! !!
!
!
!
!
!!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!!
!
!!
!!
!
!
!
!!
!!!! !!!
!!!
!
!
!!
!!
!
!
!
!!
!
!!!
!
!
!
!!
!
!
!!
!
!!
!!
!
! !!
!!!!

0
!
!
!
!! !!! ! !!!
!!! !! !!
!!
!!!!!!! ! !!!
!!
!
!!!
!
!!
!!
!
!!
!!!!
!! !
!! !
!
!!
!
!!!!!
!
!!
!
! !
!
!!
!
!!
!
!!
!
!!
!!! !!
!!! ! !
!!!! ! ! !!
! !!
!! ! ! !
!!
!!
!
! !
!
!!
!!
!
!!
!
!!
!!
!
!!!
!! !
!!!
!
!
!!
!
!!
!!
!!!!
!
! !!
0.8

chas
0.0

!
!
!
!
!!
!
!
!!
!
!!
!
!
!!!
!
!!
!!!
!! !! ! !!!
!
!!
!
!!
!!!!
!!!!!
!!
!
!!
!!! !
!
!!
!
!!
!
!!
!
!
!!
!
!!
!!
!!
!!!!!!! !
!
!
!!
!
!!
!
!!
!
!!
!
!!
!!
!
!!
!!
!
!!!! ! !!!
!!
!!
!!
!
!
!!!
!
!
!!
!!!
!
!
!!
!!!
!
!!
!!
!
!!
!
!
!!
!
!!
!
!!
!
!!
!
!!
!
!!
!
!
!
!!!
!
!
!!
!
!
!!
!!
!
!! !
!
!!
!!
!
!!
!!
!!
!
!!
!
!!
!!
!
!!
!!
!!
!
!!!
!!
!
!!
!
!
!!!
!!!
!
!
!
!!
!!
!
!
!!
!
!!
!
!
!
!!
!!
!!
!
!!
!
!!
!!
!!
!
!!
!
!!!
!
!!
!!
!
!!
!!
!!
!!
!!
!
!
!!
!
!
! !
!
!
!!
!!
!
!
!!
!
!
!!
!
!
!!
!
!
!!
!
!
!!
!
!
!
!!
!!
!
!
!!
!
!
!!
!
!
!!! ! !
!!!
!!
!!
!!
! ! !!
!
!!
!!
!
!
!!
!!
!
!!
!
!!
!
!!
!
!! !! !
!!!
!
!
!!
!!
!
!!
!
!!
!
!!
!
!
!!
!!!!
!! !
!!
!
!!
!
!!
!!
!
!!
!
!!!
!!!
!
!!
!
!!
!!!
!
!!
!!
!
!!
!
!!
!
!
!!
!
!
! !
!
!
!
!!
!
!!
!!
!
!!
!
!!
!!
!!!
!
!!
!
!!
!!
!!
!
!!!
!!
!
!
!!
!
!!
!
!
!!
!
!!
!
!
!
!!
!!
!
!
!!
!!
!
!
!!
!
!!
!
!
!!!
!
!
!!!
!!
! !
!
!
!!
!!!
!
!!
!
!
!!
!
!
!
!!
!
!!
!!
!
!!
!
!!
!
!!
!
!!
!
!
!!
!
!!
!
!!
!
!!
!!
!!
!!
!
!
!!
!
!!
!
!!
!
!!
!
!
!!
!
!!!
!!
!!
!
!!
!!
!!
!
!!

!
! ! ! ! ! !
!!
!
!!!
!!! !!
!!
!
!
!
! !
!
! ! ! ! ! ! !!!
!
!!
! !!!
!!
!
! !!
!! !
!!
!!
!!
!
!!
!
!! ! ! ! ! !!
!!!
! !!
!!! !
!
!
!! ! ! ! !!!
! !!!
!
!!!! !!
!!
!
!!

0.4 0.7
!!!
!!
! ! ! ! !!
!!
!
!
!! !!
!!
! !!
! ! ! ! !!
!!!! !! !! !
!!
!!!
!!!!! !
!
!!
!!
!
!!!
nox
!
!
!!
!
!
!
!
!!
!!
!
!!
!!!
! !! ! !
! !
! !
! ! !!!
!!
!
!!
!!
!
!
!!!!
!!
!
!
!!
!!
!!
!
!
! ! ! !
!
!!
!
!!
!
!!
!
!!
!
!!
!!
!!
!
! !
!
!
!
!!
!
!
!! !
! !
! !
! !
!!!
!!
!! !! !
!! !
!!!
!!!
!
!!
!! !!!!!
!
!!
!
!
!!
!!
! !
!
!
!
!!
!
!!!
!!!
! !
!
!!
!
!
!
!!
!!
!!
!
!
!
!!!!!
!!
!!!!!
!
!
!!
!
!
!
!!
!!
!! !! ! !
! !
! !
! !!!!
!
! !
!!!!
!
!!
!
!
!!
!!
!
! !! !! ! !!!
!!
!!
!!
!
!
!
! !
!
!
!
!!
!!
! ! !
! ! ! !
!! ! !
!! ! ! ! !!
!!
!
!
!
!
!! !
!!
!!
!
!!!
!!
! !
!
! !!
! !! ! !!
!
!!!
!!
! !
!!
!! !!!
!
!
!!
!!
!!!! !
! ! !!!!! ! !!
!!
!!
!
!!!
!
! ! !
!!! !
!!
! ! ! ! ! !!! !!
! !!
!!
!!
!!
! !!
! ! !
!!
!!
!!
!
!!
! !
! !! !
!
!!
!
!
!
!
!
!
!
!!
! !!
!
! !!
!
!!! ! !
!
!
! !!
!
!
!
!!
!!
!
!!
!
!!
!!
!
!!
!
!
!!
!!
!
!
!
!
!
!
!!!!
! !! !!!
!!!
!
!!!
!!
!
!!
!!
!
!!
!
!!
! !
!
!
!!
!!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!!
!
!
!
!
!!
!
! !!
! !
!
!
! ! !!!
! !!
! ! ! ! ! !!!! !! !!
!! ! !! !!
!!
!
!!
!
!
!!!
!
!
!
!
!
!
!!
!
!
!
!
!!
!
!!
!
!
!!
!
!!
!
!
!!
!
!!
!!
!!!!
!
!
!!
!
!!
!
!!!!
!!!!
!
! !!!!
!!
!
!!
!!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!! !!!
!
!
!
! !
!!! ! !!!
!
!!!
!
!!!
! !
!
! !
!
!!
!
!
!
!!
!!
!!
!!
!
!!
!!
!!
!
!!
!
!
!
!
!
!
!
!
!!
!!!
!!
!
!
!
!
!!!!
!
!
!!!!
!!! !! ! !
!!
!!!!!
!!
!!!
!
!!
!
!!!
!!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!!
!
!
!
!!
!!
!
!
!!
!
!
!
!
!
!
!!
!!
!
!
!
!
!
! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!!
!
!!
!
!
!
!
!
!
!!!
!! ! !!
!!!!
!
!
!!!! ! !!!!!
!!!!!!
!! ! !!! !
!!
!
!
!
!
!!
!
!!
!
! !!
!
!!
!
!!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!!
!
!
!!
!
!
!!
!
!!
!
!
!!
!! !!
!
!!
!
!
!!
!
!
!!
!
!
!
!
!
!
!
!
!!
!
!!
!!
!!
!!!
!!
!!
!!
!!
!!
!!
!!
!
!
!
!!!!!!
!!
!!
!
! !!
! !!
!!!! ! !! !!!
!
!!
!
!
! !
! !
! !!
!
!!!
!
!
!!
! !
!!
!
!
!!
!!
!!!! ! !!
!!
!
!
!!!
!
!!
!!!!!
!
!
!!
!!!
!
!
!!!
!!! !
!!! !
!!
!
!!
!!
!
!!
!!!
!! !
!!
!
! ! !
!
!
!!
!
! ! !! !!!
!
!!
!
!
! ! !
!!
!
!! !
!
!!
!!
!
!
!!
!!
!!!
!!!
!
!!!
!!
!!!!!! ! !!
!
!
!!
!!!
!
!!
!
!!
!!
!
! !
!!
!!!!!
!
! ! !
!!!
!!! ! !
!!!
! !!!
!
!!
!
!
!
!
!
!
!
!
!
!!
!
!
!!
! !!! !
! ! !
!
!!
!!
!
!
!!
!
!
!!
!
!!
!!!
!
!
!
!
!
!!
!
!
!
!
!
!
!!
!!
!!! !
!
!
!!
!!
!
!
!
!!
!!
!
!!
!!
!!
! !
!
!!
!!
!
!!
!!
!!
!!
! !
!!
!
!
!!
!!!
!! !
!! !
!!
!!
!
!!
!
!!
!
!
!!
! !
!
!
!!
!
!
!!!
!! !
!
!!
!
!!
!
!!
!
!
!
!
!! !
!!
!
!
!
!!
!
!
!
!
!!
!
!
!!!
!
!
!
! !!
! !!!
!!
!
!
!
!
!
!!
!
!
!
!
!!
!!
!
!
!
!!! !!
!
! !
!
!
!
!
!
!
!
!
!!
!
!
!
!
!!
!
!
!
!!
!!
!!
!!
!
!
!
!
!
!
!!
!!
!!
!
!
!
!
!! !
!
!!
!
!!
!
!
!
!!
!
!!
!
! !
!!!
!
!!
!
!
!
!
!!
!
!
!!
!
! !
!!
!!

!
! !! !!! !! ! ! ! !!! ! ! ! ! ! !! ! ! ! !! !
! !
! ! !
! ! ! ! ! ! !! ! !! !!!!!
!! !!
!! ! !!
!! !
!!
!! ! ! !
!!! ! !
!
! !
! !
! !!
!!!!
!
!
! !
!!! ! ! ! !! ! !!! !!!
! ! !! ! !!!! !! !
! ! !!
! !
!! !! !
4 6 8

!
!! ! ! ! ! !
!
!!
! !
! ! !
! !
! !
!
!!!!
! !!! !! !!
! !
!
!!! !!
!!!!
!!
!
! !
!
!!
!!! !! !!!! ! !
!!!! !
!! !
!
!!!
! ! !
!!!!!
! ! !!! ! !! !!
! ! !! ! !!!! !!!!
!! ! ! !! !! !!
!
!!
!! !!!
!
! ! !!!
!! ! ! !! !!
! !
!!
! !
!!!
! !!! !!! !!
!! !
! !
rm
!
!! ! !
! !! !!!
!!
! !! !
!
!!
!
!
!!!!
! ! !
! !
! !
!
!!
!!
!!
!! !
!! !
!
!!!
! !!
!!
!!!
!!!!
!!
!
!! !!
!!
!!
!!
!!
!! !
!!
! !! ! !
!
!!
!!!
! ! !!!
!!
!
!!
!
!! ! !
!
!
! !
!! !
!
!
!!
!!!!
! !! !!
!!
! !!
!
!
!
!!
!!
!
!! ! !
!!!
!
!
!!
!
!
!!
!!
! !
!
!
!!
!!
!
! ! !
! !
!
!
!!
!!
!!
!!! ! !!
!!
!! !!! !
!
!!
!
!
!!
!
!
!!
!
!!!!
!! !! !
! ! !
!
!!
!
!
!
!!
!
!!
!
!!!
!
!!
!
!!
!!!
!
! !!! ! !
!!! !
!
!!!
!!
!!
!
!!!
!
! !
!!!
!
!!!
!
!
!!
!!
!
!
!!
!
!
!
!!
! !
!
!!
!
!!
!
!
!!
!
!
!!
!
!!
!
!!
!
!! !
!!
!
! ! !
!
!!
!
!
!!
!
!
!!
!
! ! !!
!
!!
!
!!
!
!
!
!!
!
! !
!
!
!
!!
! ! !
!!
!!
!
!
!!
!!
!!
!!
!
!
!
!!
!
!!
! ! !!
!!
!!!! ! !
!
!!
!!
!
! !
!
!!
!!
!
!
!
!
!!!
!
!
!
!!
! !
!
!!
!!
!
!!!! !!
!
!!
!
!!
! !
!!
!
!
!
!!
!
!
!
!
!!
!
!
!!
!
!! !
!
!
!
!!
!
!
!
!
!!
!!! !
! !!
! !
!!
!
!
!
!!! ! !!!
! !
!
!
!!
!
!!
!!
!
!!
!!
!!
!!!!
!
! !!
!!
!!! ! !
! !
!
!
!!
!
!
!!!
!
!!
!
!
!!!!
!!
!!!
!!!!
!! !!
!
! !
!
!!
! !!
!!
!
! !
!
!!
!!
!
! !!!
!
!
!
!!
! !
!!
!!
!
!
!
!
! !
!
!
!
!!
!
!
!!
!
!
!!
!!
!
!!
!
!
!!
!
!
!
!!
!
!!!
!
!
!!
!!
!! !!
!
!
!!
!
!!
!
!! !
! !!
!
!!
!!
!
!!
!
!!
!!
!
!!
!
!!
! !
!! !
!!
! !
!
!!!
!
!!
!
!!
!
!
!!!
!!
!
!
!! !!
!!
!!
!
!!
! !
!!
!
!!!!
!!
!!
!
!
!!!
!
! !
!
!!
!
!
!!
!
!!
! !
!
!
!
!!
!!
!!
!
!!
!
!
!!
!!
!! !!
!
!
!!
!
!!
!!!!
!
!!
!
!
!!
!!
!
!
!!
!
!
!
!!
!
!! ! !
!
!
!
!!
!
!
!!
! ! ! !!
! ! !
!!!!
!
! ! !! !!
!
!!
!
!
!!
!
!
!!
!!
!!
!! !
!!
!! !
! ! !
!
!!
!!
!
!!
!
!!
!!
!!
!!
!!
!!!
!!
!!
!
!
!!
!! !
!! ! !
!
!!!
!
!!
!
!
!!
!
!!
!!!
!!
!!
!!
!!
!
!!
!
!
!!
!!
!
!
!!
!
!!!
!
!
!!
!!
!
!
!!
!
!
!! !
!
!
!!
!
!
!
!!
!
!
!
!!
!
!!
!
!
!
!!
!
!
!!
!!
!!!
!! !!! !
!!
!!
!
!
!!
!
! !
! !!
! !
!
!!
!
!
!!!
!
!!
!!
!
! !
! !! !!!
!
!
!
!!!!!!
!!
!!
!
!!!
!! !
!
!! !
!!
!! !!
!
! !!
!!
!
!
!!
!!
!
!
!! !
!
!!
!
!
!!
!!
!!
!
!
!!
!
!
!
!
!!
!!
!
!!
!!
!
!
!!
!
!!
!!!!!
! !!
!!
!!
!!
!!
!!
!!
!
!
!
!
!!
!
!
!
!!
!
! !
!
!
!
!
!
!!
!!
!
!
!!! !
!!
!
!!!
! ! !!
!!!
!
!
!!
!! ! !!!!
!
!!
!!! !
!
!
! ! !!
!!
! !
!!
!!
!!
!!
!
!!
!!
!!! !! !!! !
! !
! !!
! !
!!!
!
!!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!!
!
!!!!
!! ! !
!!
!
!
!!
!
!
!! !
!
!
!!!
!!
!!
!!
!
!
!!
! !!
!
!!
! !
!!!
!
!!
! !!!
! ! !!!!!!!
!!
!!
!
!!
!!
!
!
!
!
!!
! !
! !!
!
!!
!
!!
!
!
!!
!!
!
!!!
!!
!!
!
!
!
!!
!
!
!!!
!!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!!!
!!
! !!! !
! ! ! ! !!!! ! !!
!
!!
! !
!
!! !! ! ! ! ! !!
! !! !
!!!
! !!
!!! !!!!
!
!!! !
! !
!
! !
! !
! !
! !
!! ! !
! !
! !
! !
! ! !! ! !! ! !!! !
!
!! !
! !
! !
! ! !! !
! !
! !
! !
! ! ! !! !

!
!
!
!!!
!!! !!! !! !
!!
! !!! ! ! ! !!
!! ! ! !!
! !
!!
!
!!
!!! !!!
!!
!
!
! !
!
!!
!
!!
!!
!
!! !
!
!
!!
! ! ! !
! ! !
!!! !! ! !! ! ! !
!! !
!
!!
! !!!!!!
!!
!!!
!!
!!
!!
! !!!!!!
!
!!!
!!
!
!!
!
! !
!
!!!!
! !
!
!!
!!
!
!!
!!
!
!
!!!
!!!!! !
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!!
!
!
!
!
!!! !!
! !
!!! !!
!!!
!!
!
!
!!
!!
!!!
!!
!
!!
!!
!!!!
!! !
!
!
!
!
! !!!
!
!
!
!
!
!!!
!
!!!
!
!
!
!!
!
!!
!
!
!!
!
!!!
!
!
! !
! ! !!
!!
!
!
!
!!
!
!
!
!!
!!
!!
!
!
!
!
!
!!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!!!
!!
!! !
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!!
!
!
!
!!!!!! !!
!
!
!
!!
!
!
!
!!!
!!
!
!
!
!!!
!
! !
!
!!
!
! !
!
!!
!!
!
!!
!
! !
!!
! !
!!!!
!!
!
!!!!!
!!
! !!
!
!! !
!
!
!!
!
!
!!
!
!
!!
! ! ! !
!
!!
!! !
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!!
!
!!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!!
!!!
!
!!
!
!!
!
!
!
!
!
!!
!
!!
!
! !
!! !
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!!
!
! !!!!
!!!
!
!
!
!
!
!
!
!!
! !
! ! !!
!!
!!
!
!
!!
! !
!!! ! ! !
! ! ! !
!!
!
!!!
!
!!!!
!! !
! !!
!!
!
!!
!
!!!!!
! ! !!
!!
!
!!
!
!!
!
!!! ! !!
!!
!
!!
!! !
! !!!
!
!!
!!!!! !
!! !!! !
!! ! !
!!! !
!!
! !
! !!!!
! !
!!
!
!
! !
!
!!!
!
!!!
!
!
!
!!
!
!!!!
!!! !
!!
!!
!
!
!!
!
!!
!!
! !
!
!!!! !!
!!
!
!! ! ! !!! ! ! ! !! ! ! ! !!! ! ! !
! !
!!
!!
!
!!!
! ! !
!! !
! ! !!! !
! !!! ! ! !!
! ! ! !!! !! !
!
!!
! !!! !
!!
!!
age
!
!
! !
! ! ! ! !! ! ! !
!!
!! !
! !
! ! ! ! !!
! ! ! ! ! ! !
! !!
! !!!
!!! !!!!
!
!! ! !
!!!! !!!!!!!! ! ! !
!!!
!!
!
!
!!!
!!
!! !
!!
!
!
!!
!
!!!
!!
!!!
! !
!
!!
!!
!
!!
!
! ! !! !!
!!
!!!
!
! ! !!!
!!
!
!!! ! ! !!!
!
!!
!
!
!!
!
!!!
! !!
!!!
! !!
!
!!
! !
!!
!!
!!
!!!
!!
!!! !!
!
!!!
!!!
!!

60
!! ! !!! !!
!
!!
!
!!! ! ! !
!! !
! !!
!!
!! ! !!
!!!!! !! !!!
!! ! !
!!! !!
! ! !
!! ! ! ! !
! !!
!!!!! !!!
! !
! !!
!!
!
! !
!
! !
! !! ! !
!
!
!
!!
!!
!!
!!
!!!!
! !!
! !
!
!
!
!
! !!
!
!!
!
!!
!
!!
!
!
! !!
!! !
!!
!
!
!
!!
!!!
!
!!! !
!
!!!
!
!
!
!!
!
!!
!!!
!!
!! ! !!
!
!
!
!!
!
!
!!!
! !
!
!
!
!
!
!
!
!
!
!
!
!!!
! !!
!! !
! ! !
!
!
!
!
!
!
!!! !!
! ! !!
!
!
! !!
!
!
!!
!
!
!
!
!!
!
!!
!
!!
! !!!!
!
!
!!!
! !!!!
!!
! !!
!
!
! !!
! !
!
!!
!
!!!
!! !! !
!
! !
!
!
!!
!
!
!
!
!
!
!
!!!!
! ! !
!
! !
!!
!
!
!
!
!!!! ! !
!
!
!
!!
!
!
!
!!
!
!!
!
!!!! !
!
!!!
!
!
!
!!!
!!
! !
!
!!!
!
! !!
! !
!
! !
!
! ! !
!
!!
!
!
!
!
!
!!
!! ! ! !!!!
!!
!!!!
!!
!
!!!!
!
!!! !
!!
!!
!
!
!
!
! !
!!
!
!!
!!
!
!
!
!!
!
!!
!
! !
!
!!
!
!
!
!
!
!
!!!
! !
!!
!
!
! !
!!! !!!! !
! !! !
!
!!!!
!!
!!!!
!! ! !
!!
!
!!!
!!!
!!
!
! !!!!
!
!
!!
!
!
!!
!!!!!
!
!!!
!
!!
!!!
!
! !
!
!! !!!! !
!!
!
!
!!! ! !
!
!
!!!
!!
!
!!
!!
! !
! ! !!!
!!!!!!!!!
!!!
!!! !!!!
!!
!
!
! !
!!
!
!!
!
!!
!
!
!!
! ! !!!
!
!!
!
! ! !
!! !!
!
!
! !
!!
! !
!!
!!!
!! !!!!!!
!
! !
!!
!!
!
!
!!
!!
!
!
!
!
!
!
!
!
!
!!
!!
! !
! !
! !
!
!
!
!
!
!
!!
!
!
!
!!
!!!
!!!
!
! !
!!
!
!
!
!!
!
!!
!
!
!!
!
!!
! !
!!!
! !
!!!
!
!!
!
!!
!!
!!!
!!
!!!
!!
!
!!
! !
!!!
!
!
!!!
! !
!
!
!!!
!
!
!!
!!
!
!!
!!!
! !
!
!
! !
!
!!!
!!!
!!
!!
!
!
!
!
!
!!!
!
!! !! !
!!
!
!
!
!
!!
!
!
!
!!
!
!!
!!!
!
!!
!!! !
!
!!
!
!!
!
!!!
!
!
!!
!
!!
!
! ! !
!
!
!
! !
!!!
!
!!
! ! !! !
!!!
! !! !
!
!!!!
!
!
!
!
!!!!! !
!
!
!
!
!
!
!
!!
!
!!!
!
! !!
!
!!
!
!
!!
!
!
!!
!
!! !!!
! !
!!
! !
!!
!
!!!!!! !
!!
!
!!
!
!
!!
!!
!
!
! !
!!
!
!
!!
!
!
!
!!
!
!
!
!
!!!!
! !!!
!!
!
!
! !
!
!!!
!!
!!
!
!
!!
!
!
!!
!
!
! !
!
!
!!
!
!
!
!
!
!!
!
!
!
! !
!!
!
!!
!
!
!
!
!
!
!
!
!!
! !
!!
!! !!!
!
! ! !! !!
! !!!!!!
! !! !
!
! !
!!!
!
!
! !!!
!
! !
!
! ! !! !
!
! ! !!
!!
! !!
!!
!
!!! !!
!!!!!! !!! !!
! !! !!!
!
!
!
! !!
!
!!
! ! !
!
!!!
!! ! !

0
! ! ! ! ! ! !
! !! ! ! ! !!! !!!
! ! ! !!!! !
!! ! !
! !
!!
! ! !! !! ! !
!!!!
! ! ! ! ! !! ! !!
! !!!!
!!!!
dis
! ! ! !
!
! ! !! !
!
!!!
! !!!!!! !
!! !
8

! !
! ! !
! !
!! !
! !! ! !!
!
!!
!! !
!!
!
! ! ! !! !! !
!
! !!!!!!!
!
!! ! !!
! ! !
!!
!
!
!
!
!
!!
! !
!
!
!
!
!
!!
!
!!
!
!!
!! !
!
!!
!
!
!
!
!
!
!
!
!
!
!
!!
!!
!
!!!! ! !
!
!
!!
!
!!!!
!
!
!!
!
!
! !
!!!
!! !!!!
!
!
!!!
!!!
!!
!
!
! !
!!!
! !!!!! !!
!
!!!
! ! !
!!
!
!
!
!
!!
!
!!!!
!
! !! !
! !
! !
!
!
!
!
!
!
!
!
!
!
!!
!!
! !!
!
!!
!
!!
!
!
!
!
!
!
!!
!
!
!
!
!!
! !
! !
!
!!
!!!
!!
!!
!!!!
!!
!
!!!
!!
!!!
!
!!
!
!!
!
!
!
!!!!!!!
! !
!!
!!
!
!
!!
!
!
!
!
! !
! !
! ! !!
!!
!
!!
!
!
!!!
!!!
! !! !
! !
!
! !
!
!
!!
!!
!
!
!
!!
!!
!! !
!
!
!
!!
!
!
!
!!
!
!
!
!!!
!
!
!
!
!!!!
!!
!! !! !!!!!
!!
!
!!
!!
!
!!!!
!
!
!
!
!!
!
!
!!
!
!!
!!
!!
!
!!
!!
!
!
!!
!!
! !!
!!
!
!
! !
!
!
!
!
!
!
!
!!
!!
! !
! !!!! !!
!
!
!!!!
!!
! !
! !!
!! !
!
!
!
! !
! !!!
!
!
!
!
!
!
!
!
!
! !!
!!
!
!! ! ! !
!
!!
!!
!!
!!!
!
!!
!
!
!!
!!
!
!!
!!
!
!
!
!
!!
!
!!
!
!!!
!
!!
!!
!!
! !
!!
!!! !
! !
!! !
!!!
!
!
!
!
!
!
!!
!!
!!
!
!!
!
!
!!
!
!!
!
!
!
!
!
!!
!!!
!!
!
!!
!!
!
!!
!
!
! !
!!
!!
!
!
!!
!
!
!
!
!
!!
!
!
!!
!! !!! ! ! !!!! ! ! !!!
!! !
!!! !
!!
!!
!!
!!!
!
!
!!
!!!!
!!!!
!! !!!
! !
!
!!
! !
!!
!!!
!
!
!!
!
!
!!!
!
! !!
! !!
2

!
!
!
!
!!
!
!
!
!
!!
!
!!
!
!
!
! !!
!!
!
!! !! !! !
! !
! ! !!
!! !
! !
! !
!
!
!!
!
!!
!
!
!! ! !!!
!!
!!
!
!!
!
!
!!
!
!
!!
!!
!!!
!
!!
!
!!
!
!!
!!
!
!!!
!
!!
!! !
! !
!!
!
!!
!
!!
!
!
!
!!
!
!
!
!! !
!
!

FIGURE 8 5 A sca erp o ma r x for a four een var ab es n he Bos on da ase

164 Graphical Data Analysis with R

Parallel coordinate plots

Figure 8.6 is the default parallel coordinates plot using the function parcoord from
the package MASS. Some of the features can be seen that were identified in the
scatterplot matrix display, at least for those variables with adjacent axes. There is
quite a lot of information on the distributions of the individual variables, such as the
skewness of crim and the gaps in rad, tax, and possibly black.
Parallel coordinates are most effective used interactively, when groups of points
can be selected across all axes and be compared with the rest. This can be done in
the package iplots and its possible successor, the package Acinonyx, which is in
development. Figure 6.19 shows the same plot, giving some of the flavour of interac-
tion, with the points where rad= 24 highlighted in blue and with the variable axes
ordered by differences between those cases and the rest.

data(Boston, package="MASS")
par(mfrow=c(1,1), mar=c(2.1, 1.1, 1.1, 1.1))
MASS::parcoord(Boston)

FIGURE 8.6: A parallel coordinates plot for all fourteen variables in the Boston
dataset. Details are discussed in the text.

Heatmaps
With a heatmap each case is represented by a row and each variable by a column.
The individual cells are coloured according to the case value relative to the other
values in the column. For this purpose the variables are standardised individually,
either with a normal transformation to z scores or adjusted to a scale from minimum
to maximum. (It is possible to colour according to all values in the dataset, although
that is unwise, as it emphasises differences between the levels of different variables
rather than differences between individual cases.)
As the orderings of cases and variables may be freely chosen, it is helpful to try
to order them in an informative manner. Clustering or seriation methods can work
Getting an Overview 165

library(gplots)
heatmap.2(as.matrix(Boston), scale="column", trace="none")
Color Key
and Histogram
1000 2000 3000 4000
Count
0

!5 0 5
Column Z!Score
328
325
295
216
38
39
214
296
327
176
99
294
326
206
71
74
72
73
17
230
229
311
215
239
244
333
290
304
303
250
251
253
252
254
13
332
351
350
302
247
69
70
68
248
245
246
249
346
347
63
82
61
1
64
60
84
342
242
241
243
240
93
266
95
268
269
275
279
278
277
276
81
83
59
94
285
349
194
195
288
286
289
348
202
67
66
203
355
356
300
299
57
255
256
199
197
198
192
189
193
191
190
188
353
352
200
201
344
301
345
105
104
110
108
106
109
107
144
143
145
148
165
500
498
497
501
101
102
77
167
162
158
163
164
136
138
129
132
131
140
130
128
134
114
113
133
137
139
141
142
112
118
115
117
120
343
172
152
153
161
170
155
116
119
160
159
151
149
150
494
111
499
78
76
79
495
80
496
329
330
331
75
55
169
171
135
154
166
168
28
26
19
35
33
318
319
20
7
14
16
320
312
223
237
238
178
175
179
231
236
235
232
228
225
233
234
226
97
91
92
503
502
324
36
208
100
90
98
322
321
323
297
177
217
213
207
298
209
37
96
313
315
173
174
309
314
310
316
22
29
317
18
30
15
27
221
224
227
23
25
24
222
9
11
12
1
8
210
32
31
34
219
211
220
218
504
505
89
506
210
212
62
264
260
259
261
267
263
258
262
265
48
49
339
336
334
335
337
340
47
46
4
5 5
6
4
86
88
85
87
3
338
341
50
52
51
270
273
274
65
307
308
306
282
280
305
271
281
283
54
53
272
43
42
44
184
183
181
185
121
186
182
180
187
125
123
126
122
124
127
291
293
292
40
41
196
257
58
204
205
56
287
284
354
378
390
449
391
396
398
395
394
477
397
443
384
383
393
440
382
380
402
379
404
441
376
387
401
386
388
375
374
399
444
442
448
453
357
454
479
403
480
392
363
361
389
407
460
463
471
359
472
464
358
462
487
360
473
483
482
492
489
493
469
474
366
365
373
370
369
372
371
488
486
484
485
465
481
470
381
406
447
422
421
409
367
476
450
408
468
405
452
475
362
364
377
478
400
466
490
491
385
423
459
461
445
414
410
424
458
457
455
451
425
437
417
467
438
426
428
427
413
416
412
446
456
420
419
411
435
434
436
429
433
368
418
432
431
439
430
415
157
156
103
146
147
age

zn

ptratio

medv

crim

chas

nox

dis

rm

rad

indus

lstat

black

tax

FIGURE 8.7: A heatmap for the Boston dataset. Some features can be discerned with
a little effort (see text).

well and you just have to bear in mind that each method will give different results.
Figure 8.7 shows a heatmap of the Boston data using the package gplots.
It is difficult to see much, although certain patterns are apparent, such as the
group of relatively high values for the variable black because of the shape of the
distribution, and the blocks of equally shaded values on some variables in the lower
section of the plot. It is difficult to see much more. The colour legend top left with
the superimposed histogram of values is useful, because we can see that although
there are many different possible shades, most of the data values lie in the centre
of the scale and only a few shades have been used. Using a different colour palette
and possibly a nonlinear scale could make the display more enlightening. Experi-
166 Graphical Data Analysis with R

menting with various colour schemes and clustering methods might reveal additional
information.
All this makes heatmaps a fairly subjective tool and it is one of those graphic
displays which can be effective for particular structures in some datasets, but which
cannot be relied upon to produce good results in general.

Glyphs
With glyphs each case is represented by a multivariate symbol reflecting the case’s
variable values. As for heatmaps, each variable must be standardised in some way
first and this can influence the way the display looks a lot. The type of symbol used
is also relevant and makes a big difference. It could be the oft discussed Chernoff
faces, profile charts, star shapes, or some other form. Whatever is used must have at
least the same number of dimensions as the number of variables in the dataset and
each variable is allocated to one (or more) dimensions.
In Figure 8.8 only glyphs for the first four cases have been drawn to show some
of the details of the plot. The stars function has been used and the segments have
been coloured for better effect using a rainbow palette. You really need a big screen
or zooming capability to appreciate the display of the full dataset.

par(mar=c(1.1, 1.1, 1.1, 1.1))

palette(rainbow(14, s = 0.6, v = 0.75))
stars(Boston[1:4,], labels=NULL, draw.segments = TRUE)

FIGURE 8.8: Glyphs (stars) for the first four cases of the Boston dataset.
Getting an Overview 167

Figure 8.9 shows the result for the whole Boston dataset. It looks like that there
are several distinct groups in the dataset, as we can see groups of different shapes.
Surprisingly the data show some evidence of grouping already. As with heatmaps,
the allocation of the variables to the dimensions (in Figure 8.9 this is the ordering
of the variables round the star), the scales used, and the ordering of the cases can
strongly influence what information can be detected.

stars(Boston, labels=NULL, draw.segments = TRUE)

FIGURE 8.9: Glyphs (stars) for the Boston dataset. The bigger glyphs towards the
foot of the plot are the cases where the variable rad takes the value 24.
168 Graphical Data Analysis with R

8.4 Multivariate overviews for categorical variables

All the displays in the previous section are primarily for continuous variables, al-
though they can be useful for categorical variables sometimes too. If you want to
look at a small group of categorical variables together, then some kind of mosaic-
plot is best. This can be useful in checking experiments to see whether a study is
unbalanced, and, if so, how.
The famous barley dataset, which Cleveland reanalysed in his book [Cleveland, 1993],
has three categorical variables and one yield measurement for each combination. You
can immediately see that the experiment is balanced by drawing a mosaicplot of the
categorical variables and observing the resulting regular pattern.
The dataset foster in the package HSAUR2 has two categorical variables, the
mother’s genotype and a litter genotype. Figure 8.10 shows that the structure is unbal-
anced, as the rectangles representing the variable combinations have different sizes.

data(foster, package="HSAUR2")
mosaic(~litgen+motgen, data=foster)

motgen
A B I J
A
B
litgen
I
J

FIGURE 8.10: Genotype groups in the foster dataset using a mosaicplot. There are
not equal numbers in the different combinations.

It is more informative to use a multiple barcharts version of a mosaicplot, which

can be drawn using ggplot2’s functionality, as in Figure 8.11. Then it is easier to see
just which groups are smaller or bigger than average.
As was mentioned in Chapter 7, there are distinct limits to the numbers of cate-
gorical variable combinations that can reasonably be displayed and understood. This
is fine for monitoring experimental designs, as experiments usually only have a re-
stricted number of combinations. It can be an issue in large surveys where there
may be many classifying variables to be taken into account. The dataset HI in Ecdat
Getting an Overview 169

ggplot(data=foster, aes(motgen)) + geom_bar() +

facet_grid(litgen~ .) + xlab("") + ylab("") +
scale_y_continuous(breaks=seq(0,6,3)) +
labs(title="litter genotype by mother’s genotype")

litter genotype by mother's genotype

A
0

3
I

3
J

0
A B I J

FIGURE 8.11: Genotype groups in the foster dataset using a multiple barchart. Num-
bers of rats vary across the different genotype combinations.

includes information for 22, 272 married women on region (4 categories), race (3),
education (6), as well as three variables on insurance status amounting to 6 different
categorisations. This gives 432 combinations in total that might be of interest. There
is also a weighting for each case.
170 Graphical Data Analysis with R

8.5 Graphics by group

Sometimes there are known groups in a dataset and it is important to get an overview
of variable values split by group. There are two typical situations that arise: you
can have several grouping or conditioning variables and just a couple of variables to
display or you can have a single grouping variable and many variables to display.

Trellis graphics
Trellis graphics [Becker et al., 1996] are ideal for the first case. They were introduced
some twenty years ago to effectively display data for large numbers of subsets. Each
component plot or panel shows the same basic display for a different subset of the
data, but each has the same scaling to allow comparison. Subsets can be defined by
categorical variables, by discretisations of continuous variables and by combinations
of variables. Each of the individual panels is a conditional view of the data.
In R you can use the packages lattice or ggplot2 for this. There was an excel-
lent comparison of the two approaches on the blog Learning R [rlearnr, 2009] in
2009. Trellis graphics can be very effective and some people use them a lot. There is
extensive information about the lattice package in [Sarkar, 2008] and on the accom-
panying webpage. Information on ggplot2 is available in [Wickham, 2009] and on
the ggplot2 webpages.
Figure 8.12 shows a lattice display for the barley dataset. Apart from using filled
circles for the points rather than open circles, this plot just uses the graphics defaults.
It shows that the increasing yields across the sites from Grand Rapids to Waseca hold
for all varieties except for Peatland. You can also verify Cleveland’s observation, that
the 1932 yields are almost always less than the 1931 yields, except for Morris where
it is the other way round. This is easier to see with Cleveland’s plot, where there
are six panel plots, one for each site, rather than this plot where there are ten panel
plots, one for each variety. On the other hand, the different pattern for the Peatland
variety is easier to see with this plot. As always it is best to look at a selection of
graphics. Cleveland concluded that the Morris data for the two years must have been
switched. Recently Wright has re-examined the dataset using additional sources of
supplementary data and suggests that in the light of the variability in the data he has
found, the Morris data are quite plausible as they are [Wright, 2013].
Trellis graphics may be drawn in many different ways, depending on the choice
of panel variables and panel plot, depending on the conditioning variables and what
order they are in, depending on the order of categories within a conditioning variable
(lattice plots the sites and varieties for barley in increasing order of median yield, as
suggested by Cleveland, because that is how the factors are ordered in the R dataset),
and depending on how the individual plots are arranged on the page. How well the
plots look and how informative they are also depends very much on the size and
aspect ratio of the overall display. Draw Figure 8.12 yourself and experiment with
growing and shrinking the window in both directions.
Getting an Overview 171

library(lattice)
data(barley, package="lattice")
dotplot(site ~ yield |variety , data = barley,
groups = year, columns=2, pch=16, col=c("red","blue"),
key = list(text=list(levels(barley$year)),
points = list(pch=16, col=c("red", "blue"))),
xlab = "Barley Yield (bushels/acre) ", ylab=NULL,
main="Barley Yields by Site for ten Varieties")

Barley Yields by Site for ten Varieties

1932 !
1931 !

20 30 40 50 60

Wisconsin No. 38 Trebi

Waseca !
! ! !

Crookston ! ! ! !

Morris ! ! !!

University Farm !! ! !

Duluth !! !!

Grand Rapids ! ! ! !

Velvet Peatland Glabron No. 457

Waseca ! ! ! ! ! ! ! !

Crookston ! ! ! ! ! ! ! !

Morris ! ! ! ! ! ! ! !

University Farm ! ! ! ! ! ! ! !

Duluth ! ! !
! ! ! ! !

Grand Rapids ! ! ! ! ! ! ! !

Svansota No. 462 Manchuria No. 475

Waseca ! ! ! ! ! ! ! !

Crookston ! ! ! ! ! ! ! !

Morris ! ! ! ! ! ! ! !

University Farm ! ! ! ! ! ! !

Duluth !! ! ! ! ! ! !

Grand Rapids ! ! ! ! ! ! ! !

20 30 40 50 60 20 30 40 50 60

Barley Yield (bushels/acre)

FIGURE 8.12: A lattice plot of the barley data. Yields at six different sites for two
years are shown in ten plots, one for each variety. Yields in 1932 were generally
lower than in 1931 and there is a common pattern across sites for most varieties.

Like mosaicplots, trellis graphics can in theory include unlimited numbers of combi-
nations, in practice the individual plots become too small if you try to get everything
on one page. When trellis graphics were first introduced, applications were described
using hundreds of plots printed on many, many pages. This is a sensible approach if
you are looking for individual plots which stand out, although it is difficult to get
an overview and an idea of overall structure. For designed experiments and other
structured datasets it will often be possible to organise all plots on one page, as the
number of combinations usually remains limited.
172 Graphical Data Analysis with R

Group plots
When there is only one grouping variable, it is interesting to look at how all other
variables vary in parallel and for this a group plot can be drawn. Either there is a
column for each variable to be displayed and a row for each group or the other way
round. Mostly columns are better for comparisons, unless boxplots are used. Con-
tinuous variables may be plotted as histograms (density estimates or other displays
could be used), whereas categorical and ordinal variables may be plotted as bar-
charts. In group plots all plots for the same variables are drawn to the same scale to
aid comparison.
Figure 8.13 shows an example for the uniranks dataset for UK universities from
the GDAdata package, which was discussed in §6.6. Histograms have been drawn
for all the nine variables for each of the six groups of universities. Rather than loop
through the variables, the code constructs a long version of the dataset and then
uses facetting to arrange the plots. The scales = "free_x" option allows each
column to have its own x-axis scale.
Several interesting features can be seen: the top performance of the Russell group,
the good performance of the 1994 group, the range of performances for the univer-
sities which do not belong to any group, and the roughly equally poor showing of
the other three groups. Inspecting individual columns more closely, you can see the
sharp divisions on EntryTariff between the groups and to a lesser extent a simi-
lar effect for AvTeachScore.
Both trellis graphics and group plots are made up of lots of individual plots. With
trellis displays every individual plot is of the same type, shows the same variables,
and has the same scale, they just each show a different subset of the data. With
group plots each column is like a trellis plot, but different columns show different
variables and have different scales while each row shows the same subset. For all
continuous variables or all categorical variables each row is like a set of histograms
or barcharts respectively, treating that subset as a dataset in its own right, as described
in Section 8.2. The only differences are that the plots are drawn in one row and the
scaling of the individual plots depends on the scaling of the other plots in the same
column for the other subsets.
Group plots are not seen very often, which is perhaps surprising. They are simple
to understand and they offer effective multivariate comparisons. Since their main
advantage lies in comparisons down columns there is no drawback in drawing several
of them if there are many variables involved. Figure 8.13 is just about large enough
to include all nine variables. Had there been more variables, then they could have
been drawn in several displays over more than one page.
 AvTeach NSSTeach NSSOverall SpendperSt StudentStaffR Careers VAddScore EntryTariff NSSFeedb
8
6
4
2
0
8
6
4
2
0
Getting an Overview

8
6
4
2
0 1994 Group Guild HE
8
6
4
Million+

2
0
8
6
4
2
0
8
6
4
2
Russell Uni Alliance

0
40 60 80 100 75 80 85 90 95 707580859095 2.5 5.0 7.5 10.010 15 20 25 304050607080 2 4 6 8 20030040050060050 60 70 80

FIGURE 8.13: A group plot of the uniranks dataset. The Russell universities tend to get top scores on all criteria (it is better to have
low values for StudentStaffRatio) except for NSSFeedback. The top row representing universities who are not members of any
group have a spread of values on all criteria. The 1994 universities look to be second best to the Russell universities.
data(uniranks, package="GDAdata")
names(uniranks)[c(5, 6, 8, 9, 10, 11, 13)] <- c("AvTeach", "NSSTeach", "SpendperSt", "StudentStaffR",
"Careers", "VAddScore", "NSSFeedb")
ur2 <- melt(uniranks[, c(3, 5:13)], id.vars="UniGroup", variable.name="uniV", value.name="uniX")
ggplot(ur2, aes(uniX)) + geom_histogram() + xlab("") + ylab("") +
facet_grid(UniGroup~uniV, scales = "free_x")
173
174 Graphical Data Analysis with R

8.6 Modelling and testing for overviews

1. Transformations
If a first look shows that distributions of variables are highly skewed, it may be
constructive to transform the variables. The Box-Cox family of transformations
is a good choice. If nominal variables have many categories with small frequen-
cies, it may be helpful to combine or delete some. Otherwise, tests mentioned in
Chapters 3 and 4 may be appropriate.
2. Associations (and causality)
When variables show associations in a scatterplot matrix, then linear or non-
linear models, depending on the form of association, should be considered.
3. Linear models
If one or more panel plots stand out as being different in a trellis display, then a
linear model based on the conditioning factors might be used to confirm this.
4. Discrimination between groups
Discriminant analysis could be used in conjunction with group plots, as could
other supervised learning methods, including Support Vector Machines.

Main points
1. Sets of univariate graphics are good for giving a quick overview of variable val-
ues and distribution patterns (e.g., Figures 8.1 and 8.2).
2. Scatterplot matrices are valuable for identifying bivariate patterns even with quite
a few variables (cf. Figure 8.5).
3. Parallel coordinate plots are useful for studying groups of cases and are most
effective when they are interactive (§8.3).

4. Trellis displays are excellent for comparing data on one or two variables by sub-
sets, while group plots provide a useful overview across many variables for a
small set of subgroups (cf. Figures 8.12 and 8.13).
5. Other multivariate displays look interesting, but scaling, colours and orderings
all have to be suitably chosen and the resulting graphics can still be difficult to
decode (§8.3).
Getting an Overview 175

Exercises
1. Cloud seeding (clouds from the package HSAUR2)
What kinds of variables are there in this dataset? What plots would you recom-
mend to help people get to know the dataset?

2. Longley’s Data (longley from the package datasets)

Longley’s dataset is well known as an example for highly collinear regression.
Can you see this from the scatterplot matrix? Are there other features worth
noting?

3. US States (state.x77 from the package datasets)

What kinds of variable are there? Is there anything interesting or unusual in the
univariate distributions? Compare what information you might get from each of
the multivariate displays discussed in this chapter: a scatterplot matrix, a parallel
coordinates plot, a heatmap, and a collection of glyphs.
4. Crabs (crabs from the package MASS)
[Venables and Ripley, 2002] uses this dataset in discussing classification and dis-
crimination. The authors initially transform to a log scale and then write that “The
data are very highly correlated and scatterplot matrices and brush plots [i.e. in-
teractive graphics] are none too revealing.” Using graphical overviews, comment
on whether the log transform was a good idea and whether you agree with their
statement on the correlations.

5. Pima Indians
Data about diabetes amongst adult females of the Pima Indians is available in
R, in the packages MASS and MMST. Both use the 532 cases with complete
records. A larger version of the dataset with 768 cases is available from the UCI
Machine Learning Repository [Bache and Lichman, 2013]. Download the larger
dataset and give an overview of the differences between the cases available in R
and the rest using two groups plots, one for the continuous variables and one for
the variables npreg and type.
6. Exam results in London
The dataset Exam in mlmRev is used to illustrate multilevel modelling. Prepare
an initial graphical summary of the data and summarise your results in three main
conclusions.
7. Distance to college
The dataset CollegeDistance in AER is from a survey of high school students. If
you had to prepare a one-page summary of the main information you can find by
exploring the data, what graphics would you use?
176 Graphical Data Analysis with R

8. US traffic fatalities
Numbers of various kinds of traffic fatalities for each of six years are given for
each state in the contiguous United States in the dataset Fatalities from the pack-
age AER. There are 32 variable values for each state and year. Carry out an
initial exploratory analysis to decide what information you would present to give
people a first impression of the data.
9. Intermission
The painting Britain at Play by L.S. Lowry hangs in the Usher Gallery in Lin-
coln. Does the title match the picture well? Can you see how the British ‘played’
in the early 1940s?
9
Graphics and Data Quality: How Good Are
the Data?

Beauty is less important than quality.

Eugene Ormandy

Summary
Chapter 9 discusses data quality, including missing values and outliers, what they
are, and ways of identifying them.

9.1 Introduction
Good analyses depend on good data. Checking and cleaning data are basic tasks that
always have to be carried out and yet are rarely explicitly discussed. Textbooks—and
indeed R packages—often present datasets only after they have been cleaned and
filtered. The detailed work that has gone into getting them into their semi-pristine
condition is passed over and sometimes even swept under the carpet. One of the
problems is that there are so many different ways that data can be of poor quality, or
as Hadley Wickham put it in a contribution to R Help, paraphrasing Tolstoy: “every
messy data is messy in its own way”.
There are some general principles, some techniques which are useful fairly often,
and a whole raft of special cases. Graphics are helpful for identifying these problems
and can sometimes shed light on how they can be solved. Even if you cannot solve a
problem, it is important to know how serious it is.
Data quality problems may include multiple codings for the same category (e.g.,
Male, male, m, M, man, . . . ), measurement or data entry errors, data heaping, gaps in
the data, ambiguous definitions of variables, shifted chunks of data, and, of course,
missing values and outliers, which are both discussed in this chapter.

177
178 Graphical Data Analysis with R

9.2 Missing values

Missing values in datasets can lead a rather shadowy existence. If there are very few
of them, they may be just ignored. If there are more of them, then values may be
imputed to replace them. In either case reported results may not make clear that there
were any missing values in the dataset. Political opinion polls generally report party
support for the people who said they would vote for a particular party. The ‘don’t
knows’ may only be referred to in the small print, if at all, and can be a sizeable
proportion of the electorate.

Visualising patterns of missing values

Graphical displays can assist in summarising how many missings there are and in as-
certaining if there are patterns of missingness. In statistics textbooks there tend to be
few dataset examples including missing values, possibly because they add irrelevant
difficulties that may distract from the theory being illustrated in the examples. Real
datasets often have missings of one kind or another.
In the R package mi for multiple imputation by Bayesian methods there is a
dataset CHAIN, used earlier in Exercise 6 of Chapter 3. Figure 9.1 shows the patterns
of missings using the display provided in the package.

data(CHAIN, package="mi")
par(mar=c(1.1, 4.1, 1.1, 2.1))
mi::missing.pattern.plot(CHAIN, y.order=TRUE, xlab="", main="")

here.W1

cs37.W1

pcs.W1

age.W1

c28.W1

b05.W1

h39b.W1

FIGURE 9.1: A missing pattern plot of the CHAIN dataset in the package mi. The
rows represent the seven variables and the columns the 532 cases. Where a value is
missing on a variable for a case the corresponding cell is marked in red. The variables
have been ordered by numbers of missings and the cases clustered. All variables have
missing values for a few cases.
Graphics and Data Quality: How Good Are the Data? 179

We can immediately see the curious fact that a number of cases have no values,
just missings (the group on the left), and that only three variables have other missing
values, since only those rows have other red cells. As this display attempts to show
all entries for all cases individually, it is only feasible for smaller datasets.
With the function visna in the package extracat, the different missing value
patterns are displayed not the individual missing values. Figure 9.2 uses the same data
as in Figure 9.1 and additionally provides information on the proportions of missings
by variable and the relative frequency of each pattern. The figure also emphasises that
the cases with only the first variable missing make up the majority of the cases with
missings.

visna(CHAIN, sort="b")

haartadhr.W1
mcs37.W1
h39b.W1

b05.W1

age.W1
c28.W1

pcs.W1

FIGURE 9.2: A missing pattern plot of the CHAIN dataset using the package extra-
cat. The columns represent the seven variables and the rows the missing patterns.
The cells for the variables with missing values in a pattern are drawn in blue. The
variables and patterns have been ordered by numbers of missings on both rows and
columns (sort="b"). The bars beneath the columns show the proportions of miss-
ings by variable and the bars on the right show the relative frequencies of the patterns.
One variable has far more missings than the others. Four variables are only missing
for cases where all variables are missing.
180 Graphical Data Analysis with R

For some datasets the number of cases with no values missing may be very large.
The frequency bar for this pattern is then reduced in size to enable readers to distin-
guish differences in the frequencies of the other patterns and the bar is given a red
border to indicate this censoring.
In the CHAIN dataset there are seven variables, all of which have some values
missing. Potentially there could have been 128 (= 27 ) missing patterns, yet there
are only nine. This often happens and it makes the visna plot very efficient. In the
oly12 dataset in VGAMdata, which contains details of the athletes who took part
in the 2012 London Olympics, there are 14 variables with 3 having missing values.
All eight possible patterns of missings amongst these 3 variables arise, as Figure 9.3
shows. (This dataset was already discussed in §5.5.) The large number of date of
birth cases missing obscures the patterns involving weight and height. Redrawing
the plot without the variable date of birth, i.e., using the subset

oly12d <- oly12[, names(oly12) != "DOB"]

shows that weight is often missing on its own, while height is mostly missing together
with weight and rarely on its own.

data(oly12, package="VGAMdata")
oly12a <- oly12
names(oly12a) <- abbreviate(names(oly12), 3)
visna(oly12a, sort="b")
DOB

Nam

POB
Wgh

Hgh

Age

Evn
Sex

Brn
Cnt

Gld

Spr
Slv

Ttl

FIGURE 9.3: Missing value patterns in the London Olympics dataset (oly12). Date
of birth is missing very often and weight is more often missing than height.
 Graphics and Data Quality: How Good Are the Data? 181

Missings dependent on values of other variables (MAR)

To investigate whether missings may be MAR, rather than MCAR, you can com-
pare the subsets of the cases with and without missings. The dataset freetrade in the
Amelia package provides values of 8 variables for nine Asian countries for each of
19 years. Five of the variables have missing values, one of which being the country’s
average tariff rates, where about a third of the data are missing. Figure 9.4 displays
the missing pattern for this variable by year and by country. The countries have been
sorted by the numbers of years for which the data are missing and the table of miss-
ings has been visualised with a fluctuation diagram. The tariff rates are missing more
often for some countries than for others and there is little evidence of any pattern
over time. The data do not look MCAR.

data(freetrade, package="Amelia")
freetrade <- within(freetrade, land1 <-
reorder(country, tariff, function(x) sum(is.na(x))))
fluctile(xtabs(is.na(tariff) ~ land1 + year, data=freetrade))
1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
198 198 198 198 198 198 198 198 198 199 199 199 199 199 199 199 199 199 199
s e
pal lan nk es ys ia ea sta pin
Ne Thai SriLaIndonMala Ind Kor PakiPhilip
d a ia ia n

FIGURE 9.4: A plot of missing values on tariff rate information for nine Asian coun-
tries over 19 years. There is complete data for the Philippines and data for all coun-
tries in 1993. There are no striking patterns.

Reasons for missings and dealing with missings

Data can be missing for different reasons. It could be that a value was not recorded,
or that it was, but was obviously an error and was consequently coded as missing.
A value may be lost or incorrectly transcribed. In a survey people may not answer
because they don’t know, because they don’t want to answer, because there is no
option that matches their views, or because they were not asked. Or they may give a
ridiculous answer, which the interviewer then marks as ‘missing’.
182 Graphical Data Analysis with R

There is no standard missing code used by software (the NA used by R and other
software is common). In the past, numbers like 99 or 999 were used and caused
problems when people didn’t notice and calculated statistics and fitted models as
if those were real numbers. The version of the Pima Indians dataset offered on the
UCI Machine Learning Library uses 0 for entries the R version records as missings.
Figure 9.5 shows a missing pattern plot for the Pima.tr2 dataset from MASS. Most
of the records are complete. Quite a few cases are only missing a skin thickness value
and a few values are missing on two of the other variables.

data(Pima.tr2, package="MASS")
visna(Pima.tr2, sort="b")
npreg

type
skin

ped

age
bmi

glu
bp

FIGURE 9.5: A missing value pattern plot for the Pima Indians dataset. Only three
variables have missing values and no case is missing on all three.

The best that can be said is that all data (and documentation) should be looked at
carefully before analysis. Other major statistics packages (SAS, SPSS, Stata) offer
multiple missing codings to take account of different reasons for data being missing,
but R does not; codings for more than one type of missing have to be sorted out by
the user. As an example in practice, the US Multicenter Osteoarthritis Study (http:
//most.ucsf.edu/) has 13 different missing codes in its protocol.
Taking account of missing values can be important, especially if they are ‘not
missing at random’ (NMAR). This is a term that became popular when multiple
imputation was introduced [Rubin, 1987], the clever idea that instead of imputing a
single value to substitute for a missing value, you should use several. NMAR data are
distinguished from MCAR (Missing Completely At Random) and MAR (Missing
At Random) (where missingness may depend on the value of the missing entry).
One of the reasons the 1992 election in the UK was badly predicted was apparently
that conservative voters were reluctant to admit to intending to vote conservative
[Worcester, 1996], an example of MAR.
Graphics and Data Quality: How Good Are the Data? 183

9.3 Outliers
What is an outlier?
Outliers are cases which are far away from the bulk of the data. They may be errors
(a decimal point in the wrong place), genuine extreme values (some people are paid
much more money than the rest of us), rare values (occasionally a lot of people are
sick on the same day), unusual values (small people can be very heavy), cases of spe-
cial interest (exceptional performances), or data from some other source (including a
few small companies in a dataset on large companies). The last is sometimes referred
to in the statistical literature as contamination.
Whether points are judged to be outliers depends on the data or model with which
they are compared. A basketball player may be regarded as unusually tall amongst
a group of his own age, whilst being average within a group of his fellow basket-
ball players. Outliers are often talked about as if they are individual unusual values,
but you often observe groups of outliers, particularly with large datasets, and one
analyst’s outlying group may be another person’s population subset.
As groups of outliers get larger relative to the main part of a dataset, you could
say you are looking for outliers or you are identifying distinct subsets or you are
assessing how homogeneous the dataset is. This chapter concentrates on at most
small groups of outliers and does not consider the more complex problem of dataset
heterogeneity in large datasets. That is a topic for itself and depends particularly
heavily on the context and aims of a study.
Possible outliers on individual variables are easy to spot graphically, and bi-
variate outliers, which are not outlying on the individual variables, can be seen in
scatterplots. There can also be higher dimensional outliers that are not outliers on
lower numbers of dimensions and these are trickier to find. That is hardly surprising:
Try to think of what a three-dimensional outlier looks like which is neither a one-
dimensional outlier on any of its three dimensions nor a bivariate outlier on any of
the three pairs of dimensions.
Determining outliers analytically is not easy. Many procedures have been sug-
gested, but none are universally accepted, there are too many different kinds of out-
lier that might arise. Analytic approaches are useful for picking out extreme outliers
(and every method should be able to manage that) and for providing a first filtering
of the data to produce a set of possible outliers that can then be examined in more
detail.
It is worth identifying outliers for a number of reasons. Bad errors should be
corrected, genuine outlying values can be interesting in their own right, and many
statistical methods may work poorly in the presence of outliers. Robust methods have
been proposed for dealing with datasets with outliers. They are often computationally
demanding.
There is a fairly extensive literature on outliers beginning with the classic
text by Barnett and Lewis [Barnett and Lewis, 1994] and including more recently
[Aggarwal, 2013].
184 Graphical Data Analysis with R

Examples of outliers
Figure 9.6 shows a sample collection of plots for identifying potential outliers in a
subset of the USCrimes dataset from the package TeachingDemos. You can see a
boxplot of the murder rates in US States in 2010, a scatterplot of the rates of theft
and vehicle theft, and a parallel coordinate plot of the rates for nine different crimes.
The preparatory code for the plot involves some subsetting and restructuring. One
important point to note is that the dataset USCrimes is a three-dimensional array, not
a data frame, so it has to be converted to a table and a long data frame first. Then data
for the year 2010 are chosen and the data for the US as a whole are excluded. Next
the data frame is reformed with columns for each of the variables using the spread
function from tidyr. Finally the rate variables are selected:

library(tidyr)
data(USCrimes, package="TeachingDemos")
names(dimnames(USCrimes)) <- c("State", "Year", "Crime")
US10 <- USCrimes %>% as.table %>%
as.data.frame(responseName = "Rate") %>%
filter(Year==2010 & State != "United States-Total") %>%
spread(Crime, Rate) %>% select(State, ends_with("Rate"))

The boxplot identifies two possible outliers and one is clearly hugely different
from all the rest, the District of Columbia. This can be identified with

US10 %>% select(State, MurderRate) %>% filter(MurderRate > 10)

The scatterplot seems to show one extreme outlier on both theft rates, the District
of Columbia again. Interestingly, a boxplot of TheftRate alone shows no outliers.
The scatterplot also indicates two states looking like bivariate outliers. Neither Cal-
ifornia nor Nevada are extreme on the individual variables, but both have relatively
higher vehicle theft rates compared to their theft rates. Finally the parallel coordi-
nate plot, in which each axis has been scaled individually between 0 and 1, reveals
that one state has extremely high values on six of the nine variables, obviously the
District of Columbia and that another state has a particularly high rate of rape.

a <- ggplot(US10, aes("var", MurderRate)) +

geom_boxplot() + xlab("") +
ylab("Murder rate per 100,000 population")
b <- ggplot(US10, aes(TheftRate, VehicleTheftRate)) +
geom_point() +
xlab("Theft rate per 100,000 population") +
ylab("Vehicle theft rate per 100,000 population")
c <- ggparcoord(data = US10, columns = c(2:10),
scale="uniminmax") +
theme(axis.title.x = element_blank(),
axis.title.y = element_blank())
grid.arrange(arrangeGrob(a, b, ncol=2, widths=c(1, 4)),
c, nrow=2)
Graphics and Data Quality: How Good Are the Data? 185

! !

800
20

Vehicle theft rate per 100,000 population

Murder rate per 100,000 population

600
15

10 400 !
!
! !

!
! !
! !
! ! !
!
! !
5
! ! ! !
!!
!!!
200 ! !
! !
! !
!
!
! !
! !
! !
! ! ! ! !
! ! !
!
!! !

0
var 1500 2000 2500 3000
Theft rate per 100,000 population

1.00

0.75

0.50

0.25

0.00

ViolentCrimeRateMurderRate RapeRate RobberyRate AssaultRate

PropertyCrimeRate
BurglaryRate TheftRate VehicleTheftRate

FIGURE 9.6: Plots of the crime rates by state in 2010 for the USCrimes dataset, used
for identifying potential outliers.
186 Graphical Data Analysis with R

Univariate outliers
There have been many attempts to suggest rules and tests for determining whether
points should be considered outliers and none of them are satisfactory in all situa-
tions. Using tests for assessing whether points are outliers is difficult, as the tests are
primarily designed for testing individual points and you may want to test several.
The best-known approach for an initial look at the data is to use boxplots. Tukey
suggested marking individual cases as outliers if they were more than 1.5 × IQR (the
interquartile range) outside the hinges (basically the quartiles). Figure 1.8 showed
boxplots for the six continuous variables in the Pima Indians dataset. There are no
outliers on the first variable (plasma glucose concentration), both low and high out-
liers on blood pressure, and high outliers on the other four variables, with as many as
nine for the diabetes pedigree function.
Figure 9.7 displays boxplots for the seven continuous variables from the dia-
monds dataset from package ggplot2, also used in Exercise 7 of Chapter 3. First the
code selects the desired variables and then combines them all in a new long data
frame, so that all plots can be drawn simultaneously with the last line. Note the use
of the scales option to ensure individual scaling for each variable and the nrow
option to put all plots in one row. The formatting options are necessary to clear the
plots of irrelevant labelling.
The differences in how Figures 1.8 and 9.7 were drawn and what they show are
worth noting. Traditional base graphics were used for Figure 1.8 and the boxplots
for all the variables are shown in one display. This required that the variables were
transformed to a common scale first. Graphics from ggplot2 were used for Figure 9.7.
By pretending that all the variables were actually one long variable composed of
different subsets it was possible to draw individual plots for each of them with a
single command, using their actual scales.
Boxplots generally work well, although they tend to imply too many outliers for
large datasets such as diamonds. The reason is clear: With large datasets there are
sufficient numbers of points at intermediate outlying levels for them to be regarded
as part of the population, even though boxplots plot them as outliers. At any rate, you
can argue that it is better to identify too many potential outliers as too few. You are
probably going to review the outliers in their order of ‘outlyingness’ anyway, and so
the more extreme ones will be dealt with first.
Apart from obvious errors (e.g., digits have been transcribed, so that a tempera-
ture of 19◦ has been written as 91◦ ), you need background knowledge to judge what
kind of case an indicated outlier might be. In most applications there is information
on many other variables and this can be used to help decide. Scatterplots can be used
for looking at pairs of variables (as in Figure 9.8), while parallel coordinate plots are
good for looking at more. Perhaps the case is an outlier on several variables, either
in a consistent way (the District of Columbia in the USCrimes dataset in Figure 9.6)
or in an inconsistent way (sometimes big, sometimes small). The latter may suggest
a data shift error, where values have been entered in the wrong columns for a case.
When data have skew distributions, it is often a good idea to transform them,
perhaps to a logarithmic scale. Points which were high outliers on the original scale
Graphics and Data Quality: How Good Are the Data? 187

library(tidyr)
diam1 <- diamonds %>% select(carat, depth:z) %>%
gather(dX, dV, carat:z)
ggplot(diam1, aes("dX", dV)) + geom_boxplot() +
facet_wrap(~dX, scales = "free_y", nrow=1) +
xlab("") + ylab("") + scale_x_discrete(breaks=NULL)

carat depth table price x y z

80 60
5 ! !
!
! !
!
!
!
! ! !
30
!
!
! !
!
!
! !
!
90
!
!
! !
! !
!
! !
!
! !
!
!
9
! !
!
!
! !
!
!
4 15000
! ! !
!
! !
! !
!
!
!
70
!
! !
!
!
! !
!
!
80
!
!
! !
!
! ! !
!
40
! ! !
!
!
! !
! !
!
! !
! !
!
20
!
! !
!
!
! !
! ! !
!
!
3
!
! !
6
! !
!
!
!
! 70 !
! 10000 !

60
! !
! !
!
! !
!
! !
! !
!
! !
! !
!
!
2 !
! !
!
!
20
!
!
60
!
10
!
!
!
!
3
!
!
5000
!
!
! !
!
50
!
!
!
1
!
!
!
50 ! !
!
!

! !
!
!
!
0 0 ! 0 ! 0 !
0

FIGURE 9.7: Individual boxplots of seven of the diamonds variables. Each of the
variables has outliers, which is hardly surprising in a dataset of over 50 thousand
cases. The last three variables, x, y, and z have cases with value 0, which must be
wrong, and there are several high outliers, which must be wrong too. Both the carat
and price variables are skewed to the right.

will then no longer necessarily be outliers on the new scale and some points that
were formerly regarded as non-outliers may now be low outliers. Again, background
knowledge is essential to judge whether a transformation makes sense.

Multivariate outliers
Most outliers are outliers on at least one univariate dimension, sometimes several.
The more variables there are, the more likely it is that cases are outliers on some
dimension or other. This implies that the more variables you have, the more stringent
the rules should be for deciding which cases might be outliers. In practice you usually
start with the most extreme outliers and work inwards.
Multivariate measures depend on robust estimation of the parameters of the data.
For small datasets the efficiency of calculating these robust statistics is good and
there can only be a few outliers anyway. At least, you would think so. The famous
stackloss dataset, a favourite amongst robust statisticians, has only 21 points and
probably all of them have been declared an outlier at one time or another in some
article or other [Dodge, 1996].
188 Graphical Data Analysis with R

Large datasets are more complicated. It is much more likely that they are het-
erogeneous, the data quality is probably poorer, and there may well be a variety of
outliers and outlying groups of diverse kinds. Whether this has much influence on
an analysis is another matter. The diamonds dataset shows an example. The first plot
in Figure 9.8 is a scatterplot of y and z (actually width and depth). There are some
obvious errors (values of 0 and excessively high values). Restricting the ranges gives
the second plot, which excludes 23 of the almost 54, 000 points. Now some other
cases appear to be in error (the three low values on depth), while others look unusual
compared to the bulk of the data, bivariate outliers that are not outlying on either of
width or depth alone.

a2 <- ggplot(diamonds, aes(y, z)) + geom_point() +

xlab("width") + ylab("depth")
d2 <- filter(diamonds, y > 2 & y < 11 & z > 1 & z < 7)
b2 <- ggplot(d2, aes(y, z)) + geom_point() +
xlab("width") + ylab("depth")
grid.arrange(a2, b2, ncol=2)

! !
30 !
!
!! !!
6
! !
! !!
!!! !!
!
!! !!!
!!
!!
!
!!
!
!!!
! ! !
!
!!
!!
!
!!
! !
!
!!
! !
!!
!!!!!!
! !!
!!
!
!
!
!!
!
!
!!
!!
!
!!!! !
!!
!!
!
!!!
!!!
!!!
!
!
!!
!
!!
!
!
!
!
!
!!
!
!
!!!
!
!
!!
!!!
! !
!
!!
!
!
!
!!
!
!
!
!!
!
!
!!
!
!
!!
!
!
!!
!
!
!!
!
!
!
!
!!
!
!
!
!
!!
!
!
!
!
!
!!
!
!
!
!
!
!!
!
!!
!
!!!
!
! !!!!
!
!!!!
!
!
!!
!
!!
!
!!
!
!
!!!
!
!!
!
!!
!
!
!!
!! !
!
!
!
!
!!
!
!
!!
!
!
!!
!
!
!!
! !
!
!
!!!
!
!!
!
!!!!
!
!
!
!!
!
!!!
!!
!
!!
!
!!!!!!
20
! ! !! !
!!
!
!
!
!
!!
!
!
!
!
!!
!
!!
!
!
!!
!
!!
!!
!
!
!
!
!!
!
!!!
!
!!
!
!
!!
!
!!
! !
!!! !!
!!
!
!
!!
!
!!
!
!
!!!
!
!!!
!
!!
!!
!
!!
!
!!
!
!!!
!!!
!
!
!
!!
!
!
!
!!
!
!
!
!!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!!
!
!
!
!!
!
!!
!
!!!
!!!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!!
!!
!!
!!!
!
!!
!
!!
!
!!
!
!!
!
!!
!!
!!
!!
!
!
!!!
!!
!
!
!
!!
!
!!!
!
!
!
!!
!
!
!
!!
!!
!
!
!
!!
!
!
!
!!
!
!!!!
depth

depth

!!! ! !!!!!
!
!!
!
!
!
!
!!
!
!
!
!
!!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!
!!!
!!
!
!
!
!
!
!!
!
!
!
!
!!
!
!
!
!
!
!
!
!!
!
!
!
!
!!
!
! !! !!!
!
!!
!
!
!
!
!!
!
!!
!
!
!
!!
!!!
!
!
!
!
!
!!
!
!
!
!
!
!!
!!
!!!
!!
!
!
!
!
!!
!
!
!
!
!!
!
!
!
!!
!
!
!!
!
!
!
!!
!
!!
!
!!!!
!!!!!
!!
!
!!
!!
!
!!!
!
!!
!!
!
!
!!
!!
!
!!
!!
!
!! !!
4
!
!!!
!
!!
!!
!
!
!
!!
!
!
!!
!
!
!!
!
!!!
!
!
!!
!!
!!
!
!
!!
!
!!
!
!!
!!!
!
!!
!!
!
!!
!
! !
!
!!
!
!!
!!
!
!!
!
!
!
!
!!
!
!
!
!!
!
!
!
!!
!
!!
! ! !! !!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!!
!!
!!
!!
!
!
!!
!
!!
!!
!
!!
!!
!
!!
!
!!
!!
!
!
!!
!
!
!
!!
!!!
!!
!
!!
!!
!
!
!!
!!
!!
!
!!
!
! !
!
!!
!
!!
!!
!
!!
!
!!
!
!!
!!
!
!
!!
!!
!!
!
!
!
!
!!
!
!
!
!!
!!
!! !
!
!
!!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!!!
!!
! !
!!
!
!
!!
!
!
!!
!
!!
!
!!
!!
!!
!!
!!
!
!!
!
!!
!
!!
!!
!
!!
!!
!!
!!
!!
!
!!
!
!
!!
!! !
!!
!!
!
!
!
!!
!
!
!
!
!!
!
!
!
!
!
!
!!
!
!
!
!!
!
!
!
!! !
!
!
!!
!
!!
!
!
!!
!!
!
!
!!
!
!
!
!
!!
!
!
!
!
!
!!
!
!
!
!
!!
!!
!!
!!
!
!!
!
!
!
!
!!
!
!
!
!
!
!!
!
!
!
!
!
!
!!
!
!
!
!!
!
!
!
!!
!
!!
!!!
!!!
!
!
!!!
!
!!
!
!!
!!!
!
!
!!!
!!!
!!
!
!!
!!
!
!
!!
!!
!
!!
!!!
!
!
!!
!!
! !
!
! !
!!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!
!
!
!
!!
!!
!
!
!
!
!
!
!
!
!
!
!
!
!!
!!
!!
!!
!!!
!
!
!
!
!
!!
!!
!
!
!
!
!
!!
!!
!
!
!
!
!
!
!
!
!!
!!
!
!!
!
!
!
!!
!
!
!
!
!
!
!
!
!
!
!!
!!
!
!
!
!
!
!
!
!
!
!!!!!
!
!
!
!!!
!!
!
!
!!
!
!
!!!
!
!
!
!!
!!
!!!
!!
!
!
!
!!
!
!
!!
!
!!
!
!!
!!
!!
!
!
!!
!!
!
!
!!!
!
!
!
!
!
!!
!
!!
!!
!
!
!!
!
!!
!
!!
! !
! !
!!!
!
!
!
!!
!
!!
!
!
!
!
!
!
!!
!
!
!!
!!
!!
!!
!!
!
!!!
!!
!
!
!
!
!!
!
!!
!!
!!
!
!!
!!
!!
!!!
!
!
!!
!
!
!
!
!!
!
!!! !!
10 !
!!
!
!
!
!!
!
!!
!
!
!
!
!
!!
!
!
!
!
!!
!
!
!!
!!
!!
!
!
!!
!
!
!
!
!
!!
!
!
!
!
!!
!
!
!
!!
!
!!
!
!!
!
!
!!
!!
!!
!
!!
!
!!
!!
!
!!!!
!!
!!
!
!!
!
!!
!!
!
!!
!
!
!
!
!!
!
!!
!
!
!!
!
!
!
!!
!
!
!
!! !
!
! !
!
!
!!
!!
!
!
!
!!
!
!
!!
!!
!
!
!!
!!
!
!
!!
!!
!
!!
!!
!
!
!
!!
!!
!!
!
!
!!
!!
!!!
!
!
!
!!
!
!
!
!
!!
!
!!!
!
!
!!
!
!!
!
!!
!
!
!!
!
!
!
!
!!
!!
!!
!
!
!
!
!!
! !
!
!!
!
!
!!
!
!
!
!!
!!
!!
!
!!!
! !
!!
!
!
!
!
!
!!
!
!!
!
!
!!
!
!!
!
!
!!
!!
!!
!!
!!
!
!!
!
!
!
!!
!
!!
!
!!
!!
!!
!!
!
!!!
!
!
!
!
!
!
!!
!
!
!!
!
!
!
!!
!
!
!
!! !!
!! !
!
!
!
!!
!
!!
!
!!
!
!!!
!
!
!
!!
!! !
!
! !
!
!
! !
! !
2
!
!!
!
!
!
!
!
!!
!
!!!
!
!
!!
!
!
!
!
!
!
!!
!
!
!! ! !
!!
!
!
!
!
!!!
!
!
!
!
!
!!
!
!
!
!!!
!
!
!
!
!!
!
!
!
!
!
!
!!
!
!
!
!
!
!!
!
!
!
!
!
!!
!
!
!
!!
! !
!! !
!
0 ! ! !
!!! !

0 20 40 60 5 7 9
width width

FIGURE 9.8: Plots of y (width) and z (depth) from the diamonds dataset with and
without 23 outliers and rescaled accordingly. Without the extreme outliers the plot is
more informative and additional cases look suspicious.

The statistics of the dataset including and excluding the original 23 outliers barely
change, only the graphics are dramatically affected and that is because of the univari-
ate outliers. Note how Hadley Wickham’s dplyr package is used for calculating the
statistics.

diamS1 <- summarise(diamonds, meanx = mean(x), sdx = sd(x),

meany = mean(y), sdy = sd(y), corxy = cor(x,y))
diamS2 <- summarise(d2, meanx = mean(x), sdx = sd(x),
meany = mean(y), sdy = sd(y), corxy = cor(x,y))
Graphics and Data Quality: How Good Are the Data? 189

diamS1

# meanx sdx meany sdy corxy

# 1 5.731157 1.121761 5.734526 1.142135 0.9747015

diamS2

# meanx sdx meany sdy corxy

# 1 5.731605 1.119402 5.733428 1.111272 0.9986573

Figure 9.8 also illustrates how points can be outliers on two dimensions without
being outliers on the individual dimensions.
The decisions as to which points are outliers can be problematic and may depend
on an implicitly assumed model. In Figure 9.9, showing the olive oil dataset used in
two exercises in earlier chapters, you could regard points as outliers that are far from
the mass of the data, or you could regard points as outliers that do not fit the smooth
model well. Some points are outliers on both criteria.

data(olives, package="extracat")
ggplot(data=olives, aes(x=oleic, y=palmitic)) + geom_point() +
geom_density2d(bins=4, col="red") + geom_smooth()

!
!!
!
!
! !
! !
!!
!
!
!!
!! !
! ! !! !
1500
! ! !
! ! !! !
!! ! !
! ! !
! !!!! ! ! ! !
!! ! !!!!! !!! !!! ! ! !
!
! !! !
! !
!! !!!! !!! !!
! !!!
! !
!! !
!! ! ! ! !! !! !! !
! ! ! !! !! ! ! ! ! ! ! !!
! !
!!
!
!
!
!!!
!! ! !
! !!!! ! ! ! !! !
! !
! !
!
!!!! !!! !!! ! !!!!!!! ! !!! !
!
!!!!
!!! !
! !! ! ! ! !! ! !!
!
!!
!
! !! !
!
!
!
!! ! ! !
! ! ! ! ! !!
! !
!
!!
! ! !! !!!
!!
! ! ! ! !! ! !! ! !
!! ! !
palmitic

!!
! ! ! ! !! ! !
! ! ! ! !
! ! ! ! !
! ! !! ! ! ! ! !! !! ! ! ! !
!! !! ! ! ! !! ! !! ! !
! !! ! !! ! ! !! ! !!!!! !
! !!!! ! !!
! !!! ! ! ! !
!
! ! ! ! !
!!!!!!!! !!! ! !!
! ! ! ! !
!
!! ! ! ! ! ! !
!! !! !! ! !
!!! ! !!!! ! ! ! !! ! !
!! !!!
! !
!!!!! !!
!!!!!
! !! ! ! !!
! !
!!!!!
!
! ! ! !! !
!
!! ! !! ! !!
!!!! ! ! ! !
!!
!
!
!!
! !! ! ! ! !!!! ! !! ! !
!!
!!
!
! ! !
!
!!!
!!!! !! ! !
!!
!
! ! !!! !!! ! !
! ! ! ! ! !! !
1000 !
! !
! !!!!
!!! !
!
!!
!
!! !
! ! ! !
! ! !
!

6500 7000 7500 8000 8500

oleic

FIGURE 9.9: A scatterplot of two of the variables from the olives dataset with con-
tour lines from a two-dimensional density estimate and a loess smoother superim-
posed. Some points to the top left are outliers according to the density estimate, but
not according to the smoother.
190 Graphical Data Analysis with R

Higher dimensional outliers are tough to spot visually when they are not extremes
on lower dimensions, and it is an open question how often such cases occur in
real datasets. Suggested methods of finding them include using a robustified Maha-
lanobis distance and there have been several proposals how to do this in the literature
[Ben-Gal, 2005]. Graphical displays such as parallel coordinate plots and scatterplot
matrices can then be helpful for establishing why points have been identified as out-
liers.
A simple illustration is given in Figure 9.10, using the Boston dataset yet again,
where the boxplot shows that there are low outliers on the variable ptratio, the
pupil teacher ratio. Drawing a parallel coordinate plot of all the variables and colour-
ing the three cases sharing the lowest value gives the plot on the right.

data(Boston, package="MASS")
a <- ggplot(Boston, aes("var", ptratio)) + geom_boxplot() +
xlab("") + ylab("Pupil-teacher ratio") +
scale_x_discrete(breaks=NULL)
Boston <- within(Boston, pt1 <- ifelse(ptratio < 13, 1, 0))
oc <- order(Boston$pt1)
b <- ggparcoord(data = Boston[oc,], columns = c(1:14),
scale="uniminmax", groupColumn="pt1") +
theme(axis.title.x = element_blank(),
axis.title.y = element_blank())
grid.arrange(a, b, nrow=1, widths=c(1,4))

1.00

20.0
0.75
pt1
Pupil!teacher ratio

1.00

0.75
17.5
0.50
0.50

0.25

0.00
15.0 0.25

12.5 ! 0.00

crim zn indus chas nox rm age dis rad tax ptratio black lstat medv

FIGURE 9.10: A boxplot of ptratio from the Boston dataset and a parallel co-
ordinate plot of all the variables with the three cases having the lowest value for
ptratio coloured light blue. The three areas are very similar and are not outliers
on any of the other variables.

Categorical variables divide datasets into subsets and often cases can be outliers
within a subgroup but not on the whole dataset or the other way round. For a sin-
gle subsetting variable, the best graphical approach is to plot boxplots for each of
the categories. Figure 9.11 illustrates this for Sepal.Width in the iris dataset (cf.
§1.3), where the boxplot for the three species together suggests four outliers, while
Graphics and Data Quality: How Good Are the Data? 191

a <- ggplot(iris, aes("boxplot for all", Sepal.Width)) +

xlab("") + geom_boxplot() +
scale_x_discrete(breaks=NULL)
b <- ggplot(iris, aes(Species, Sepal.Width)) +
geom_boxplot() + xlab("")
grid.arrange(a, b, nrow=1, widths=c(1,2))

4.5 4.5
! !

!
!
4.0 4.0
!

3.5
Sepal.Width

3.5
Sepal.Width

3.0 3.0

2.5
2.5
!
!

2.0
2.0 !

setosa versicolor virginica

FIGURE 9.11: Boxplots of Sepal.Width for the iris dataset as a whole and for
the three species separately. Four possible outliers are identified in each plot, while
only one outlier is identified in both.

the boxplots for the three species separately also suggest four outliers, although only
one of the points is an outlier in both displays.

Categorical outliers
It is easy enough to understand what is meant by an outlier on any continuous scale,
but categorical outliers are rather different. The term could refer to cases which have
some rare value on a categorical scale (e.g., purple hair), but that is more a matter of
defining the categories appropriately for the analyses to be carried out.
The more interesting situation is where there are a number of categorical vari-
ables and certain combinations occur very infrequently or should not occur at all. In
his study of data cleaning, [Malik, 2010], Waqas Ahmed Malik used a dataset from
the Pakistani Labour Force Survey of 2003. He found a few cases of people who
were divorced or widowed and who were classified on another variable as being the
spouse of the head of household. Those cases were clearly in error on at least one of
the variables. In another analysis he found people with low education and employ-
ment status recorded as being corporate managers. In a dataset with around 140, 000
cases these kinds of issues are bound to arise. Fluctuation diagrams (cf. §7.4) are
useful visual tools for finding such cases.
192 Graphical Data Analysis with R

Dealing with outliers

Identifying cases as potential outliers, whether with graphical or with analytic meth-
ods, is relatively straightforward, what to do about the information is not. Outliers
need to be inspected and reviewed. Sometimes they may be interesting in their own
right (e.g., telephone companies have regarded unusual records as indicating fraud),
sometimes they may be recognisable special cases (e.g., Bob Beamon’s long jump at
the 1968 Olympic Games), and sometimes they may be just errors.
If the cases are obviously wrong, then those values may be discarded or, if possi-
ble, corrected. You can decide to discard whole cases or just the unacceptable values,
and background knowledge is essential for these kinds of decisions. If only individ-
ual values are discarded, they may be treated as missing and substitute values may
be imputed for them in one way or another. Imputation can be a tricky business and
the appropriate expertise is called for. Whenever data are discarded, it is a sensible
idea to record what has been done and why, to ensure that discussion of results is
complete. When two people analyse the same data, but handle outliers in different
ways, it can save a lot of time to have complete details of what has been done when
you compare results.
Some statistics are little affected by outliers, medians for instance, but graphics
are always affected, even by a single case, as default scales have to be drawn to
incorporate all the data. (Of course, this only applies to outliers which are outlying
on at least one dimension, cf. Figure 9.8. Solely high dimensional outliers do not
affect scaling.) One extreme value in a million can reduce the rest of the dataset to
a column in a histogram or to a heavily overdrawn point in a scatterplot. To see the
structure of the main part of the dataset you have to zoom in, effectively ignoring
the outliers. It would be helpful if graphics gave some clue when some points are
excluded like this; you can easily forget about outliers that are not visible.
Keeping outliers and discarding them are the two alternative extremes. In mod-
elling, robust methods attempt to reduce the effect of outliers by calculating a weight-
ing for each case. Outliers that are far away may have little or no weight attached to
them, whereas cases which are neither far away nor part of the main pattern of the
data may be assigned a larger weight, although still lower than cases in the middle
of the data. Each robust method will produce different weightings, and the choice of
parameters within a method will also have an effect. The jury is still out on how best
to proceed.

A possible strategy for outliers

‘Extreme’ outliers are likely to be errors or data from a quite different population
and should be discarded or replaced with imputed values. ‘Ordinary’ outliers may be
of some interest in their own right. Quite often they will affect statistical modelling
less than one might fear, although it is as well to check what influence they have
using a robust approach, if that is possible. Initially an exploration of the dataset
with graphical methods and some simple outlier rules (e.g., a robust Mahalanobis
distance) should suffice to get a feel for the data.
Graphics and Data Quality: How Good Are the Data? 193

1. Plot the one-dimensional distributions. Examine any ‘extreme’ outliers to see if

they are interesting in their own right, if they are obvious errors which can be
corrected, or if they are values which should perhaps be discarded.
2. For outliers which are ‘extreme’ on one dimension, examine their values on other
dimensions. This can help to decide what to do about them. Consider imputing
values for outlying values you intend to discard.
3. After dealing with the ‘extreme’ values, examine further potential one-
dimensional outliers, i.e., those cases which are not ‘extreme’, but are marked
as outliers by a boxplot or by some other rule. For these cases check the val-
ues on other dimensions using multivariate graphics, parallel coordinate plots, or
sploms. Consider discarding cases which are outlying on more than one dimen-
sion.
4. Cases which are outliers in higher dimensions, but not in lower ones, should be
considered next. They may be identified in sploms (bivariate outliers), or in par-
allel coordinate plots, but it is probably best to use a robust multivariate approach
and then inspect the possible outliers graphically to see why the analysis picked
them out.
5. Finally, consider whether outliers should be looked for in subsets of the complete
dataset. Drawing boxplots by grouping variables is a good way of doing this.
And, as always, any potential outliers found should be examined in conjunction
with the case values on other variables and in the context of the study aims.

9.4 Modelling and testing for data quality

1. Missing values
In principle some kind of binary pattern test could be carried out. In practice it
is not done, probably because either there are not many missings or they have a
readily interpretable pattern.
2. Univariate outliers
Many tests have been suggested and a helpful place to start is Barnett and Lewis
[Barnett and Lewis, 1994]. Some unvariate tests are implemented in the package
outliers.
3. Multivariate outliers
To use a criterion to identify a point as an outlier or to test for an outlier implies
some kind of model of ‘standard’ data. No test is fully satisfactory, so much
depends on the definition of ‘good’ data, on the possible numbers of outliers, and
on the context. Different models and procedures will lead to different results.
194 Graphical Data Analysis with R

4. Robust approaches
Modern approaches rely on robust estimation of the ‘good’ part of the data
and, as with all robust methods, there is a wide range of possibilities involv-
ing clever heuristics of one kind or another. Robust Mahalanobis distances are
key. Packages implementing the methods include robust base, mvoutlier, and
CerioliOutlierDetection.

Main points
1. Quality of data can be a major issue and there are many ways the quality of
a dataset may be poor. All manner of different approaches may be helpful in
identifying and solving the problems.
2. Missing value pattern plots are good for identifying possible structure in missings
(cf. Figures 9.2 and 9.3).
3. Outliers are difficult to define and pin down (§9.3). They may be compared to
models or densities (Figure 9.9) or by subgroups (Figure 9.11).

4. Individual extreme values are easy to spot, groups of outliers are more difficult
to determine. There can be many different reasons for outliers, sometimes they
are errors, sometimes they are important special cases.
5. Graphical displays are useful for finding univariate outliers (Figure 9.7) and bi-
variate ones (Figure 9.8).
6. Sploms and parallel coordinate plots can be helpful for studying potential outliers
(Figure 9.10), for instance ones identified by robust approaches.
Graphics and Data Quality: How Good Are the Data? 195

Exercises
1. Lung cancer
The dataset lung in the package survival has 228 patients and 10 variables. What
missing value patterns are there?

2. Academic Performance Index

The dataset apipop in the package survey comprises 6194 Californian schools.
There are 37 variables in all.
(a) Are the missing values missing at random or are there patterns?
(b) Does excluding the variables with a majority of missing values change the
picture much?
3. Diamonds
Discuss the various kinds of outlier identified in a scatterplot of the variables
carat and width from the diamonds dataset discussed in Figures 9.7 and 9.8.
4. Beamon’s longjump
The dataset MexLJ has the best longjumps by the 14 finalists in the 1968 Mexico
Olympics. Is Bob Beamon’s famous jump really so very different from those of
the other competitors? Draw an appropriate plot for investigating this.
5. Pearson heights
Pearsons’ height data for fathers and sons were considered in §3.3 and §5.4.
(a) Draw a scatterplot of the heights. Are there any cases you would regard as
outliers?
(b) Overlay your plot with a bivariate density estimate. Which cases do you
think would be regarded as outliers under this model?
6. Forbes2000
This dataset from the HSAUR2 package includes data for the sales, assets, prof-
its, and market value for each of 2000 companies in 2004. Are there many uni-
variate outliers? What about bivariate outliers?
7. Forbes2000 again
The three variables, sales, assets, and marketvalue, are all related to
some extent to the size of the company. If they are transformed to logs, their
distributions look far less skewed.
(a) How many cases are still possibly outliers?
(b) Given that some companies reported losses, the variable profits can
only be transformed to logs by excluding companies who failed to make
profits. Which additional companies might then be considered to be out-
liers?
196 Graphical Data Analysis with R

8. Wheat
Rothamstead is the longest running agricultural research station in the world and
is probably best known to statisticians because Fisher worked there from 1919
to 1933. In 1910 a study of wheat yields was carried out there and Figure 9.12
shows the grain and straw yields. Which of the points in the scatterplot might be
outliers? Why?

!
! !!
! !

5.0 !
!
! ! !
!
! ! !
! ! ! !!
! !
! !
! !!! !!! ! !
! !
! ! !! !
! ! !!
4.5 ! !
! !! ! ! ! !
! ! !! ! !
! ! !
!!! ! ! !! ! !
!
!! ! ! !!! !! ! ! ! !
! ! !
! ! !! ! !! ! !!! ! !
!!! ! ! ! !! !! !! ! !
! ! !
!! !!
! !
! !! ! ! ! !! !
! ! ! !!! ! !!
!! ! !!! !!! !
!! ! !
!
! ! ! ! !!! !!
! ! !!!!!! ! !! !
! ! ! !! !! ! ! !! !! ! ! !
! ! ! ! ! !!! ! ! ! !
4.0
grain

!
! !! ! !!!
!
!! !! !!! ! !!! !! !! !
! !! ! ! !!!
! !! !! !
! !! !!! ! ! ! !! !
! ! !!
! !! !! ! !
! ! ! ! !!! ! ! !
! !! !! ! ! !!!! ! ! ! !
! !! !
!! !! ! !! ! ! ! !!
! !! ! !! !! !! !! !! !
!
! !! ! ! !! ! ! !
! !!!!! ! !!! ! !
! !! ! ! ! ! ! ! ! !! !
! !
3.5
! !! ! ! !
! !! ! ! !
! !! !!! ! !
! ! ! !!!! ! !! ! ! !
! !!
! !! !! !! !
! !! ! !
!
! !!! !
!! ! !
!
! ! ! ! !
!! !! !
!! !
! ! ! !! !
3.0
!
! !
!
!
! !
!
!

4 5 6 7 8 9
straw

FIGURE 9.12: A scatterplot of the yields of grain and straw of 500 plots in the
mercer.wheat.uniformity dataset from the package agridat.

9. USCrimes 2010
Create a subset of the USCrimes dataset in the TeachingDemos package contain-
ing just the data for 2010. As the District of Columbia is an extreme outlier on
several of the rates, remove it from the dataset and carry out a graphical outlier
analysis.
(a) How many univariate and bivariate outliers do you find?
(b) Are they the same as the ones found in the dataset with the District of
Columbia included?
10. Boston housing
(a) Compare a boxplot and a histogram of the tax variable. Which is more
useful for investigating outliers and how many outliers are there?
(b) Draw a parallel coordinate plot of all the variables and colour the cases
with a high value for tax. Are these cases different from the rest on other
variables?
11. Intermission
The Mona Lisa by Leonardo da Vinci hangs in the Louvre museum in Paris. Is
the Mona Lisa so different from other portraits of women?
10
Comparisons, Comparisons, Comparisons

Shall I compare thee to a summer’s day?

William Shakespeare (Sonnet 18)

Summary
Chapter 10 covers making comparisons using graphics, complementing more formal
statistical methods of comparison.

10.1 Introduction
Many scientific studies are carried out in a two-group design using a treated group
and a control group. The groups are then commonly compared on the basis of the
difference between their two means, using a two-sample t-test, assuming that the
samples are drawn from normal distributions with equal variances. There may be
other interesting features in the data, which cannot be formally compared because
the sample sizes are too small. This should not stop us from looking at the data, and
perhaps plotting the data should be compulsory if a t-test is carried out, just to remind
people of the assumptions they are implicitly making and to make them think about
what features they might be missing.
Many different features might be seen by looking at the data underlying t-tests.
There may be evidence of differing variability, there may be outliers, either in relation
to the whole dataset or in relation to one of the groups, there may be clustering
in the data, there may be skewness, there might be favoured values. A t-test tells
you nothing about any of these features and when the sample sizes are small, the
situation the test was originally developed for by Gosset, there is too little data to
be sure whether such features exist. As sample sizes get larger, it makes sense to
look at more than just the difference between the means. Graphics and statistics are
complementary approaches.

197
198 Graphical Data Analysis with R

The dataset bank in the package gclus concerning forged and genuine Swiss
banknotes was introduced in §5.6. Carrying out t-tests of the differences in means for
the various banknote measurements comparing the genuine and forged notes gives
a range of significances. Tests on the variables Diagonal and Right both give
p-values of < 2.2 × 10−16 . (This p-value limit may vary from machine to machine.)
The left-hand side of Figure 10.1 shows histograms of the Diagonal values
for the two groups of notes and the right hand side shows histograms for the variable
Right. The groups barely overlap for the Diagonal measurements, although there
is one outlier amongst the genuine notes. On the other hand, the distributions of the
Right measurements overlap a lot and it is mildly surprising that the means are so
significantly different.

data(bank, package="gclus")
bank <- within(bank,
st <- ifelse(Status==0, "genuine", "forgery"))
c1 <- ggplot(bank, aes(x=Diagonal)) +
geom_histogram(binwidth=0.2) + facet_grid(st~.)
c2 <- ggplot(bank, aes(x=Right)) +
geom_histogram(binwidth=0.1) + facet_grid(st~.)
grid.arrange(c1, c2, ncol=2)

20 15

15
10
forgery

forgery

10

5
5

0 0
count

count

20 15

15
10
genuine

genuine

10

5
5

0 0
138 139 140 141 142 129.0 129.5 130.0 130.5 131.0
Diagonal Right

FIGURE 10.1: Histograms of the Diagonal measurements on the left and the
Right measurements on the right for the forgery and genuine groups in the Swiss
banknote dataset. t-tests confirm that the differences in means between the two
groups are highly significant. The graphics show that the distribution patterns dif-
fer.
Comparisons, Comparisons, Comparisons 199

10.2 Making comparisons

“At the heart of quantitative reasoning is a single question: Compared to what?”
[Tufte, 1990]. Numbers are of little use on their own, they have to be compared. In
practice you are always making comparisons when you look at graphics, sometimes
with explicit formal hypotheses, sometimes with implicit expectations.
If you are surprised by what you see, it is doubtless because you had antici-
pated seeing a different picture. For instance, a scatterplot with a completely random
pattern can be very informative, if you believed that the data would be strongly cor-
related. An absence of information can also be information. In Conan Doyle’s story
“Silver Blaze”, it was the fact that the dog did not bark in the night that Sherlock
Holmes found curious. It is a good idea to think about what a graphic might look like
before you draw it. We are all experts at explaining what we see—once we have the
display in front of us.
Sometimes you compare new data to accepted standards, sometimes you compare
new data with old data, mostly you compare data from different groups within the
same dataset. Which comparisons you make, and why, can be crucial to the success
of an analysis and there are often very many comparisons that might be considered.
In a study of fuel use of cars you could compare this year with last year or earlier
years, you could compare use in town traffic with motorway use, you could compare
cars of different sizes and of different manufacturers.
There is also the issue of how you make the comparisons. Do you compare values
or, if suitably paired data are available, study differences? And if you study differ-
ences, do you consider absolute differences or relative differences?
Groups may also be compared with descriptive statistics like medians, maxima,
other quantiles, ranked values, or measures of variability. Medians might be used
in comparing incomes, minima in comparing lap times, quantiles or ranked values
in comparing top groups, and so on. Often the aim is to detect differences between
groups, although sometimes it is to see if groups can be taken to be sufficiently sim-
ilar to one another. For instance, an analysis of variance requires that the variability
in each group should be approximately the same.
Of course, graphics are only part of the story; statistical comparisons have to
be made. Usually means are compared, and the central limit theorem is often used,
rightly or wrongly, as an all-embracing justification for that kind of testing. Alterna-
tively non-parametric tests may be a possibility. Graphically there is more flexibility
and it is possible to compare whole distributions, although there is less objectivity.
One person’s “it is obvious from the graphic that. . . ” may be another person’s “that
could have arisen by chance and should be checked”.
200 Graphical Data Analysis with R

Types of comparison
With categorical data such as the Titanic dataset, discussed in Chapter 7, the main
statistics for comparisons are rates and percentages. You might want to compare:
1. male and female survival rates;
2. the survival rate of the passengers compared to that of the crew;
3. survival rates between the three passenger classes;

4. survival rates by sex within class or by class within sex;

5. the survival rates with those of other sinkings (there has been research comparing
the Titanic and Lusitania disasters [Frey et al., 2011]).
With all these possibilities it is important to choose appropriate comparisons. Given
the survival rate of male adults in the third class, you could compare it with the rate
for female adults in the third class or with the rate for male adults in the other classes.
If there had been more male children on board you could further compare the male
adult survival rate in the third class with the male child survival rate in the third class.

Comparisons may be
specific such as the male adult survival rate in the third class compared to the female
adult survival rate in the third class;
general such as comparing the male adult survival rate in the third class to the
overall survival rate (which includes that subgroup) or to the survival rate for all
others;
at different levels comparing males and females, comparing male adults and fe-
male adults, comparing male adults and female adults in the third class.
Each comparison requires a different model with different standard errors and the
interpretation of the results can be tricky given the large number of alternatives avail-
able for what you can compare.
In this chapter the emphasis is on using graphics to make comparisons. With
formal statistical comparisons you have an explicit statement of what is to be com-
pared. With graphical comparisons you decide what is important depending on what
you see and on what your expectations were. In effect you may be unconsciously
making many implicit comparisons, deciding what is worth looking at in more detail
and what not. People may have different opinions of what they can see in a plot,
although hopefully they will agree on the major features.
Comparisons, Comparisons, Comparisons 201

Comparing like with like

Comparisons should be fair, and that requires paying careful attention to several is-
sues. Any differences identified should be due to the factor you want to investigate.
Comparable populations Samples provide information on the populations they
are drawn from. Mortality and morbidity statistics are often reported as ‘age-
adjusted’ to try to make them comparable across populations with different age
structures.
Comparable variables In surveys, questions on the same topic may be framed in
quite different ways. In business, markets may be defined differently by different
companies, so that their sales figures are difficult to compare.
Comparable sources Data from different sources may be collected in different
ways according to different rules. Opinion polls from different firms can dif-
fer systematically. Drink-driving laws and how they are implemented may vary
widely between countries.

Comparable groups The gold standard in clinical research is the randomised con-
trol trial. Participants are allocated to treatments at random and, where possi-
ble, neither they nor the researchers know which treatment they receive (double-
blind). This should go some way to ensuring that the groups are similar in all but
the treatment they receive.
Comparable conditions If the effects of a particular factor are to be compared, then
we need to ensure that the effects of all other possible influences are as far as
possible eliminated. In the traditional discussions of the Hawthorne experiment
it is concluded that the improvements in performance were probably due to the
participants’ being observed, not a factor the experimenters had originally taken
into account.

Comparable measurements School marks in the United Kingdom are usually on a

scale of 0 to 100. In Germany schoolchildren get grades from 6 (worst) to 1.

Standardising data Sometimes data are standardised to allow comparisons. In the

decathlon event athletes’ performances in each of ten disciplines are converted
to point scores according to internationally agreed formulae, so that high jump
performances can be compared with shotput performances, with sprint times and
so on. Financial data collected over time are usually adjusted for inflation.
202 Graphical Data Analysis with R

10.3 Making visual comparisons

Comparing to a standard
Michelson’s measurements of the speed of light (used in Exercise 4 in Chapter 1) are
shown in Figure 10.2 with a red dotted line added, showing the current estimate of
the speed of light adjusted for travelling through air. Although the range of values
obtained by Michelson contains the ‘true’ value, a 95% confidence interval based on
these data does not. Using

data(Michelson, package="HistData")
tc <- t.test(Michelson, mu=734.5)

gives a confidence interval of 836.72 to 868.08.

ggplot(Michelson, aes(x=velocity)) + geom_bar(binwidth = 25) +

geom_vline(xintercept = 734.5, colour="red",
linetype = "longdash") +
xlab("Speed of light in kms/sec (less 299,000)")

15

10
count

600 700 800 900 1000 1100

Speed of light in kms/sec (less 299,000)

FIGURE 10.2: Michelson’s data for the speed of light from 1879. Most of his ob-
servations were higher than the relevant currently accepted value, which would be
734.5 on this scale (marked in red).

Comparing new data with old data

There are at least two datasets with petrol consumption data for cars in R, mtcars
with information on 32 models from 1973-4 and Cars93 from the MASS package
with information on 93 car models on sale in the USA in 1993 (also used in §5.4).
Figure 10.3 compares a histogram of the miles per gallon reported for the cars in the
earlier dataset with a histogram for the miles per gallon in city driving reported for
Comparisons, Comparisons, Comparisons 203

the cars in the later dataset. The horizontal scales have been chosen to be comparable
and the histograms have been drawn one above the other rather than side by side for
easier comparison of the distributions. Even though city driving is more demanding
than driving overall, the cars in the later dataset perform a little better.

data(Cars93, package="MASS")
c1 <- ggplot(mtcars, aes(mpg)) + geom_bar(fill="blue") +
xlim(10,50) + xlab("mpg for 32 cars from 1973-4")
c2 <- ggplot(Cars93, aes(MPG.city)) +
geom_bar(fill="red") + xlim(10,50) +
xlab("mpg in city driving for 93 cars from 1993")
grid.arrange(c1, c2, nrow=2)

4
count

0
10 20 30 40 50
mpg for 32 cars from 1973!4

10
count

0
10 20 30 40 50
mpg in city driving for 93 cars from 1993

FIGURE 10.3: Comparing miles per gallon distributions for cars from 1973-4 and
1993. The later cars appear to have a slightly better performance and it should be
taken into account that the mpg figures for them are only for city driving.

A comparison of the means with a t-test

tf <- t.test(Cars93$MPG.city, mtcars$mpg)

shows that the means are not significantly different, with a p-value of 0.067 and a
95% confidence interval of -0.16 to 4.71.
204 Graphical Data Analysis with R

Comparing subgroups
Figure 10.1 showed comparisons of two of the variables from the Swiss Banknote
dataset. Histograms are effective for comparing two groups and can reveal a lot of
data details. They are less effective when there are several groups.
Boxplots work well for more groups. Figure 10.4 displays the palmitic vari-
able for the nine areas in the olives dataset, which was also used in §9.3. Having the
area labels directly beneath their corresponding boxplots helps a lot, as does colour-
ing by region. For presentation purposes it could be worth ordering the areas by their
medians or ordering them within the three regions by their medians.

ggplot(olives, aes(Area, palmitic, colour=Region)) +

geom_boxplot() + theme(legend.position="bottom")

1800
!
!
!
!
!
!

1500
!
!

! ! !
palmitic

!
1200

900

600 !

Calabria Coast!Sardinia
East!Liguria
Inland!Sardinia
North!Apulia Sicily South!Apulia Umbria West!Liguria
Area

Region North Sardinia South

FIGURE 10.4: Boxplots of the palmitic data by area for the olives dataset. Dif-
ferences in levels and variability as well as some outliers can be observed. There
is much more variability in the Southern region than in the other two regions, both
between and within areas.

Density estimates or possibly distribution functions can be used too, though for
a smaller number of groups. Figure 10.5 shows boxplots, density estimates, and em-
pirical distribution functions of the palmitic variable for the three regions in the
olives dataset. Each display picks out the main feature (that the values for the South-
ern region are generally higher than those for the other two regions). The boxplots
and density estimates are more informative than the empirical distribution functions.
If you amend the code above using Area instead of Region you can see how the
comparisons of the nine areas would look using these plots. Although some features
Comparisons, Comparisons, Comparisons 205

stand out, there are too many groups for an unambiguous allocation of colours, so
that it is not always easy to tell which feature belongs to which area.

o1 <- ggplot(olives, aes(Region, palmitic, colour = Region)) +

geom_boxplot() + theme(legend.position = "none")
o2 <- ggplot(olives, aes(x=palmitic, colour = Region)) +
geom_density() + ylab("density") +
theme(legend.position = "none")
o3 <- ggplot(olives, aes(x=palmitic, colour = Region)) +
stat_ecdf() + ylab("cdf") +
theme(legend.position = "bottom")
grid.arrange(o1, arrangeGrob(o2, o3, nrow=2),
ncol=2, widths=c(1, 2))

1800 0.0100
!
!
!
!

0.0075
!
!
density

0.0050
1500

!
0.0025
!

!
! 0.0000
!
600 900 1200 1500 1800
palmitic

1200 palmitic

1.00
!

!
! 0.75
! !
!
!
!
! !
!
cdf

900 0.50
!
!
!
!

0.25

0.00
500 1000 1500
palmitic
600 !

North Sardinia South Region North Sardinia South

Region

FIGURE 10.5: Three alternative displays of the distributions of the palmitic vari-
able for the three regions in the olives dataset. The boxplots on the left emphasise the
outliers and show that the olive oils from the South have higher values than the oils
from the other two regions, while the values for the Sardinian region are very close
together. The density estimates in the upper right plot show much the same infor-
mation, although with less emphasis on the outliers, and reveal additionally that the
distribution for the Southern region is skewed towards lower values. The empirical
distribution functions in the lower right plot do not show much more than that the
Southern region has higher values.
206 Graphical Data Analysis with R

Comparing time series (Playfair’s import/export data)

William Playfair drew many impressive displays, including several showing Eng-
land’s imports and exports with other countries. The data for England’s trade with
the East Indies between 1700 and 1780 have been estimated [Bissantz, 2009] from
the graphic in the first edition of Playfair’s “Commercial and Political Atlas”, pub-
lished in 1786. The corresponding graphic in the third edition [Playfair, 2005] goes
up to 1800.
There are several ways imports and exports could be compared over the years.
Playfair plotted both series in the same display and coloured the area between them
to give some idea of the balance of trade. The top plot of Figure 10.6 is a redrawing
of his plot.
Cleveland suggested plotting the difference between imports and exports
[Cleveland, 1994] and this is shown in the middle plot of Figure 10.6.
You might also consider plotting the relative difference rather than the absolute
difference and this has been done in the final plot of Figure 10.6, where the rel-
ative difference has been calculated by dividing the difference by the average of
Imports and Exports. Depending on the goals of the analysis, either Imports
or Exports alone could have been taken as the base.
Modern economic analysts would probably consider adjusting for inflation.
There are estimates of eighteenth century inflation available by year (e.g.,
[James, 2014]), but the cumulative effects are on a much smaller scale than the dif-
ferences observed here.
This historical context of these data should not be ignored. The period includes,
amongst other events, the War of the Spanish Succession (1701-14), the South Sea
Bubble (1720), the Irish Famine due to the “Great Frost” (1740-41), the Seven Years’
War (1756-63), and the American Revolutionary War (1775-83).

data(EastIndiesTrade,package="GDAdata")
c1 <- ggplot(EastIndiesTrade, aes(x=Year, y=Exports)) +
ylim(0,2000) + geom_line(colour="red", size=2) +
geom_line(aes(x=Year, y=Imports),
colour="yellow", size=2) +
geom_ribbon(aes(ymin=Exports, ymax=Imports),
fill="pink",alpha=0.5) +
ylab("Exports(red) and Imports(yellow)")
c2 <- ggplot(EastIndiesTrade, aes(x=Year,
y=Exports-Imports)) + geom_line(colour="green")
c3 <- ggplot(EastIndiesTrade, aes(x=Year,
y=(Exports-Imports)/((Exports + Imports)/2))) +
geom_line(colour="blue")
grid.arrange(c1, c2, c3, nrow=3)
Comparisons, Comparisons, Comparisons 207

2000
Exports(red) and Imports(yellow)

1500

1000

500

0
1700 1720 1740 1760 1780
Year

!200
Exports ! Imports

!400

!600

!800

1700 1720 1740 1760 1780

Year
(Exports ! Imports)/((Exports + Imports)/2)

0.0

!0.5

!1.0

!1.5

1700 1720 1740 1760 1780

Year

FIGURE 10.6: England’s trade with the East Indies in the eighteenth century. The
top plot is a redrawn version of Playfair’s plot showing that imports were always
higher than exports. The middle plot shows the balance of trade and highlights a dip
in the 1760’s, which is hard to see in the top plot. The lowest plot shows the relative
balance of trade and suggests that recent deficits are lower, compared to the deficits
in the twenty years around 1720.
208 Graphical Data Analysis with R

10.4 Comparing group effects graphically

The famous barley dataset was considered briefly in §8.5. The aim of the study was
to compare ten varieties of barley by looking at yields in two successive years at
each of six testing station sites. The yields for the two years are a little different, with
those for 1931 looking higher than those for 1932, as can be seen in Figure10.7.

data(barley, package="lattice")
ggplot(barley, aes(yield)) + geom_histogram(binwidth=5) +
ylab("") + facet_wrap(~year, ncol=1)

1932

15

10

0
1931

15

10

0
20 40 60
yield

FIGURE 10.7: Histograms of yields for the barley dataset in the lattice package for
the years 1931 and 1932. The values for the earlier year look higher on average with
the distribution appearing to be shifted to the right compared to the following year.

It is the differences in yield by variety that are mainly of interest and drawing
ten histograms would not be practical, even if there was enough data to justify it.
Figure 10.8 shows parallel dotplots of the 12 values for the ten varieties in the up-
per plot and confidence intervals for the variety means in the lower plot. The plots
suggest that there is little difference between the varieties. Note the use of the %>%
operator from the magrittr package to build a sequence of operations in a readable
order. magrittr is imported through loading the dplyr package.
Comparisons, Comparisons, Comparisons 209

c1 <- ggplot(barley, aes(x=variety, y=yield)) +

geom_point() + ylim(10,70)
barl1 <- barley %>% group_by(variety) %>%
summarise(N = n(), mean = mean(yield),
sd = sd(yield), se = sd/sqrt(N))
lims <- aes(ymax = mean + 2*se, ymin=mean - 2*se)
p1 <- ggplot(barl1, aes(x=variety, y=mean)) +
geom_point() + ylim(10,70) +
geom_errorbar(lims, width=0.2)
grid.arrange(c1, p1)

!
!

60 !
! !

! !
! ! ! !
! ! ! ! !
!
!
! !
! ! ! !
yield

! ! !
! !
40 !
!
! !
! !
!
! !
! !
! ! ! !
! ! ! !
! ! !
!
! ! !
! ! ! !
!
! !
! ! !
! ! ! ! !
! !
! ! !
! ! ! !
! ! ! !
! !
! !
! ! !
! !
! ! ! !
! !
! !
20 !
! ! !
! !

!
!
!

Svansota No. 462 Manchuria No. 475 Velvet Peatland Glabron No. 457 Wisconsin No. 38 Trebi
variety

60
mean

40 ! !

! !
!
! !
! !
!

20

Svansota No. 462 Manchuria No. 475 Velvet Peatland Glabron No. 457 Wisconsin No. 38 Trebi
variety

FIGURE 10.8: Yields for the barley dataset by variety. The individual data points for
each variety are shown in the upper plot and confidence intervals for the means in
the lower one (the intervals are ±2SE). The vertical axis scales have been chosen to
include all the data and to be the same for both plots. The varieties appear to produce
similar yields.
210 Graphical Data Analysis with R

The displays in Figure 10.8 treat all values equivalently, ignoring both the sites
and the years. Figure 10.9 shows confidence intervals for yields by year for each site.
It is obvious that there are clear differences between the sites and between the years
within the sites. The odd pattern for the Morris site that Cleveland remarked on, with
the data possibly reversed for the two years, can also be seen.
The mutate function is used to put the two years in ascending chronological
order, which looks more natural. It appears that the levels for the three variables,
variety, year, and site, in the dataset in lattice have been ordered by increas-
ing yields. You can confirm this by drawing boxplots of yield by each of the vari-
ables.

barl2 <- barley %>%

mutate(Year = factor(year,
levels = c("1931", "1932"))) %>%
group_by(site, Year) %>%
summarise(N = n(), mean = mean(yield),
sd = sd(yield), se = sd/sqrt(N))
lims <- aes(ymax = mean + 2*se, ymin=mean - 2*se)
ggplot(barl2, aes(colour=Year, x=site, y=mean)) +
geom_point() + geom_errorbar(lims, width=0.2) +
ylim(10,70) + theme(legend.position = ’bottom’)

60

!
mean

! !
40
!

!
!
! ! !

20
!

Grand Rapids Duluth University Farm Morris Crookston Waseca

site

Year ! 1931 ! 1932

FIGURE 10.9: Confidence intervals for mean barley yields by year for each site. The
data for 1931 are lower than for 1932 at all sites barring Morris.
Comparisons, Comparisons, Comparisons 211

To identify differences between the varieties, both site and year have to be taken
into account to make an effective comparison. A linear model shows that variety
is indeed significant and Figure 10.10 shows the interval estimates for the variety
coefficients for this model. The default in R is to set the first coefficient to 0, so the
plot shows that significance is primarily due to the last two varieties (Wisconsin No.
38 and Trebi) being higher than the first (Svansota).

m1 <- lm(yield~site+year+variety, data=barley)

library(coefplot)
coefplot(m1, predictors="variety", lwdOuter=1) + ggtitle("") +
ylab("") + xlab("yield difference from Svansota")

varietyTrebi !

varietyWisconsin No. 38 !

varietyNo. 457 !

varietyGlabron !

varietyPeatland !

varietyVelvet !

varietyNo. 475 !

varietyManchuria !

varietyNo. 462 !

!5 0 5 10 15
yield difference from Svansota

FIGURE 10.10: Interval estimates (±2s.e.) for the coefficients from a linear model of
yield for the barley dataset. The top two varieties have coefficients that are obviously
higher than 0, the value for the Svansota variety (not shown). The other varieties have
positive coefficients, although not clearly higher than 0.
212 Graphical Data Analysis with R

10.5 Comparing rates visually

As Chapter 7 shows, for example in Figure 7.1, shading bars of the same height
but different widths according to subgroup proportions is a good way of comparing
rates and the biggest groups get the most attention, being the widest bars. However,
there is no information on the possible significance of any differences between the
proportions. A statistical alternative would be to plot confidence intervals for the
proportions, treating them as estimates of unknown parameters.
Besides the advantage of offering some statistical guidance, there are two dis-
advantages, one major and one minor that have to be borne in mind. The minor
disadvantage is that the larger groups will have smaller intervals, thus attracting less
attention. The major disadvantage is that individual confidence intervals will not nec-
essarily help assess the significance of the particular comparisons you might wish to
make.
Figure 10.11 shows confidence intervals for male survival rates by class for the
Titanic, assuming independent binomial distributions for each class. It is an alterna-
tive display for the right-hand plot in Figure 7.1, but only for the males. The dif-
ferences between the survival rates for the male first-class passengers and the other
male groups are obviously significant; the significances of the other differences are
unclear.
The code for Figure 10.11 constructs the intervals by hand, so to speak. A new
dataset for just the males is defined. Then the necessary marginal totals are calculated
for finding the rates and from them the limits are calculated.

0.5

0.4
Male survival rate

0.3

!
0.2
!

0.1

0.0

1st 2nd 3rd Crew

Class

FIGURE 10.11: Approximate confidence intervals for male survival rates by class
for the Titanic dataset. The first-class males had significantly better survival chances
than the other males.
Comparisons, Comparisons, Comparisons 213

t1 <- data.frame(Titanic)
t1m <- t1[t1$Sex=="Male",]
xt1 <- xtabs(Freq ~ Class, data=t1m[t1m$Survived=="Yes",])
xt2 <- xtabs(Freq ~ Class, data=t1m)
surv <- xt1/xt2
survS <- (surv*(1-surv)/xt2)^0.5
su <- data.frame(cbind(surv, survS))
su$Class <- rownames(su)
lims <- aes(ymax = surv + 2*survS, ymin=surv - 2*survS)
ggplot(su, aes(x=Class, y=surv)) + geom_point() +
geom_errorbar(lims, width=0.1, colour="blue") +
ylim(0,0.5) + ylab("Male survival rate")

Model-based confidence intervals for the rates could be calculated using logistic
regression models of the males’ data and the results would be roughly the same.
The dataset would have to be rearranged first and after the model fitting the results
converted to probability scales. Other logistic models could be fit as well, modelling
the females as well as the males. A model with the interaction of Class and Sex
then gives the same results, while a model with only the additive terms leads to
different results. With displays of interval estimates you always have to know how
the estimates were calculated.

library(reshape2)
t1c <- dcast(t1m, Class+Sex~Survived, sum)
m1c <- glm(cbind(Yes, No) ~ Class,
family = binomial, data=t1c)
p1c <- predict(m1c, se.fit=TRUE)
lowc <- p1c$fit - 2*p1c$se.fit
highc <- p1c$fit + 2*p1c$se.fit
estp <- 1/(1+exp(-p1c$fit))
lowp <- 1/(1+exp(-lowc))
highp <- 1/(1+exp(-highc))

Alternatively, you could plot coefficient interval estimates from a model using co-
efplot, for instance with the following code. This uses a model form which includes
no intercept, so that all the coefficients of interest are included.

t1e <- dcast(t1, Class+Sex~Survived, sum)

m1f <- glm(cbind(Yes, No) ~ 0+Class*Sex,
family = binomial, data=t1e)
coefplot(m1f) + xlab("") + ggtitle("")

Rates can possibly depend on many, many factors. How baseball batting aver-
ages may depend on situational factors is discussed in [Albert, 1994], where eight
factors including home/away, day/night game, groundball/flyball pitcher, scoring po-
sition/none are considered, and even more are suggested. Specific hypotheses require
the appropriate underlying models for drawing intervals.
214 Graphical Data Analysis with R

10.6 Graphics for comparing many subsets

Trellis displays (cf. §8.5) are an effective means of looking at displays of many sub-
sets at the same time. Big differences for particular subsets stand out; smaller dif-
ferences are more difficult to spot. Rearranging the order of the panels can help.
Figure 10.12 is a trellis display for the olive oil dataset showing scatterplots of the
variables palmitic and palmitoleic for the nine areas.

data(olives, package="extracat")
ggplot(olives, aes(palmitic, palmitoleic)) +
geom_point() + facet_wrap(~Area)

Calabria Coast!Sardinia East!Liguria

200
!
!
! ! !!!!
!
!!! !!!
!!
!! ! !
!!
!
!
!!!
!!
! !!!!
!
!! !!!! !! !!
100 !! ! !! !
!!!!!!
!
! ! !
!
!!
!
!
!!!
!!
! !!!!!! !!
!
! !!! !! !
!!!!!
!!! !
! ! ! !!!
!!!!
!! !
! !
!!!!
! !
!!

Inland!Sardinia North!Apulia Sicily

200
palmitoleic

!
! ! !
! !

! ! !! ! !
! !!
!!!
!!
100
!!!!
!! !
!
! !
!
!!
!
!
!!
!
!
!!!
!
!
!
!!!
! ! ! !!!
!
!!!
! !
!! !!
!
!
!!!!
! ! !
! !!!!!
!! !! ! ! !
!!! !
!!
!! ! !!!!! !!
!
!
! !
! ! !
! !
!

South!Apulia Umbria West!Liguria

!
! !
!
! !!
!! ! !
!
!! !! !!!
! !
!!! !
!!! !!!
200 ! !!
!!!
!!
!!
!
!
! !!
!
!
!
!
!
!
!
!
!
!!
!!
! !!
!!
!
!
!!
!
! ! ! ! !
!!!
!
!!! !!
!
!
!
!
!
!
!
!
!!!
!
!!!
!
!
!
!
!
!!
!
!!
!! !
!!
!!!!!
!
!!!!!
!
!! !
!! !
!!!
! !!
!!!
!!
!
!
!
!
!
!
!
!!!!
!
! !! !!
!!
! !! !
!! !
! !!!
! ! !!
100
! !!! !
!!!!
! !
!
!!
!!
!!!!!!!
!
!
! !
!
! ! !
!
!!
!!!
! !!
!
!
!!!!
!
!
!
!

600 900 1200 1500 1800

600 900 1200 1500 1800
600 900 1200 1500 1800
palmitic

FIGURE 10.12: A trellis display of scatterplots for two of the variables in the olives
dataset. The data are in close clusters for several of the areas in different parts of the
plot. In two of the areas, Sicily and South Apulia, there are strong linear associations.
Comparisons, Comparisons, Comparisons 215

Trellis graphics can be drawn in R using the lattice package or the facetting
options in ggplot2. It is interesting to compare the trellis graphic with a single scat-
terplot of the two variables in which the areas have been allocated different colours,
Figure 10.13. Some features are easier to see, others not.

ggplot(olives, aes(palmitic, palmitoleic)) +

geom_point(aes(colour=Area)) +
theme(legend.position = "bottom") +
guides(col = guide_legend(nrow = 2))

!
!
!!

! !
!
!
!!!
! !
! ! ! !
! !
! !
!! !
! !!
! ! ! !
! ! ! !! !
!
!!! ! ! !
!! !! !
200 !! !
! !! !!! !!
! !
!
! !! !
!! ! !!
! ! ! !!! ! !!!
!
!
! ! ! !! !!!!!! ! !
! ! !
!! !
!!
! !
! !
! ! !
! !! !! ! ! !!
!!! ! !
!! ! !
! ! !!!
! !! ! !
!! ! !! !! !! !
palmitoleic

!! ! ! ! !! ! ! ! ! !
! ! !! !
!
!!!
!
!!
!!! ! ! ! !
! ! ! !!! ! !!
! !! !! !! ! !
! ! !
! ! !
!! !
! !! ! !!
!!
! ! !
! ! ! !
! ! ! !
!!
! ! !! ! !!
! !
! ! ! ! !! !
! ! !! !! !!!
!!
!
! ! ! !!!!! ! ! ! !
!
! !
! ! ! !!! ! ! ! !!
! !! !!! ! !
100
!!! !
! !!
! !!!!! ! !!! !
!! !!! ! !
!
!
!! !!! !!!!!! !!
! !! ! !! ! ! !
!
!! !!!
!!! !!
!
! ! !
! ! !
!!! !
!
!
! !! ! ! ! !!
! !! ! ! ! ! !
! ! !! ! ! ! !!!!
!!!!! ! !!! !
! !
! !!!!! ! ! ! ! !!
! ! ! !! ! !
!
!
! !!!!! !!!!!
! !!
!!
! !! ! !!!!
!! !
!
!! !! !!! !
! !! !
!! ! !! ! ! ! !
! !
! ! !
! !
!!

!
!

600 900 1200 1500 1800

palmitic

! Calabria ! East!Liguria ! North!Apulia ! South!Apulia West!Liguria

Area
! Coast!Sardinia ! Inland!Sardinia ! Sicily ! Umbria

FIGURE 10.13: A single scatterplot for the data from Figure 10.12 with the points
coloured by the variable Area. Overall there is a linear relationship between the
variables palmitic and palmitoleic driven by the South Apulia region. Al-
though the areas are separated to some extent, the density of points, overplotting,
and the difficulty of distinguishing some of the colours make detailed interpretation
tricky.
216 Graphical Data Analysis with R

With interactive graphics you can select individual groups in a scatterplot and
see them highlighted, while leaving the other cases in the background, providing
context. A similar effect, but for all groups simultaneously, can be achieved by draw-
ing a trellis plot with each panel displaying both its particular group highlighted in
the foreground and also all the other cases in grey in the background. Figure 10.14
demonstrates the idea. It uses the function facetshade from the extracat pack-
age, which supplies the necessary additional facetting option for ggplot2 objects.
The alpha function from the scales package is used to set the alpha-blending of the
points in the background.
It is now easier to judge the relative positions of the groups compared to the rest
of the data than it was in Figure 10.12. A similar approach was used with some of
the parallel coordinate plots in Chapter 6.
Figures 10.12, 10.13, and 10.14 offer three different ways of comparing scatter-
plots by groups. Personally I prefer Figure 10.14 to Figure 10.12, especially when
alpha is chosen carefully, which may require a little experimentation. Figure 10.13
and Figure 10.14 are alternatives for different situations. The former is fine for well
separated groups and when little display space is available (for whatever reason).
The latter is better, when groups overlap. As always with facetting with no prede-
fined order (the default is alphabetic), it is worth giving some thought as to whether
another ordering might be more informative. For this dataset a grouping by the vari-
able Region would make sense.

10.7 Graphics principles for comparisons

Graph size Each graphic in a comparison must be drawn to the same size and as-
pect ratio. Comparing graphics of different sizes is possible, just more difficult.

Common scaling The importance of using the same scales for groups being com-
pared, common scaling, has been mentioned several times. It is easy to forget
when using default options for displays. In general, both the horizontal and ver-
tical axes should be the same for all graphics to be compared, although, as so
often, the exception proves the rule. Histograms of two groups of different sizes
may be better drawn with their own vertical axes if it is the form of the distribu-
tion that is to be compared rather than the data frequencies.
Alignment Graphics to be displayed can be aligned vertically, which is good for
comparing variability and form, or aligned horizontally, which is good for com-
paring the levels of peaks and troughs. Probably it is best to look at both. Either
is much better than trying to compare without alignment.

Single and multiple windows Plotting all groups in a single window can aid or
hinder comparison. In a single graphic well-separated groups can easily be dis-
tinguished, whereas overlapping groups can be difficult to detect. Small multi-
Comparisons, Comparisons, Comparisons 217

data(olives, package="extracat")
library(scales)
fs1 <- facetshade(data = olives,
aes(x = palmitic, y = palmitoleic), f = .~Area)
fs1 + geom_point(colour = alpha("black", 0.05)) +
geom_point(data = olives, colour = "red") +
facet_wrap(f=~Area, nrow=3) + theme(legend.position="none")

Calabria Coast!Sardinia East!Liguria

200
!
!
! ! !!!!
!
!!! !!!
!!!
! ! !
!!
!
!
!!!
!!
! !!!!
!
!! !!!! !! !!
100 !! ! !! !
!!!!!!
!
! ! !
!
!!
!
!
!!!
!!
! !!!!!! !!
!
! !!! !! !
!!!!!
!!! !
! ! ! !!!
!!!!
!! !
! !
!!!!
! !
!!

Inland!Sardinia North!Apulia Sicily

200
palmitoleic

!
! ! !
! !

! ! !! ! !
! !!
!!!
!!
100
!!!!
!! !
!
! !
!
!!
!
!
!!
!
!
!!!
!
!
!
!!!
! ! ! !!!
!
!!!
! !
!! !!
!
!
!!!!
! ! !
! !!!!!
!! !! ! ! !
!!! !
!!
!! ! !!!!! !!
!
!
! !
! ! !
! !
!

South!Apulia Umbria West!Liguria

!
! !
!
! !!
!! ! !
!! ! ! !!!
!! ! !
! !!!!!
!!
200 ! !! ! ! !
!!!
!
!!!
!!
!!
!
!
!!!!
!
! !
!! ! !!! !!
! !! ! !
!!!
!!!!
!
!
!
! ! !!
!!
!!
!
!!!
!!!!
! !! !!
!!!!! !
!!
!
!!!
!
!!!
!
!!! !
!
!! ! !
!!!
!!
!! !!!!
!!!
!!
!
!
!
!
!
!
!
!!!!
!
! !
! !!
!!
! !! !
!! !
! !!!
! ! !!
100
! !!! !
!!!!
! !
!
!!
!!
!!!!!!!
!
!
! !
!
! ! !
!
!!
!!!
! !!
!
!
!!!!
!
!
!
!

600 900 1200 1500 1800

600 900 1200 1500 1800
600 900 1200 1500 1800
palmitic

FIGURE 10.14: Another version of the trellis display in Figure 10.12. All cases are
drawn in every panel with those for the corresponding subgroup drawn in colour
on top and the remaining cases drawn in grey with alpha-blending underneath. This
kind of plot shows the subgroups in context and makes it easier to assess the rela-
tive shapes of the subgroups. The West-Liguria oils have slightly higher palmitoleic
values than the others and the East-Liguria outlier is an outlier for the whole dataset.
218 Graphical Data Analysis with R

ples, with one display for each group, are the better alternative for overlapping
groups. It can be helpful to draw the rest of the cases in the background to provide
context as shown in Figure 10.14.
Colour and shape Distinguishing groups is helped considerably by using unique
colours for each group. Shape and size are sometimes used for points, but colour
is much more effective.
What is being compared A feature in a plot may subconsciously be compared with
the rest of the data, with another part of the data, with several other parts of the
data, with all the data, with the data that are graphically visible (when there
are overlapping points, for instance), or even with remembered data that are not
shown. If something interesting is found, it is valuable to determine which of
these comparisons led to your thinking it was ‘interesting’.

10.8 Modelling and testing for comparisons

1. Comparing means
The default for comparing two means is the t-test. Even if the necessary as-
sumptions are not ideally met, very low p-values are good evidence for real dif-
ferences. You just have to decide what ‘very low’ means, although presumably
everyone would agree that the p-values reported in §10.1 are.
2. More complex comparisons
When additional factors are taken into account, comparisons are made using lin-
ear models. Testing should then take into account the large number of tests that
are carried out.
3. Comparing rates
Logistic regression is used for comparing two rates while including the possible
influences of other explanatory variables. Proportional odds models may be used
when there are ordinal rates.
4. Non-parametric approaches when standard assumptions do not hold
There is a range of non-parametric tests which may be applied, including
Wilcoxon for comparing two means and Kruskal-Wallis for non-parametric anal-
ysis of variance.
Multiple testing Looking for several kinds of difference at once amongst many dif-
ferent subgroups means that something of potential interest is bound to appear. This
can even happen when looking at differences
 in rates between only a few groups:
given m groups there are at least m2 possible paired comparisons, and still more if
comparisons of combinations of groups are included. The pragmatic approach is to
examine the most extreme differences first and to bear in mind that the significance
of any supporting statistical tests carried out will be affected by multiple testing.
Comparisons, Comparisons, Comparisons 219

There are formal methods for counteracting explicit multiple testing, such as
Bonferroni corrections and FDR (false discovery rates). Details can be found in
[Bretz et al., 2010] and [Benjamini, 2010]. These approaches do not cover graphi-
cal analyses. There are so many features that might be seen that you cannot be sure
just how many tests are implicitly being carried out.

Main points
1. There is more to comparing groups than comparing their means. Quantiles, vari-
ability, and distributional patterns may also be compared—at least graphically
(§10.1).
2. Fair comparisons need comparable data (populations, sources, variables, mea-
surements, groups) and comparable graphics (size, aspect ratio, scaling, align-
ment) (§10.2).
3. Different comparisons show different aspects of datasets (Figure 10.6).

4. Comparisons may be made at different levels. Selecting the right comparison

requires careful thought. Choosing appropriate conditioning variables is often
important (Figures 10.8 and 10.9).
5. Appropriate intervals are needed for testing differences. Each comparison of in-
terest may require its own model (§10.5).
6. Trellis scatterplots in which the rest of the data are plotted in grey in the back-
ground are effective for comparing overlapping subgroups (Figure 10.14).
220 Graphical Data Analysis with R

Exercises
1. Swiss banknotes
Consider the bank dataset from the gclus discussed in §10.1.
(a) How do the distributions of the variable Length differ for the two groups
defined by the Status variable? Draw histograms, boxplots, and empir-
ical distribution functions. Which display do you find most informative?
What are the advantages and disadvantages of the three displays in this
application?
(b) Are the mean lengths of the two groups significantly different?
2. Petrol consumption
In §10.3, the petrol consumption of cars was compared for two datasets using
miles per gallon figures. In most European countries, consumption is measured
the other way round, as litres needed to drive 100 kilometres.
(a) Draw comparative plots of petrol consumption, measured in gallons needed
to drive 100 miles, for the two datasets. What features, if any, are notable
in the plots?
(b) Carry out a t-test comparing the two means. Discuss your result in conjunc-
tion with the result of the t-test carried out in §10.3.
(c) A major influence on petrol consumption is the weight of a car. Draw scat-
terplots of MPG.city and 1/MPG.city against Weight for the Cars93
dataset. What conclusion do you draw and which scatterplot do you prefer?
3. Barley (corrected?)
Cleveland suggested that some of the data in the barley dataset analysed in §10.4
is possibly wrong. Construct a new version by switching the data for the Morris
site for the two years.
(a) Draw revised versions of the first two figures from the chapter. How differ-
ent are your plots compared to the original ones?
(b) Refit the linear model and plot the interval estimates. Do the conclusions
about the differences between the varieties change with this revised dataset?
4. Balance of trade
England’s balance of trade with the East Indies was discussed in §10.3.
(a) The relative balance of trade was calculated using the average of imports
and exports. How would that graph look if you used (i) imports and (ii)
exports as the denominator? How would you interpret the graphs and what
headlines would you give them in a newspaper article?
(b) What is the relation of imports to exports between individual countries
nowadays? Find data for your own country and draw time series of the
balance of trade between it and its three most important partners over the
last twenty years. How have you defined “most important”?
Comparisons, Comparisons, Comparisons 221

5. Diamonds
The diamonds dataset from the ggplot2 package includes information on over
50, 000 round cut diamonds.
(a) The variable color has categories ranging from D (best) to J (worst).
Draw a plot showing how price varies with color. Are you surprised by
the result? What might explain it?
(b) Draw a plot of the color coefficient estimates with 95% confidence inter-
vals. What conclusions would you draw?
6. Swiss banknotes (again)
Consider the variables Right and Left, measurements of the edge widths of
the notes.
(a) What do the distributions of the differences between these measurements
for each note look like for the two groups? Are the differences significantly
different from zero?
(b) The measurements Bottom and Top for the margin widths might also be
expected to be close to equal for each note. Are they and does the difference
relate to the edge width differences?
(c) Instead of using absolute differences, proportionate differences could be
used. Draw a plot to compare the scales of the proportionate differences
for the edges and margins. What denominator would you suggest? Do you
think the data are reported precisely enough for these analyses?
7. Olkin95
There are data on 70 different studies of thrombolytic therapy after acute my-
ocardial infarction in the Olkin95 dataset in the meta package. (This dataset was
also used in Exercise 2 in Chapter 5.)
(a) Plot the event rates for the experimental groups against the corresponding
rates for the control groups. What does your plot show?
(b) The sizes of the studies should also be taken into account. Draw a scat-
terplot of the rate differences in each study against the size of the study,
using the total number of participants for the size. (This is a kind of funnel
plot.) What conclusions would you draw from your plot? How much does
it matter, if at all, that the experimental and control groups are not always
the same size?
8. Intermission
The Judgement of Paris by Rubens hangs in the National Gallery in London.
How would you compare it with paintings of the same scene by, amongst others,
Cézanne, Renoir, Raphael, and Watteau?
9. Intermission (extended)
Manet’s painting Le déjeuner sur l’herbe is in the Musée d’Orsay in Paris. Monet
painted a version a couple of years later and Picasso was inspired to make over
one hundred drawings and some twenty-seven paintings of the scene. How do
the versions by Picasso and Monet compare with Manet’s?
11
Graphics for Time Series

The Government are very keen on amassing statistics—they collect them, add
them, raise them to the nth power, take the cube root and prepare wonderful
diagrams.

Josiah Stamp

Summary
Chapter 11 discusses drawing graphics for displaying the information in time series.

11.1 Introduction
Graphic displays are excellent for showing time series data and they are used exten-
sively. Several of Playfair’s graphics are of time series, and his displays of Balance of
Trade data are well known [Playfair, 2005]. An example was shown in Figure 10.6.
Economic and financial data are obvious applications for time series, but weather
data, sports performances, and sales results all involve time series data too.
Graphics are particularly helpful for studying several time series at once. If each
of a set of series can be transformed to a common scale, then they can all be plotted
on the same display and it is possible to compare trends, changes in levels, and other
features. If series show a common pattern for only part of the time period displayed,
then that will be visible in the graphic while it may not be apparent in a statistical
summary of the data.
Two entertaining and surprisingly informative time series graphics applica-
tions are Name Voyager [Wattenberg, 2005], which displays patterns of choice
of baby names in the United States over the last 130 years, and Ngram Viewer
[Google, 2010], which allows you to examine how word and phrase use developed
over time in books. Much more can be found on visualising time in Graham Wills’s
book of the same name [Wills, 2012].

223
224 Graphical Data Analysis with R

11.2 Graphics for a single time series

Consider the average number of goals scored per game in each of the first 46 seasons
of the German Bundesliga from 1963/64 to 2008/09. In the first two seasons there
were only 16 teams instead of 18, and in 1991/92 there were 20 teams. For that one
year two extra teams were included to allow teams from the former East Germany
to join. The averages can be calculated by aggregating the goals reported by team in
the dataset Bundesliga in vcd and then dividing by the number of games. To display
these data, there are several decisions to be made:
Symbol Should the data be represented by points and/or lines or by bars?
Scale What scale should be used for the y-axis, from what minimum level to what
maximum level? Including zero on the y-axis, which is often rightly recom-
mended in other circumstances, conveys the level of goals scored, but variability
and possible trends may be shown less well.
Aspect ratio The look of a time series display is affected strongly by the aspect
ratio, that is the length of the y-axis to the length of the horizontal time axis. The
temptation to make a series appear flat (lengthen the time axis) or steep (shorten
the time axis) should probably be resisted. [Cleveland and McGill, 1987] recom-
mends choosing an aspect ratio that has the slopes of interest in the display at an
angle of 45◦ . You just have to decide what the slopes of interest might be.
Trend Should a trend estimate in form of a smoother be added to the display? And,
if so, which smoother should you use?
Gaps If the series is regular, how should gaps in the data be represented? Gaps
could be time points when values were not recorded for some reason or they
could be periods where there were no data to record, for instance holidays when
shops are shut.
And there are additional, less critical, but still important decisions, such as the la-
belling of the time axis, the use of colour, and annotations referring to events at
particular times.
Figure 11.1 shows the goals data using points and lines, a vertical scale slightly
larger than the minimum and maximum, an aspect ratio with width twice height and
including a spline smoother with pointwise 95% confidence intervals. The time axis
is labelled by the first year of each season, i.e., the average goals per game for season
1970/71 are plotted at t = 1970. The plot shows that the average was up near 3.5 in
the 1970s and early 1980s, then declined to under 3.0 and has been fairly constant
since. There is no evidence that the introduction in 1995 of three points for a win
instead of two made any difference to the numbers of goals scored.
Graphing the series for the average home goals and average away goals separately
(Figure 11.2) reveals the interesting fact that the variation depended mainly on the
numbers of home goals scored.
Graphics for Time Series 225
Bundesliga goals per game by year

4.0

! !
! !
3.5
! !
!
! ! !
!
! ! !
!
! ! ! !
! !
!

3.0 !
! ! !
! ! ! ! ! ! !
! ! !
! ! ! !
!
! !
!
!
!
2.5

2.0

1970 1980 1990 2000 2010

FIGURE 11.1: A time series of the average number of goals scored in the German
Bundesliga over the first 46 seasons from 1963/64 to 2008/09 with a gam smoother
added. There were 16 teams in the first two seasons and 20 in season 1991/92; oth-
erwise there were always 18 teams. The average number of goals scored has been
fairly stable since 1990, but at a lower level than in earlier years.
data(Bundesliga, package="vcd")
goals <- Bundesliga %>% group_by(Year) %>%
summarise(hg=sum(HomeGoals), ag=sum(AwayGoals),
ng=n(), avg=(hg+ag)/ng)
library(mgcv)
ggplot(goals, aes(Year, avg)) + geom_point() + geom_line() +
stat_smooth(method = "gam", formula = y ~ s(x)) +
ylim(2,4) + xlab("") + ylab("") +
ggtitle("Bundesliga goals per game by year")

2.5

2.0

1.5

1.0

0.5

1970 1980 1990 2000 2010

FIGURE 11.2: The average numbers of home (red) and away (blue) goals scored in
the Bundesliga.
ggplot(goals) + geom_line(aes(Year, hg/ng), colour="red") +
geom_line(aes(Year, ag/ng), colour="blue") +
ylim(0.5, 2.5) + xlab("") + ylab("")
226 Graphical Data Analysis with R

10.0

7.5

5.0

2.5

0.0
1970 1980 1990

FIGURE 11.3: Unadjusted quarterly (West) German unemployment rates from 1961
to just after the unification of the two Germanies. Unemployment was very low in
the 1960s and early 1970s, apart from a short spell at a higher level in 1967. There
were distinct jumps in unemployment in the mid 1970s and in the early 1980s due to
the oil crises of 1973 and 1979. Unemployment was declining at the end of the series
from the high levels of the mid to late 1980s. The series shows a strong seasonal
pattern with higher levels in winter than in summer.
library(zoo)
data(GermanUnemployment, package="AER")
ge <- as.zoo(GermanUnemployment)
autoplot(ge$unadjusted) + xlab("") + ylab("")

The Bundesliga dataset is an ordinary dataframe, but many time series datasets
in R are provided in special time series classes, like the German quarterly unem-
ployment data in the AER package. You can just use a default plot function or,
alternatively, you can use the zoo package. It offers functions to convert other time
series classes to its zoo class and an autoplot function to produce ggplot ob-
jects that can then be amended with other ggplot2 options. This has been done in
Figure 11.3 for (West) German unemployment data. The series shows a regular sea-
sonal pattern and some relatively sharp changes in level between periods of fairly
steady unemployment levels. The comparable unemployment rate in 2014, some 20
years later, is a little over 6%.

11.3 Multiple series

Various situations can arise when you have multiple time series, and they may need
to be treated in different ways. You can have related series for the same population,
such as death rates from different causes, and these can be plotted together or sep-
arately. You can have the same series for different subgroups, such as overall death
Graphics for Time Series 227

rates for men and women, and they might be plotted together in a single display or
individually in a trellis display. You can have series on quite different scales for the
same population, for instance various economic indicators for a country, and these
can be plotted in a single display with some kind of scale standardisation or each in
their own window with their own individual scale.

Related series for the same population

The dataset Nightingale in HistData includes numbers of deaths in each month and
monthly annualised hospital death rates for three different causes of death for the
British Army in the Crimean War, from April 1854 to March 1856. The annualised
rates have been calculated by multiplying the deaths in a month by 12 and dividing by
the Army strength that month. Figure 11.4 shows the three rate series plotted together
and it is obvious that deaths from disease dominate and that the winter of 1854/55
was a very bad time. That winter included the first half of the Siege of Sevastopol
and the Charge of the Light Brigade at the end of October 1854. To put the numbers
in context, 156 members of the Light Brigade died at the Charge [Wikipedia, 2014],
while 503 members of the Army died in hospital from disease in that October—and
that was one of the months with a relatively low death rate from disease!

!
1000

750
Death rates

500 !

! !
!

250 !
!
! !
!
!
!

! !
! !
0 ! ! ! ! ! !

1854!07 1855!01 1855!07 1856!01

Date

FIGURE 11.4: Annualised monthly hospital death rates per 1, 000 in the British
Army from disease (black line and points), wounds (red dashed line), and other
causes (blue dotted line) in two years of the Crimean War, 1854-1856.
data(Nightingale, package="HistData")
ggplot(Nightingale, aes(Date)) +
geom_line(aes(y=Disease.rate)) +
geom_point(aes(y=Disease.rate), size=2) +
geom_line(aes(y=Wounds.rate), size=1.3, col="red",
linetype="dashed") +
geom_line(aes(y=Other.rate), size=1.3, col="blue",
linetype="dotted") + ylab("Death rates")
228 Graphical Data Analysis with R

Instead of drawing each time series individually, there is an alternative approach,

which can be more efficient. You first construct a new ‘long’ dataset of three vari-
ables: the time values needed, the values of the different time series all in one vari-
able, and a grouping variable of the time series labels. The following code would
produce a similar plot to Figure 11.4. The first line ‘melts’ the new dataset and the
second part uses the grouping variable to draw the time series. There is no particular
saving for this plot, but there would be if you wanted to display more series together.

library(reshape2)
Night2 <- melt(Nightingale, id.vars="Date",
measure.vars=c("Disease.rate", "Wounds.rate", "Other.rate"),
variable.name="NightV", value.name="Nightx")
ggplot(Night2, aes(Date, Nightx,
colour=NightV, group=NightV)) + geom_line() + ylab("")

Same series for different subgroups

You can have time series of the same variable for several countries, giving time series
that can be analysed together on the same scale, for instance GDP. Then you would
have to decide if it is the total GDP’s that are of interest (so the figures for the United
States swamp those of many smaller countries), or GDP related to population size.
An example was given in Figure 6.8 for the corn yields in the US by state over time. It
is obviously sensible to study the bigger states first, but patterns over time for smaller
states cannot be seen. More than one plot is needed, possibly an additional one for
only the smaller states or one for each state separately using individual scaling.

Series with different scales

There may be several different time series for the same country: GDP and, say, im-
ports, exports, and unemployment levels. Comparing patterns of movement in these
series on the same graphic means finding some way to plot different variables on a
common scale. One year can be chosen as a baseline and given a value of 100 with
other years’ values being transformed accordingly. That is often done with share
prices and other financial variables. Alternatively, all series can be standardised by
their respective means and standard deviations. This has the advantage of keeping se-
ries’ levels and variabilities comparable but the disadvantage of changing with every
additional time point.
Figure 11.5 shows an example of starting series off at the common value of 100.
The data are the daily closing share prices in 2013 for four computing companies:
Apple, Google, IBM, and Microsoft. The data were downloaded from Google Fi-
nance using the package FinCal. For this plot the vertical scale limits have been
determined by default, using the overall minimum and maximum of the four series.
Scale limits and the aspect ratio influence the look of time series a lot and it is usually
worth experimenting with other alternatives. Apple started the year badly, but recov-
ered to its original value. Google and Microsoft did very well. IBM started all right
Graphics for Time Series 229

and then fell away and ended up about 10% down. Note the two sharp falls (Apple
in January, Microsoft in the summer) and one steep rise (Google in the autumn).
A time series may have to be transformed to make it comparable with other series,
as here, or even to make it more comparable with itself. Many monetary time series,
such as salary levels or indeed GDP, are usually adjusted to take account of inflation.

140

120

100

80

Jan 2013 Apr 2013 Jul 2013 Oct 2013 Jan 2014

Apple Google IBM Microsoft

FIGURE 11.5: Closing share prices for four US computing companies in 2013. Each
share price was set to 100 at the start of the year and the other prices transformed ac-
cordingly. Investing at the beginning of 2013, it would have been best to buy Google
and Microsoft. Different conclusions would be drawn with other starting points.
library(FinCal); library(reshape2)
sh13 <- get.ohlcs.google(symbols=c("AAPL","GOOG","IBM","MSFT"),
start="2013-01-01",end="2013-12-31")
SH13 <- with(sh13, data.frame(date = as.Date(AAPL$date),
Apple = AAPL$close,
Google = GOOG$close,
IBM = IBM$close,
Microsoft = MSFT$close))
SH13a <- lapply(select(SH13, -date), function(x) 100*x/x[1])
SH13a <- cbind(date = SH13$date, as.data.frame(SH13a))
SH13am <- melt(SH13a, id="date", variable.name="share",
value.name="price")
ggplot(SH13am, aes(date, y=price, colour=share,
group=share)) + geom_line() + xlab("") + ylab("") +
theme(legend.position="bottom") +
theme(legend.title=element_blank())
230 Graphical Data Analysis with R

One plot versus many

Displaying multiple series on the same plot works well if there are few crossings
and if the individual series have low variability. Figure 11.4 is effective for showing
the high death rates due to disease, but it would be less good for comparing the two
series of death rates due to wounds and other causes. The four share price series in
Figure 11.5 can be kept separate fairly easily, especially thanks to the use of colour.
Figure 6.8 is also good at providing an overall picture, but does not represent series
for individual states well. When there are many series, then a trellis plot could be
used, with one panel for each series, possibly with the other series displayed with
alpha-blending in the background, as was used in Figure 10.14.
The collection of datasets hydroSIMN in the package nsRFA includes the annual
flows of 47 hydrometric stations in Piemonte and Valle d’Aosta. The data cover 65
years from 1921 to 1985, although only one series is complete and some are recorded
for just a few years. Figure 11.6 shows all the series in one display. There appear to
be some common features, such as the peaks in 1960 and 1962, and a few series
seem to be partially grouped by colour. The hydrometric station code was made a
factor so that more than one colour would be used, and the colours were assigned
sequentially according to the numeric code. Stations with similar codes are generally
close together spatially, which could explain the possible patterns.
Figure 11.7 displays the individual series. It is now much easier to see particu-
lar patterns and when data are available for each station. As well as indicating the
differing levels and variability of series, the individual plots reveal the patterns over
time better and suggest some possible outliers, for instance the first and last values
for station 3.
 Graphics for Time Series

FIGURE 11.6: Annual flows for 47 hydrometric stations in Piemonte and Valle d’Aosta in Italy. There are common peaks, and stations
that are spatially close are likely to have similar patterns. An accompanying map of where the stations are located would be very helpful.
data(hydroSIMN, package="nsRFA")
annualflows <- within(annualflows, cod <- factor(cod))
ggplot(annualflows, aes(anno, dato, group=cod, colour=cod)) +
geom_line() + xlab("") + ylab("") + theme(legend.position="none")
231
232 Graphical Data Analysis with R

1 2 3 4 7 8 9
3000

2000

1000

10 11 12 13 14 15 16
3000

2000

1000

17 18 19 20 21 22 23
3000

2000

1000

24 25 26 27 28 29 30
3000

2000

1000

31 32 33 34 35 36 37
3000

2000

1000

38 39 40 41 42 43 44
3000

2000

1000

45 46 47 48 49
3000

2000

1000

1920194019601980
1920194019601980
1920194019601980
1920194019601980
1920194019601980

FIGURE 11.7: Annual flows for 47 hydrometric stations in Piemonte and Valle
d’Aosta in Italy drawn individually. The changes over time are easier to see than
in Figure 11.6 and individual patterns are clearer.
ggplot(annualflows,
aes(anno, dato, group=cod, colour=cod)) +
geom_line() + xlab("") + ylab("") +
facet_wrap(~cod) + theme(legend.position="none")
Graphics for Time Series 233

11.4 Special features of time series

Data definitions
Time series datasets differ from other datasets because of their dependence on time.
The data have a given order and individual values are not independent of one another.
In the simplest form there is a sequence of measurements of a single variable equally
spaced in time, for example the annual GDP of a country. Even for this apparently
straightforward situation it may be misleading to say “simplest”. The definition of
GDP might have changed over time, the value of money certainly has due to inflation,
and the country itself might have changed. German data before and after 1990 can
only be compared with caution, due to the reunification of the former East and West
Germany.
In standard statistical analyses it is reasonable to assume that all data are defined
similarly and are therefore directly comparable. That is not something that can be
automatically assumed for time series. Unemployment statistics provide an interest-
ing example, as the definition of who is unemployed varies across countries and is
changed quite frequently within countries. The definition of unemployment in the
United Kingdom changed over twenty times in the period from 1979 to 1993 when
Margaret Thatcher was prime minister ([Gregg, 1994]). The change almost always
resulted in a decline in the number of unemployed. The one change in definition in
the other direction had actually been decided by the previous Labour government and
only came into operation during Thatcher’s premiership.

Length of time series

Time series can be short, for example, the annual sales of a new smart phone model,
or long, for example the temperature values recorded every minute at a weather sta-
tion over many years. Sometimes the short-term details of long series can obscure
long-term trends, sometimes they are of particular interest. Plotting series on differ-
ent time scales can be informative and for some financial time series such as ex-
change rates you could plot each minute, each day, each month, or each year. When
you plot values for longer periods, there is a choice of what to plot for each period.
You could take the final value, a value from the middle of the period, an average
value, perhaps even a weighted average of some kind. Alternatively you could plot
some kind of smooth. There are many possibilities.

Regular and irregular time series

In §6.5 parallel coordinate plots were used to plot regular time series, i.e., time series
which are recorded at equally spaced time points. Hourly data, daily data, yearly
data are often of this kind, but there are many series which are not. A history of a
patient’s temperature or blood pressure will rarely be based on a sequence of equally
spaced time points. Political opinion polls are more frequent near elections than at
234 Graphical Data Analysis with R

other times. Sometimes treating the time points as equally spaced can be acceptable,
often definitely not.
Plotting the time scale of an irregular time series accurately is important, and this
is a situation where packages which handle times and dates correctly are a valuable
help, especially as they allow you to plot several series with quite different time
points on the same display. For example, a doctor might want to compare series of
readings for different patients over a year, although he saw them at different times
and different numbers of occasions.
With irregularly spaced series you cannot really speak of missing values and it is
usual to just join successive time values. With regular series, it is a different matter.
Missing values can then be awkward: should you leave a gap or simply join the points
before and after with a straight line?
A related issue is time units that are thought of as the same, but are actually
different, like months. February is always shorter than all other months of the year.
Even the same months in successive years can be of different lengths. Retail sales
could be some 3% higher for this month compared to the same month last year,
because there was one extra shopping day.

Time series of different kinds of variables

Most time series are assumed to be of continuous variables, although you can also
have time series of nominal variables (for instance, a person’s state of health or each
year’s most popular fashion colour) or of discrete variables.
One of the most famous datasets in statistics is von Bortkiewicz’s dataset on
deaths from horsekicks in corps of the Prussian army. It is usually used to illustrate
the Poisson distribution, as in §4.5, but the numbers of deaths are given over 20
consecutive years, so the data can also be viewed as a set of 14 time series, one for
each corps. Figure 11.8 shows the total deaths each year. A barchart has been used
to reflect the fact that the data are sums, single values for each year. The initial rise
in deaths could be due to chance, better reporting, or larger corps sizes. The relevant
data on corps sizes does not seem to be reported anywhere, despite the dataset being
used in so many statistical textbooks.

data(VonBort, package="vcd")
horses <- VonBort %>% group_by(year) %>%
summarise(totalDeaths=sum(deaths))
ggplot(horses, aes(year, totalDeaths)) +
geom_bar(stat="identity") + ylim(0,20)

Outliers
Outliers in time series can be different from outliers for other kinds of datasets. They
are not necessarily extreme values for the whole series, just unusual in relation to the
pattern round them. A high temperature value in winter can be in the middle of the
distribution of values for the whole year and yet still look way out of line for that
Graphics for Time Series 235

20

15
totalDeaths

10

1875 1880 1885 1890 1895

year

FIGURE 11.8: The number of soldiers killed by horse kicks in 14 corps of the Prus-
sian army over 20 years at the end of the nineteenth century.

time of year. The famous lynx dataset (Figure 11.9) offers an interesting example.
(Functions from the zoo package have been used as the dataset is a time series object
of class ts.)
Should the two lower values in 1914 and 1915 be regarded as outliers, where
something different was happening? The fact that this coincides with World War
I makes it tempting to look for some kind of connection. Unusual values like that
are easy to spot, although others may prove more difficult. Distances of points from
curves are judged by the shortest line which can be drawn from the point to the
curve. For time series, you have to assess the vertical distance between the point and
the curve and this is hard to do for steeply rising or falling curves.

!
!
!
6000 !
as.zoo(lynx)

!
!
4000
!
! !
! ! !
! ! !
! ! !
!! !
!
! ! !! ! ! !
! ! !
!
2000 ! ! !
!
! !
!
! ! !! ! !
! ! ! !
! ! !
! ! !
! ! ! ! ! ! ! !
! ! ! ! ! ! ! !!!
!! !! ! !!!! !!! ! ! !!! ! ! ! !! !
0 !! !!!! !!!!! !! !!!!

1820 1840 1860 1880 1900 1920

Index

FIGURE 11.9: A time series of the number of lynx trapped on the Mackenzie River
in Canada. There appear to be two kinds of cycles.
data(lynx)
library(zoo)
autoplot(as.zoo(lynx)) + geom_point()

Individual extreme values can influence scales and make the rest of a plot unin-
formative. This applies to time series just as much as to histograms, scatterplots, and
parallel coordinate plots. However, scales for time series are unusual in that they are
more likely to be influenced by values not included in the plot. If you display only
236 Graphical Data Analysis with R

the last couple of years of the exchange rate between the dollar and the euro, you
may still want to let the vertical axis scale reflect the full history of the exchange rate
since 2002.
The usual principle applies that it is best to draw several displays, zooming in
to inspect details that may otherwise be hidden and zooming out to see the overall
context.

Forecasting
There are two main reasons for studying time series: to try to understand the patterns
of the past and to try to forecast the future. There are a number of alternative ways of
displaying forecasts and it is good practice to make clear where the data end and the
forecast begins, possibly using a change of background shading, a gap, or a dotted
line into the future. Any forecast based on a model should include prediction intervals
and including them in the display also emphasises where forecasts begin.
Figure 11.10 shows a two year forecast of the German unemployment data from
Figure 11.3 using an exponential smoothing state space model. Detailed information
and alternative forecasting models can be found in Rob Hyndman’s forecast package.
As forecasts reach further into the future, the prediction intervals get wider and
wider, which can make it tempting to use narrower ones. This temptation should be
resisted. It is a sad truth that no matter how wide forecast intervals are, they may not
be wide enough.

10

0
1965 1970 1975 1980 1985 1990 1995

FIGURE 11.10: A forecast of (West) German unemployment for the years 1992 and
1993, based on the data from the previous 30 years. Prediction intervals of 80% and
90% are shown.
library(forecast)
par(las=1, mar=c(3.1, 4.1, 1.1, 2.1))
fit <- ets(ge$unadjusted)
plot(forecast(fit), main="")
Graphics for Time Series 237

Seeing patterns
It is remarkable how ready people are to see patterns in time series and to overlook
features that are inconsistent with the supposed patterns. When series are plotted to-
gether, it is surprising, and even somewhat disturbing, how often people see causal
relationships that may at best be associations due to other factors or just to the pas-
sage of time itself. And if you look at enough series, there are bound to be some that
have something in common.
Yule gave an early warning example, the association between standardised mor-
tality rates and the proportion of Church of England marriages over 46 years
([Yule, 1926]). Despite this, [Coen et al., 1969] reported amongst other results that
they had found that the production of motor cars in the United Kingdom (seasonally
adjusted) led the United States S&P stock market index by six quarters.
It would be tedious to list all the various issues to watch out for with time series,
but it is well to remember that there are many.

11.5 Alternative graphics for time series

The best way to display time series is usually with lines joining the individual values.
Only plotting the points can work as long as the variability around any trend is low.
When a time series is short and what is being reported is a sequence of statistics rather
than a continuous measurement, barcharts can be a good alternative. Figure 11.8
gives an example.
Parallel coordinate plots display regular time series well and can be readily used
for plotting multiple series (§6.5). Their particular advantage lies in allowing you to
inspect distributions of values across series at individual time points using boxplots.
When data are recorded daily and there are expected to be strong calendar effects,
calendar plots can be a useful display. They are drawn just like a calendar, showing
the twelve separate months and the weekly structure of the year. The rectangle for
each day is coloured according to the value of the variable being displayed, so they
are like a heatmap. Calendar plots pick out patterns in the weekly and monthly struc-
ture well, so they are good for checking the effects of weekends or special dates.
As they use a colour scale, they work best when differences are large. The openair
package offers a calendar plot and examples with ggplot2 can be found on the web.
The association between two time series can be studied by plotting the two series
together or by drawing a scatterplot, plotting one series against the other. That is what
Yule did for the example mentioned in the last section [Yule, 1926].
238 Graphical Data Analysis with R

11.6 R classes and packages for time series

Time and date variables require special treatment. Hours, minutes and seconds need
to be handled properly, as do months, changes from winter to summer time and back,
leap years, and time zones. R provides several packages which deal with these issues,
for instance lubridate, and the time series task view [Hyndman, 2013] is a good
reference. A number of packages offer special time series classes and the time series
task view is again the best source for information. There are tools to convert time
series objects from one class to another, as in the package zoo, to facilitate using
different classes and the xts package offers a single general class covering all.
Plotting tools in time series packages are particularly useful for labelling the
time axis appropriately, something which can be a difficult and frustrating task if
carried out from scratch. The website [TimelyPortfolio, 2013] gives a short historical
summary of time series plotting possibilities in R together with an example of the
same financial series plotted by the various packages discussed.

11.7 Modelling and testing time series

1. Single time series with regularly spaced time points
Time series models generally assume equally spaced intervals between data
points. Early on, attempts were made to decompose time series into four com-
ponents: trend, seasonal, cyclical, and residual. Later on came the ARIMA mod-
els (autoregressive integrated moving average) promulgated by Box and Jenkins.
With increasing interest in financial time series have come GARCH models (gen-
eralised autoregressive conditional heteroscedasticity) and developments from
them. And so it goes on, with everyone seeking the holy grail that will enable
them to forecast the future successfully (and profitably).
2. Short, irregular time series
Time series model fitting generally requires a lot of data. When series are both
short and irregular, some kind of smoothing is a possible option. Figure 11.1
shows an approach using spline smooths, which has the advantage of providing
confidence intervals as well.
3. Multivariate time series
Modelling of time series has concentrated on single time series. Modelling sev-
eral time series together is a complex problem and progress has been slow.
[Tsay, 2014] gives a thorough overview.
Graphics for Time Series 239

Main points
1. There are many different factors to bear in mind when drawing and interpreting
time series (§11.2).
2. Single time series plots can show a lot of information (e.g., Figure 11.3).
3. Multiple series can be drawn in a single display to make comparisons easy (Fig-
ures 11.4 and 11.5).
4. Dates and times have tricky properties and it is best to take advantage of packages
that can deal with them (§11.6).
5. Time series are a different kind of data and need to be treated specially. The
data are not i.i.d. variables. Graphics are good for irregular time series and for
displaying multiple time series. Models have difficulties with both. (§11.4)

Exercises
More detailed information for the datasets is available on their help pages in R.
1. Air miles
The dataset airmiles is a time series of the miles flown annually by commercial
airlines in the US from 1937 to 1960.
(a) Before plotting the graph, think about what shape you would expect it to
have. Plot the series and comment on the differences between what you get
and your expectations.
(b) Which aspect ratio conveys the information you find in the series best?
(c) Do you think the graph looks better as a line graph (as suggested on the R
help page for the dataset) or with points as well?
(d) Might plotting a transformation help you to look more closely at the early
years or would zooming in be sufficient?

2. Beveridge Wheat Price Index

The Beveridge index of wheat prices covers almost four hundred years of Euro-
pean history from 1500 to 1869 and is available in the dataset bev in tseries.
(a) Plot the series and explain why you have decided to plot it in that way.
(b) Are there any particular features in the series which stand out? How would
you summarise the information in the series in words?
(c) Many important historical events took place over this time period, including
the Thirty Years’ War, the English Civil War, and the Napoleonic Wars. Is
there any evidence of any of these having an effect on the index?
240 Graphical Data Analysis with R

3. Goals in soccer games

The Bundesliga dataset was used in §11.2.
(a) Plot graphs of the rates of home and away goals per game over the seasons
in the same plot. What limits do you recommend for the vertical scale?
(b) Other possibilities for studying the home and away goal rates per game
include plotting the differences or ratios over time and drawing a scatterplot
of one rate against another. Is there any information in these graphics that
is shown better by one than the others?
(c) Can you find equivalent data for the top soccer league in your own country
and are there similar patterns over the years?
4. Male and female births
Important early demographic analyses were carried out on English data from the
seventeenth century. The Arbuthnot dataset in the HistData package includes
data on the numbers of male and female christenings in London from 1629 to
1710.
(a) Plot the number of male christenings over time. Which features stand out?
(b) Why do you think there was a low level of christenings from around the
mid-1640’s to 1660?
(c) Two low outliers stand out, in 1666, presumably because of the Great Fire
of London and the plague the previous year, and in 1704. A possible ex-
planation for the 1704 outlier is given on the R help page for the dataset.
Compare the data values for 1674 and 1704 to check the explanation.
5. Goals in soccer games (again)
Consider the numbers of goals scored by each team.
(a) How would you plot the annual average goals per home game for each team
in the Bundesliga over the 46 seasons in the dataset? Would you choose a
single graphic or a trellis display? Only one team has been a member of
the Bundesliga ever since it started, Hamburg. How do you think the time
series of teams with incomplete records should be displayed?
(b) You could compare the annual home and away scoring rates of particular
teams by plotting the two time series on the same display or by drawing a
scatterplot of one variable against the other. Using the two teams Hamburg
and Bayern Munich, comment on which display you think is better. Do the
displays provide different kinds of information?
6. Deaths by horsekick
Plot separate displays for each of the 14 corps in the von Bortkiewicz dataset
(VonBort in vcd).
(a) Do any of the patterns stand out as different?
(b) 11 of the 14 corps had no deaths in the first year (1875). Could this be worth
looking into?
Graphics for Time Series 241

7. Economics data
The package ggplot2 includes a dataset of five US economic indicators recorded
monthly over about 40 years, economics.
(a) If you plot all five series in one display, is it better to standardise them all
at a common value initially or to align them at their means and divide by
their standard deviations? What information is shown in the two displays?
(b) Alternatively you could plot each series separately with its own scale. Do
these displays provide additional information and is there any information
that was shown in the displays of all series together that is not so easy to
see here?
8. Australian rain
The dataset bomregions in the DAAG package includes seven regional time se-
ries of annual rain in Australia and one time series averaged over the country.
(a) Can all seven regional series be plotted in one display or are individual
displays more informative?
(b) Are there any outliers in the series and do they affect the scales used ad-
versely?
(c) Is there any evidence of trend in the series? Are there cyclical effects?
9. Tree rings
The package dplR includes several tree ring datasets, including ca533. There are
34 series of measurements covering 1358 years in all from 626 to 1983. Note
that no time variable is given, just the information that the data were recorded
annually. The actual time range can be found from NOAA’s tree ring database
website.
(a) Plot all 34 series in separate displays. Are there any common features?
(b) There are at least two series with much higher maxima than the others.
Compare a display excluding these series, but still retaining the same scal-
ing for all the plots, with a display where each series is plotted with its own
scale. What are the advantages and disadvantages of the two approaches?
10. Intermission
Salvador Dali’s painting The Persistence of Memory is in the New York Mu-
seum of Modern Art. Do you think the distorted clocks could be interpreted as
alternative models of time series?
12
Ensemble Graphics and Case Studies

Show and tell.

Common expression

Summary
Chapter 12 discusses using ensembles of graphics to explore datasets and contains a
set of case studies for readers to investigate.

12.1 Introduction
When you first look at a dataset it is a good idea to draw many graphics to get a
feel for the data. It is unnecessary to worry about labelling, legends, and the like, the
aim is to quickly gain understanding of the information available. Resizing windows,
varying aspect ratios, and redrawing displays with other formatting options all aid the
process, and the windows can be casually spread around the screen or discarded. The
many displays are for a single analyst’s use, your own.
Saving graphics for future use or for presenting any features discovered to others
is quite a different matter. Then you have to think about choosing effective versions
of each graphic, of grouping them in a structured order, and of combining them in an
ensemble, possibly linking them with text to tell a story. Carefully laid out combi-
nations of graphics are primarily for presentation purposes. Tidying up graphics and
planning a good layout can take up too much time during an exploratory analysis
and, more importantly, a different kind of thinking is involved. Deciding on the tech-
nical details of drawing a display is not the same activity as studying the content of
graphics displays to discover information. Drafting issues and computing concerns
should not hamper graphical reasoning during exploratory analyses.
The examples in this chapter are all for presentation, intended to be readily under-
standable and to look reasonably presentable. A display of a whole raft of exploratory
graphics would be complicated to explain and would represent only a momentary
snapshot of an exploratory analysis. It is rarely crucial to describe how a result was

243
244 Graphical Data Analysis with R

obtained, but it is essential to be able to draw graphics which show that there is
evidence to support the result.
Figure 12.1 shows three plots from the coffee dataset in the package pgmm. There
are 43 coffees of two varieties, Arabica and Robusta, and the Robusta coffees are
marked in red. The barchart shows how many coffees of each type there are, and
that the great majority are Arabica. The parallel coordinate plot shows that there
are several variables on which the two types differ and that the differentiation is
particularly obvious on the variables Fat and Caffine [sic]. Finally, the scatterplot
of those two variables confirms the main conclusion drawn from the pcp and presents
it more definitively. In principle, all the information in the three graphics can be seen
in the pcp. In practice, the barchart and scatterplot are valuable for emphasising the
conclusions.
The code for Figure 12.1 is involved. Some of it is to specify the colours, some
to reduce the amount of labelling, and some to draft the layout. The most important
component of the pcp plot is the use of order to ensure that the lines for the group
of interest are plotted on top of the lines for the rest of the data. Note that the spatial
arrangement of the graphics and using a consistent style for them are just as impor-
tant as the drawing of the individual graphics. For presentation you have to think of
graphics as part of a story, not in isolation.

FIGURE 12.1: An ensemble of plots of the coffee dataset. Most coffees are of type
Arabica (barchart). The Robusta variety coffees, coloured red in all three plots, are
clearly different from the Arabica ones on two variables and fairly different on two
others (pcp). The Robusta coffees have high levels of Caffine [sic] and low levels
of Fat (scatterplot).
Ensemble Graphics and Case Studies 245

data(coffee, package="pgmm")
coffee <- within(coffee, Type <- ifelse(Variety==1,
"Arabica", "Robusta"))
names(coffee) <- abbreviate(names(coffee), 8)
a <- ggplot(coffee, aes(x=Type)) + geom_bar(aes(fill=Type)) +
scale_fill_manual(values = c("grey70", "red")) +
guides(fill=FALSE) + ylab("")
b <- ggplot(coffee, aes(x=Fat, y=Caffine, colour=Type)) +
geom_point(size=3) +
scale_colour_manual(values = c("grey70", "red"))
c <- ggparcoord(coffee[order(coffee$Type),], columns=3:14,
groupColumn="Type", scale="uniminmax",
mapping = aes(size = 1)) +
xlab("") + ylab("") +
theme(legend.position = "none") +
scale_colour_manual(values = c("grey","red")) +
theme(axis.ticks.y = element_blank(),
axis.text.y = element_blank())
grid.arrange(arrangeGrob(a, b, ncol=2, widths=c(1,2)),
c, nrow=2)

To get a feeling for an exploratory analysis, think about how you might take a
first look at the coffee dataset. Everyone will draw their own set of initial graphics,
possibly including some of the following. You have to open a new graphics device for
each new plot (e.g., with dev.new()) or earlier ones will just be overwritten. Not
all of these graphics are nice to look at, although they can be improved by resizing
the windows appropriately and you can then better gain information from them.
#Barcharts of the categorical variables
ggplot(coffee, aes(factor(Variety))) + geom_bar()
#Comment: More Arabica than Robusta
ggplot(coffee, aes(Country)) + geom_bar()
#Comment: Many different countries (with odd spellings)

#Checking for missings

visna(coffee) #Comment: There are no missings

#Mosaicplot of the categorical variables

mosaic(Variety ~ Country, data=coffee)
#Comment: Only one country has both varieties

#Distributions of the continuous variables

boxplot(scale(coffee[,3:14]))
#Comment: A few outliers, mainly symmetric, try histograms?

#Scatterplot matrix of the continuous variables

pairs(coffee[,3:14], pch=16)
#Comment: A few interesting patterns, too many plots
246 Graphical Data Analysis with R

12.2 What is an ensemble of graphics?

Many of the examples up till now have used single graphics, sometimes in conjunc-
tion with alternative graphics. It is overly ambitious to think you can discover or
represent all the interesting information in a dataset with one graphic, and for real
applications it is certain that you will have to draw a large number of them. Perhaps
everyone does this in practice, even if it is rarely described or discussed. In the dis-
tant past this may have been due to the difficulty of drawing graphics, when that had
to be done by hand, and in the more recent past to restrictions of space for print-
ing graphics in published work. Neither of these constraints exists today. Graphics
may be drawn quickly and easily, and printed publications may be supplemented by
material provided on supporting webpages.
Ensemble modelling methods have become increasingly used in statistics and
machine learning. Combining many models, sometimes quite simple ones, is often
more effective than a single complex model. The same applies to graphics. Ensem-
bles of graphics are groups of displays for presenting several different aspects of a
dataset simultaneously. They should be viewed as a group, with each contributing
something else to the overall picture. You might have
• Several plots of the same type for the same variable(s)
It is often informative to draw several histograms with different binwidths, for
example Figures 3.2 and 3.3. Groups, modes, and favoured values stand out with
some scalings more than with others.
• Several different plots for the same variable(s)
Boxplots, histograms and density estimates emphasise different features of the
data, for example Figures 3.11 and 3.12.
• Several plots of the same type for different subgroups (small multiples)
Trellis plots are the standard when the combined subgroups cover all the dataset,
e.g., Figures 10.12 and 10.14. Sets of subgroup plots are good for making com-
parisons.
• Several plots of the same type for different though comparable variables
If exam marks are available for a number of different subjects (as in the math-
marks dataset in SMPracticals), then a set of histograms, one for each exam, all
scaled the same, give a good overview and enable comparisons across subjects.
Plots of multiple time series also fall into this category when they are plotted
individually (as in Figure 11.7).
• Several plots of the same type for different variables
Figure 3.9 presents all the one-dimensional marginal distributions for the Boston
housing dataset. The distributional shapes give an initial impression of the types
of information the variables provide, some grouped, some with outliers, some
skewed. Scatterplot matrices are another example and can be seen in Figures 1.9
and 5.13.
Ensemble Graphics and Case Studies 247

• A variety of plots for different variables

Information derived from studying one set of variables may be checked by
looking at other variables. Sometimes the same types of displays are useful,
sometimes others are. The information you discover in different parts of the
dataset should be put together to construct an overall picture. Examples in-
clude Figure 9.6 and Figure 12.1. Many scientific reports use ensembles of plots
and you can find plenty of examples in issues of the weekly scientific jour-
nal Nature. For presentation purposes it can be helpful to combine plots to tell
a story. Some of the best information visualisation presentations use this ap-
proach, as you can see in the work done by the New York Times graphics group
[New York Times, 2011]. Some of these presentations are put together to form a
narrative flow, an attractive way of presenting information in a convincing man-
ner.
Combining graphics of different sizes and aspect ratios in one display needs plan-
ning. It is probably best to sketch what you want first and then choose suitable units
to match the layout. Histograms tend to be wide and not high, scatterplots square,
boxplots narrow and high, pcp’s short and very wide, and the form of barcharts de-
pends on the number of categories. R is flexible enough to allow any mixture, but the
success of the design depends on you.
The term ’ensemble of graphics’ implies a static collection of graphics put to-
gether to present a coherent story. In practice, with exploratory graphics, ensembles
are more a matter of keeping several balls in the air at once, keeping track of a range
of lines of thought in parallel as a graphical analysis proceeds. How is the feature
apparent in one graphic expressed in another? Are the changes in one variable con-
sistent with the changes in another? Is there an interesting feature, which needs to
be checked with a new display? Which features are most striking and should be fol-
lowed up? What additional graphics are needed to attempt to answer the new ques-
tions which have arisen? With the benefit of the knowledge you have discovered, it
can be simple enough to put together the graphics that reveal the information clearly
and convincingly, you just have to find the information first.
248 Graphical Data Analysis with R

12.3 Combining different views—a case study example

The Fertility dataset in the AER package concerns a study of women with at least two
children. It was a large study with some 250, 000 women involved. The interest was
in which mothers had more than two children. The explanatory variables provided
include the genders of the first two children, the mother’s age and race, and how
many weeks she worked in the previous year (1979).
Figure 12.2 shows an ensemble of plots for the dataset. The code makes use
of ggplot2 objects to simplify drawing several plots of similar type and uses the
gridArrange function from the gridExtra package to design the somewhat elab-
orate layout.
As expected, the numbers of women with two or more children rise with age,
although it is perhaps surprising that the pattern is so regular before dropping off
for the two oldest cohorts. The same pattern exists for all race groups (because of
overlaps there are six), as can be seen using

data(Fertility, package="AER")
ggplot(Fertility, aes(x=age)) + geom_bar(binwidth=1) +
facet_wrap(~ afam+hispanic+other, scales="free_y", ncol=8)

The function facet_wrap ignores any empty combinations while the option
scales="free_y" scales the y axis of each graph individually, so that distri-
butional shapes are compared rather than absolute frequencies.
That the majority of women either work or do not work is to be expected. Using
a function like

with(Fertility, prop.table(table(work)))}

shows that just over a third reported something in-between.

From the barcharts in Figure 12.2 you can see that most of the women had just
two children, that boys were as usual slightly more common than girls, and that most
women in the study were Caucasians.
The numbers of women in the study increase with age and so does the proportion
of them with more than two children. This can be seen in Figure 12.3. Interestingly,
more women have three children if the first two have the same sex and this is irre-
spective of whether the first two were boys or girls (Figure 12.4).
The Fertility dataset is very large and so it is possible to look at the combination
of the effects of age and the gender of the first two childen. Figure 12.5 shows that
the pattern is pretty much the same for all ages.
Ensemble Graphics and Case Studies 249

125000
30000

100000

20000 75000

50000
10000
25000

0 0
20 25 30 35 0 20 40
age Weeks worked in 1979

250000 250000 250000 250000 250000 250000

200000 200000 200000 200000 200000 200000

150000 150000 150000 150000 150000 150000

100000 100000 100000 100000 100000 100000

50000 50000 50000 50000 50000 50000

0 0 0 0 0 0
noyes female
male female
male noyes noyes noyes
has more children first child second child African!American Hispanic other race

FIGURE 12.2: Plots of the variables in the Fertility dataset. The numbers of women
with two or more children rose steadily with age, except for the eldest two years. The
majority of women worked every week of the year or not at all. Fewer women had
more than two children than had exactly two. There were slightly more boys than
girls for both the first and second child. Most women in the study were Caucasian.
data(Fertility, package="AER")
p0 <- ggplot(Fertility) + geom_bar(binwidth=1) + ylab("")
p1 <- p0 + aes(x=age)
p2 <- p0 + aes(x=work) + xlab("Weeks worked in 1979")
k <- ggplot(Fertility) + geom_bar() + ylab("") + ylim(0,250000)
p3 <- k + aes(x=morekids) + xlab("has more children")
p4 <- k + aes(x=gender1) + xlab("first child")
p5 <- k + aes(x=gender2) + xlab("second child")
p6 <- k + aes(x=afam) + xlab("African-American")
p7 <- k + aes(x=hispanic) + xlab("Hispanic")
p8 <- k + aes(x=other) + xlab("other race")
grid.arrange(arrangeGrob(p1, p2, ncol=2, widths=c(3,3)),
arrangeGrob(p3, p4, p5, p6, p7, p8, ncol=6),
nrow=2, heights=c(1.25,1))
250 Graphical Data Analysis with R

doubledecker(morekids ~ age, data = Fertility,

gp = gpar(fill = c("grey90", "green")),
spacing=spacing_equal(0))

morekids

no

yes

22232425 26 27 28 29 30 31 32 33 34 35 age

FIGURE 12.3: A spinogram of age with the proportion of women having more than
two children highlighted. This proportion increases with age.

doubledecker(morekids ~ gender1 + gender2, data = Fertility,

gp = gpar(fill = c("grey90", "green")))

morekids

no

yes

female male female male gender2

female male gender1

FIGURE 12.4: A doubledecker plot showing the numbers of women having two girls,
girl boy, boy girl, or two boys for their first two children. The proportion of women
having more than two children is highlighted. Mothers of same sex children are more
likely to have more than two.
doubledecker(morekids ~ age + gender1 + gender2,
data = Fertility,
gp = gpar(fill = c("grey90", "green")),
spacing=spacing_dimequal(c(0.1,0,0,0)))

morekids

no
Ensemble Graphics and Case Studies

yes

fmf fm
fmfmf mfemfemfemfemfemfem femfem fem fe m fe mafem
mafemmafemmafemmafemma fe m fe mafe m fem
ma gender2
fmf mfemfem fe m femmafemma femmal fem malefemamale fema male femalmale femamale femamale gender1
222324 25 26 27 28 29 30 31 32 33 34 35 age

FIGURE 12.5: A doubledecker plot by gender pattern for the first two children and age. The proportion of women having more than two
children is highlighted and you can see that the same pattern of higher rates for same sex families holds true for all ages.
251
252 Graphical Data Analysis with R

12.4 Case studies

The following case studies are for you to investigate, so you can test your graphical
skills. More information on the datasets involved can be found on the correspond-
ing help pages in R. Full information on how and why the data were collected is
sometimes provided. Ideally for any study there should be a clear set of goals and
close cooperation with subject area experts. Even if you have well-defined goals, it is
still sensible to do some exploring around. Sometimes issues of data quality become
apparent, sometimes a question about an odd minor feature leads to much more in-
formation being provided about other parts of the dataset. As Tukey famously put it,
data analysis is detective work, and we all know that apparently innocuous clues in
detective stories can lead to major discoveries.
Each case study begins with a group of direct questions about the dataset, which
should be relatively straightforward to answer. One or more open-ended questions
follow that will hopefully encourage you to see what else you can discover. Although
analyses should be carried out graphically, you can also think about how any conclu-
sions could be checked with models.

Moral statistics of France

The Guerry dataset in the package HistData includes a range of information for each
of the 86 departments of France around 1830.
1. There are two variables relating to donations, Donations (to the poor) and
Donation_clergy. Draw histograms of the two variables and comment on
the differences between them.
2. What would you expect the scatterplot of the last two variables, Prostitutes,
numbers of prostitutes in Paris by department of birth, and Distance, distance
from Paris, to look like? Does a scatterplot of the variables match your expecta-
tions?
3. Draw a boxplot of the variable Infants, which refers to the population per
illegitimate birth. Which of the departments marked as outliers do you think are
definitely outliers? If you invert the variable to get a measure of illegitimate births
by population and draw a boxplot, other departments are identified as possible
outliers. Which of these do you think are definitely outliers? Which variable
would you judge to be more appropriate to use?
4. Some of the variables are expressed as inverses of rates, so that higher is better.
Quite a few variables are only given as ranks, so that lower is better and there
is only ordering information, but no distributional information. Draw a parallel
coordinate plot of the rank variables and select the departments with big cities
(category 3:Lg in the variables MainCity). How do these departments differ
from the others, if at all?
Ensemble Graphics and Case Studies 253

5. What additional features can you discover in the dataset?

Airbags and car accidents

There are 26217 cases of police-reported car crashes in the nassCDS dataset in the
package DAAG.

1. Draw a histogram of the variable weight. The R help file for the dataset says
that the observation weights are ”of uncertain accuracy”. Is there any evidence
of this? What graphics would you draw to investigate which cases have high
weights and which have very low weights?
2. How does the availability of airbags depend on the age of the vehicle?
3. How does death rate depend on vehicle speed?
4. How does death rate vary with the variables seatbelt, airbag, deploy,
and frontal? Which orderings of the variables and of the categories within
the variables give the most convincing graphic?
5. Are there other interesting patterns in the data worth presenting?

Athletes’ blood measurements

The ais dataset in the package DAAG contains five blood measurements and six
physical measurements for 202 male and female athletes.

1. Draw histograms of the eleven variables. Are there many different forms or are
they all roughly normal?

2. Boxplots of the variables suggest that there are a number of outliers. How many
of these are outliers on more than one variable?
3. Draw a parallel coordinate plot of the eleven variables with the females coloured
red. Do the females have lower values than the males on all the variables? Which
variables look most promising for distinguishing between males and females?
4. There are ten different sports listed, nine for the women and eight for the men.
Is there anything particularly distinctive about the patterns for any of the sports,
either for both genders together or singly?
5. If you were to choose three graphics to represent the most interesting information
in this dataset, which would you choose and how would you present them in a
summary report?
254 Graphical Data Analysis with R

Marijuana arrests
The Arrests dataset in the package effects contains information on over 5000 arrests
on the streets of Toronto over five years. There are six variables describing the ar-
rested person, a variable giving the year, and a binary variable released saying
whether the person was released with a summons or not.
1. Are there any features worth mentioning in the distributions of the individual
variables? Which graphics show these features clearly?
2. How does the variable released depend on the others, taking them one at a
time? Which graphics would you select to show this?

3. What about the dependence of released on the variables colour,

employed, and citizen in combination? Which plot would you draw to in-
vestigate this?
4. Are there other features of the dataset worth reporting?
5. If you wanted to summarise the information you have found on one page, which
graphics would you choose and how would you present them as a group?

Crohn’s disease
Crohn’s disease is a chronic inflammatory disease of the intestines. The package
robustbase includes a small dataset CrohnD of 117 patients in three groups, one
placebo group and two different drug groups.
1. Draw plots to show there are equal numbers in the three treatment groups, but
not equal numbers by gender or country.
2. Which plot would you draw to show the counts for all gender, country, and treat-
ment combinations?

3. The key outcome variable is the number of adverse events, nrAdvE. What
graphic would you choose to display its distribution? What about a set of three
graphics, comparing the distributions of the numbers of adverse events for the
three treatment groups?
4. Draw histograms of the other five variables, including ID. Explain why you
chose the binwidths you use and describe what information you can see in the
displays.
5. If you had to write a short report on your analysis, what would you include?
(There is some interesting background information in [Brant and Nguyen, 2008]
and there are plenty of other sources of information you might find useful.)
Ensemble Graphics and Case Studies 255

Footballers in the four major European leagues

Statistics has even reached soccer nowadays and the dataset EURO4PlayerSkillsSep11
in the package SportsAnalytics includes 43 pieces of information on 1851 soccer
players in September 2011.
1. Draw a scatterplot of the variables Attack and Defence. Are there any obvi-
ous outliers? What might be done about them?
2. There is a variable called Position saying whether players are goalkeepers,
defenders, midfielders, or forwards. Which group of players scores highest on
the variable Agression?
3. Draw a graphic showing the relationship between the variables Foot and Side.
What are your conclusions? Do your conclusions vary depending on a player’s
position?
4. How would you display graphically how goalkeepers differ from other players?
5. Consider only the subset of players who are goalkeepers. What sort of values
are they given for variables such as the four concerned with passing? Do these
distributions depend on the league in any way?
6. The dataset has many variables and so there is much that might be investigated.
Choose one of the four leagues and investigate if there are any noteworthy dif-
ferences between the teams. If teams are distinctive in some way, check whether
they are also distinctive if you take the other leagues into account as well. (The
data are provided to support a soccer video game and the way the players are
evaluated is not clearly documented. This may account for some of the minor
features you can find.)

Decathlon
Both from a sporting point of view and from a statistical point of view, the decathlon
is a fascinating event. Athletes must compete in ten different disciplines over two
days. Their performances in each event are transformed to a common scale and
summed to produce a total score.
The transformation formulae [IAAF, 2001] were last changed in 1985 by a com-
mittee including the academic and former decathlete, Viktor Trkal. That their recom-
mendations have continued to be used for so many years is a tribute to their work.
Viktor Trkal [personal communication] reported that the committee used principles
they believed a satisfactory system should follow and data provided by various coun-
tries on decathlon performances to guide their work. One principle that was not re-
spected by the previous system was that an improvement in performance should be
worth more points the higher the level that is improved on.
There is an excellent webpage for decathlon results, based in Estonia, where
individual scores and much additional information can be found for performances
over many years [Salmistu, 2013]. The dataset used here, the Decathlon dataset in
256 Graphical Data Analysis with R

the package GDAdata, is a subset of that data for the years 1985 to 2006 covering
almost 8000 performances. The criteria for a score being recorded is that the total
points should be higher than 6800 and that it should be the best performance of an
athlete in a particular year. This means that the top decathletes can only appear at
most once a year, although they can and do appear in several different years.
1. What plots would you choose to study the variable Totalpoints and how
would you describe the distribution?
2. Use boxplots to examine how the distribution of Totalpoints develops over
time. What conclusions would you draw?
3. Draw a scatterplot of the variables m100 and P100m, the actual times in seconds,
and the points awarded for the 100 metres. How close is the formula to a linear
function in the range covered by the data? Does the same apply to the pole vault
event?
4. Draw parallel coordinate plots of the raw data (m100 to m1500) and the points
data (P100m to P1500). Draw boxplots for the points variables. What infor-
mation can you see in each plot? Which plot is most useful for comparing the
performances in the different disciplines?
5. Pcp’s of the points variables on a common scale could be sorted by the maximum
values, the medians, the IQR’s, or by those statistics for a selection, such as
for the most recent year. Choose any two you find interesting and explain your
choice.
6. For the second question you looked at the development of the total points scored
over the 22 years. What about the development of scores for the individual dis-
ciplines? You might compare the maximum, the median, the 95% quantile, the
tenth best performance or the twenty-fifth best performance for each year. Which
do you think would be suitable? Create the necessary dataset to plot the statistics
you chose over time. What conclusions do you draw from your plot?

Intermission
Rembrandt’s The Night Watch hangs in the Rijksmuseum in Amsterdam. How might
you divide it up into individual groups and scenes? How well do individual scenes
combine to form the whole?
13
Some Notes on Graphics with R

Those are my principles, and if you don’t like them. . . well, I have others.

Groucho Marx

Summary
Chapter 13 discusses using R for graphics and passes on some individual pieces of
information that have been helpful for examples in the book.

13.1 Graphics systems in R

There are many different ways of achieving the same goals in R and you have to
decide for yourself which works best for you. This may depend on flexibility, speed,
elegance, or ease of use. The standard R graphics are called base graphics and were
written to match the graphics available in S. They offer extensive control over the
look of the displays. This makes them very flexible and suitable for presentation pur-
poses. S graphics were advanced for their time, but computing has moved on consid-
erably since then in both hardware and software terms. An alternative graphics sys-
tem, grid graphics, implemented in the package grid, is described in [Murrell, 2011].
This provides a systematic underlying structure and increases user control further.
Both of the popular packages ggplot2 and lattice make use of grid.
R is like a spoken language, in that there are a huge number of ideas you can
express and there are multiple ways of doing that. In graphics terms this means
that there are many different displays available and often several alternative ways
of drawing them. For graphics examples you can peruse the R Graphical Manual
(rgm3.lab.nig.ac.jp/RGM/), which collects R graphics that developers and users have
drawn and the code written to produce them. As for all the different ways of drawing
particular plots, just think of the various histogram and scatterplot matrix possibili-
ties. There is a task view for graphics, [Lewin-Koh, 2013], which lists many packages
and gives a helpful overview. In this book the package ggplot2 has mostly been used.

257
258 Graphical Data Analysis with R

13.2 Loading datasets and packages for graphical analysis

Some datasets are supplied in the package datasets and are directly available. Others
are supplied as parts of packages and you have to have the package installed to get
the dataset. The dataset can then usually be loaded using

data(datasetName, package = "package_name")

Naming the package ensures that you get the dataset you want. The same dataset
can be contained in more than one package in different formats and yet the iden-
tical name. And in some cases quite different datasets have identical names in dif-
ferent packages (e.g., movies and barley). Note that it is not necessary to load a
package with library(package name) to make a dataset in the package avail-
able. Loading a package unnecessarily can have the potentially irritating side-effect
of having functions unintentionally (and confusingly) overwritten.
In a similar way, some functions in different packages have the same name, al-
though they do quite different things. To avoid loading a complete package you can
use package::function. This was done in the code for Figure 9.1, since the
function was only needed for that single example. Had the package mi been loaded,
it would have automatically loaded arm, which in its turn would have loaded MASS.
There could then be a confusion between the functions select in MASS and in
dplyr. This kind of problem is becoming less frequent as packages ’import’ other
packages rather than ’depend’ on them.
Graphics may require the correction of values in datasets, the calculation of new
variables, or a restructuring of the data. R can handle as many datasets simultane-
ously as you wish, so it is essential to keep track of a dataset as changes are made.
The best approach is to make all changes in R code and give the resulting dataset(s)
new name(s). It is best to keep close tabs on which dataset or subset you are currently
working with.
The examples in this book have been run using the package versions available at
the end of 2014. Code chunks affected by subsequent changes in packages will be
updated accordingly on the book’s webpage, rosuda.org/GDA.

13.3 Graphics conventions in statistics

In drawing graphics it is important to be aware of the conventions and standards that
are used. Not all will agree on these and additionally, as the classical phrase elegantly
expresses it, the exception proves the rule. Nevertheless, most of the following are
generally accepted and have been used without comment in the book. If something
about a graphic surprises you, then perhaps you will find the explanation here.
Some Notes on Graphics with R 259

1. In a scatterplot of causally related variables the dependent variable is drawn on

the vertical axis and the explanatory variable on the horizontal axis.
2. Numbers increase to the right and up.
3. The x-axis crosses the y-axis at y = 0. When it doesn’t, this should be made clear.
4. Scales are linear, and when they are not this should be emphasised.
5. Time is usually drawn on the horizontal axis, progressing from left to right.
6. Graphics are always drawn to show all the data. If some cases are out of range,
this should be explicitly stated.
7. Aspect ratios (the relationship of the height of a graphic to its width) should be
chosen to make the slope of lines of interest about 45°, an idea first discussed
thoroughly in [Cleveland et al., 1988].
8. Points usually represent individual cases and areas represent counts or weights.
9. Vertical bars represent frequencies of continuous variables when there are no
gaps between them, and frequencies of categorical variables when there are.
10. Distinct colours are used to represent groups, while shading or continuous spec-
tra are used to represent scales.

13.4 What is a graphic anyway?

In this book graphics are visual displays of data, summarising information about
datasets. A graphic may show information about only one variable or about several
and groups of simple graphics together are often more effective than a single complex
graphic. Some graphics are easier to interpret and understand than others.
Most people are more ready to look at a graphic and try to understand it than
to look at numbers or statistical output and try to understand that. However, it is
probably also true that while people will tell you all too willingly that they do not
understand statistics and cannot do simple maths, they will not tell you that they do
not understand a graphic. Alan Fletcher makes a similar point in his stimulating book,
“The Art of Looking Sideways” [Fletcher, 2001], where he remarks that people will
say they have no ear for music, but will never admit to not being able to appreciate
paintings. It is always a good idea to discuss graphics with others and to find out
what they see and you do not—and vice versa.
There are many different graphic types used in data analysis: histograms, bar-
charts, dotplots, boxplots, scatterplots, and so on. Each type represents the data for
one or more variables from a dataset in a graphical form. Sometimes each case is
represented individually (as in a scatterplot), sometimes areas represent groups of
cases (as in a barchart or a histogram).
260 Graphical Data Analysis with R

When interpreting any graphic you have to make sure you understand how the
plot has been defined and which data are shown. In general the plot types have stan-
dard definitions, which everyone follows, even if there are some exceptions. Several
different definitions of boxplots are in use and however much we may feel that the
classic Tukey definition is the one to use, we have to be aware that others may think
otherwise.
It is tempting to think of graphics as just their graphics part, but there is much
more to them than that. There are several components to a graphic above and beyond
the graphical form itself:
Title There could be a title describing the graphic or directing your attention to some
information in the graphic.
Caption The caption should explain what is shown, possibly also giving the data
source. Captions should be detailed enough that the graphic can pretty well stand
on its own. Longer is usually better than shorter. A picture may be worth a thou-
sand words, but you need at least some words to describe and explain it.
Labels Well-labelled axes with understandable variable names (and, where appro-
priate, the units the data are measured in) make it easier to concentrate on the
graphic, as there is no need to search for the information elsewhere.
Scales Uncluttered scales with nicely chosen numerical divisions that have mean-
ing for the data help readers understand the orders of magnitude of the data on
display.
Legend If there are different groups in the display with different colours or shapes,
then a legend is valuable for defining them.
Annotations If a particular feature is to be highlighted, such as an outlier or a gap
in the data, then an annotation can be added on the plot itself.
Accompanying text Graphics should always be discussed in the accompanying text
or at the very least referred to. This provides an opportunity to comment on
particular features in more detail and to add supplementary details. Ideally this
text and the graphic should be close together. (Making sure this happens in a
book is more difficult than you would think!)
Some of this applies more to presentation graphics than to exploratory graphics. Ana-
lysts exploring data will have much of the information at their fingertips and not need
these details, they are looking for overall structure not exact numbers. Where the ad-
ditional components do become relevant is when you discover an interesting piece
of information and want to save it. The swift graphic that gave you the idea to-day
may look a bit of a mystery in two weeks’ time. Information that has been unearthed
should be cleaned up and summarised and proper presentation graphics prepared,
including most, if not all, of the components listed above. References to books with
advice on drawing presentation graphics are given in §2.1, while [Unwin, 2008] is
one of many articles that attempt to provide a summary of pertinent advice.
Some Notes on Graphics with R 261

13.5 Options for all graphics

There is an astonishing number of different types of graphics you can draw for dis-
playing data. Each also offers a broad range of individual options that can be chosen
to modify its display even more. Amongst all this variety, there are options that apply
equally to every graphic that should always be kept in mind.

Window size and shape

Size Plots look different at different sizes. Very small plots can be too cramped to
be easy to see and extremely large plots can be difficult to digest. Often the same
display leaves a different impression at different sizes. When comparing graphics
of the same type it is imperative to plot them in the same size of display.
With presentation graphics there can be restrictions because of page size for
printing or because of webpage structure if the graphics are displayed on the
web. With exploratory graphics there is no reason for not experimenting with a
range of sizes to ensure that you can find the best view for a plot. The sizes of
displays in this book have been chosen to convey the information in the plots,
while attempting to ensure that any text associated with the plots is on the same
or the facing page. They have also been restricted by the size of the pages.
Aspect ratio Plots look more polished when they have a sensible aspect ratio, the
ratio of the height to the width of a plot. They are also easier to interpret. His-
tograms which are broad and flat look squashed as if someone has trodden on
them. If they are tall and thin, it is surprisingly difficult to judge the distributional
shape. Both are actually unusual in practice, so try it yourself. If the histogram
looks good to you in a default window, try it in a window which is twice as wide,
but half as high, and in a window which is half as wide, but twice as high. One
is too wide, one is too tall, and one may be just about right.
With boxplots it is another matter. One of their main advantages is their efficient
use of space, a boxplot should be narrow and tall. Unfortunately many users of R
seem to just use the default window shape, which results in wide, squat boxplots.
Everyone is entitled to draw whatever plots they like in the privacy of their own
analyses, but it is extraordinary how many of these boxes make it into print. Two
wide boxplots side by side are not as easy to compare as two of appropriate
width and they just do not look nice. A satisfactory solution is often to plot them
horizontally rather than vertically.
With scatterplots it is usually best to use a square display, although sometimes
other forms are better. If you crush one of the axes by reducing either the horizon-
tal or vertical axis to almost nothing, you almost get a one-dimensional dotplot.
Occasionally that can be useful (and quicker than drawing a new dotplot). A
fine graphical example of the importance of getting the aspect ratio right can
be found in a Doonesbury cartoon (http://doonesbury.slate.com/
strip/archive/2013/09/10).
262 Graphical Data Analysis with R

Scales
Scaling Designing scales is difficult. Ideally you want sufficient discrimination, but
not too much. You want to cover the full range of the data, and use rounded inter-
pretable values. R uses the pretty algorithm designed for S [Becker et al., 1988]
and in general it works impressively well and you can usually rely on the default
choices. Several other algorithms have been suggested and can be found in the
package labeling.
The way the pretty algorithm is implemented in base graphics has a curious side-
effect, which you might call “walking the plank”. To avoid drawing labels outside
the range of the data (thereby increasing the size of the plot), while at the same
time keeping ‘pretty’ values for the labels, the highest and lowest labels may not
cover the full data range. This is a good idea, but what is not such a good idea is
that the axes are only drawn as far as the final labels. Histogram bars to the far
right can look as if they are falling off the edge or “walking the plank”.
Common scaling for comparisons If two or more graphics are to be compared, it
is essential to use common scaling, i.e., the same scaling for all. This includes
the window size being the same if the plots are in different windows. Ensuring
that windows are the same size can be achieved by using the default window
size or by drawing new windows of a specific size using the height and width
parameters. And, naturally, you should align the plots so that the comparisons
you want to make are direct. Alignment is easier if carried out by R for you within
a window, rather than trying to align separate plot windows by eye yourself.
Ensuring that scales are the same needs some preparatory work to check the
value ranges for the axes, either by carrying out some initial test plotting or by
doing the necessary calculations. For count and density scales, the vertical axes,
the range needed can depend on binwidths (histograms) or bandwidths (density
estimates). Precalculation of plot objects may be necessary. For categorical vari-
ables it means ensuring that the categories are in the same order in each plot and
possibly adding one or more categories with zero counts to some of the variables
to be compared. Test plotting is quick and can be informative in other ways, so
that fits in with an exploratory approach. Calculations are useful for presentation
graphics.

Scale limits The same plots look different with different scale limits. Sometimes
there is good reason to include zero, even though it may be far from the data,
sometimes not. Individual outliers can distort scales, leading to a lot of empty
space and crowding most of the data into a small area.
Plots may be drawn in a box without a horizontal axis at zero. This can be mis-
leading if readers think the lower boundary of the box is the zero line. It is advis-
able in such cases to add a faint grid line at zero.
Some Notes on Graphics with R 263

Text
Labelling (abbreviations) Variable and category names can be informative and
quite long. The need for Fortran’s restriction to variable names that had at most
six characters belongs to the distant past. This means that labelling for some
plots suffers from overlapping text, as there is no way you can fit exorbitantly
long names together for a barchart with many categories or for a parallel coordi-
nate plot with many variables.
There are two solutions. The first is to create new variable names or new cate-
gory names with appropriate abbreviations yourself. In both cases it is sensible
to make a copy of the original dataset and make the changes in the copy. The sec-
ond solution is to use R’s abbreviate function. This works from left to right
removing spaces and vowels until it reaches the allowed length, providing that no
duplicates result. Again it is best to create a new dataset with the abbreviations.
abbreviate is a considerable help, but does not work well on non-English
names. Some manual adjusting afterwards may be worthwhile.
Font sizes The way the labellings of plots and other graphics texts look on screen
and the way they look in print can be quite different. A lot depends on the original
size of the plot and how it is changed for the printed page. This is something that
is important for presentation graphics. It is at worst an irritant for exploratory
graphics if the labels are too small to read or if they are so large that they overlap
with each other or interfere with the rest of the graphic.
Annotations With presentation graphics it is sometimes a good idea to explicitly
draw attention to a particular feature of a plot and use some form of annotation,
adding some text and perhaps an arrow pointing to the feature. For best results it
may be necessary to perfect the plot with specialist graphics software. Otherwise
you can use some of the functions in base graphics, annotate in ggplot2 or
write your own function with grid. For exploratory graphics annotation is not
necessary.

Colour and appearance

Colour Colour is a difficult topic for a wide variety of reasons. Fortunately there
are several R packages that offer help. Two in particular are worth mention-
ing. RColorBrewer provides the colour palettes recommended by ColorBrewer
[Brewer, 2013] for thematic maps. With colorspace you can map between differ-
ent colour spaces including the recommended HCL [Zeileis et al., 2009]. There
is also a good package vignette for colorspace. If colour is used, it is obviously
best if possible colour blindness of viewers of the graphics is taken into account.
Unfortunately default colour selections for some packages could be better, so
you have to check.
One of the palettes used in this book is the colour blind one from the pack-
age ggthemes and another is from the ggplot2 package based on palettes rec-
264 Graphical Data Analysis with R

ommended by ColorBrewer [Brewer, 2013]. Palettes can be set at the start of a

session. If you use R’s default colour settings your displays will look different.
Colour is a powerful tool that should be used with care. Robert Simmon has
written an attractive series of short web essays called “The Subtleties of Color”
[Simmon, 2014], which offers some thought-provoking ideas on the topic. The
best colour graphics are informative and tasteful. Anything less than the best can
look kludgy and even be misleading (and you may well feel that this applies to
some examples in the book).
‘Themes’ Not all plots look the same in this book, as they have been drawn with
different R functions and packages. If you stick to one package, such as ggplot2,
you can get a consistent look to your plots. You can specify ‘themes’ to ensure
that the non-data parts of your plots are all in the same style and colour. I included
the colourblind palette from ggthemes as part of the grey theme and also made
the plot.background the same grey as the panel.background.
You have to specify the data parts of the plots separately, either with a general
function at the beginning of a session or for each plot individually. For instance,
I find the default black bars for barcharts and histograms too strong and set the
default bar colour at the start of a session to achieve a standard grey colour, while
ensuring that the borders of the bars can still be seen.

update_geom_defaults("bar",
list(fill = I("grey70"), colour = I("grey40")))

13.6 Some R graphics advice and coding tips

A number of issues come up again and again in superficially different, but actually
similar situations. The same principles generally apply, and this section comments
on a few of them.

To get a new graphics window

(This assumes you are working with software which permits more than one graph-
ics window. The popular RStudio interface basically only allows one.) The function
dev.new() opens a default window whose size will depend on the R you are us-
ing. The units used, pixels, inches, or cms, can depend on your settings as well. When
you want a bigger window than the default, or one with a particular aspect ratio, then
specify the height and width:

dev.new(height=900, width=800)

Giving an exact size is useful when graphics suffer from overplotting in a default
Some Notes on Graphics with R 265

window, which can often happen with multivariate graphics, when you want to ensure
that all windows in a group have the same non-default size, or when you are preparing
graphics for a report or presentation.

Resizing windows
Sometimes the default window size is inappropriate for a graphic (e.g., text which
overlaps in a small window will be fine in a larger one) or the aspect ratio makes the
plot look wrong (e.g., overly wide boxplots). Rather than trying to predict the right
height and width in advance it is simpler to just drag the window and choose the best
size and aspect ratio by eye. You can then find out the size of the active window using
dev.size(”px”) or dev.size(”in”).

Default plots
One of R’s strengths is the broad range of control options that most functions give
users. It must be off-putting for beginners to look at the help for drawing a histogram,
surely a simple task, and to wonder what on earth they have got themselves into,
facing such a long list of options. Fortunately the complementary strength is that R’s
defaults are generally good. You can use hist(x) and get a reasonable plot and
plot(foo) is useful for a wide range of objects foo, even if disappointing for a
single numeric vector and not often helpful for a dataset. Not all of the defaults are
as good as they could be, but that is hardly an issue when it is relatively easy to fix it
yourself. You do not need to beware of defaults, but you do need to be wary of them.

Points in scatterplots or in other point plots

Scatterplot points are plotted as open circles by default in base graphics. For small
datasets this is a matter of taste, for larger datasets it can look confusing (opinions
may differ). Using one of the options pch=16,19,20 in the plot function gives
filled circles.
When scatterplots suffer from overplotting because of multiple cases with exactly
the same values, various options have been suggested: special symbols to represent
the numbers of cases as in sunflower plots (this is only reasonable for medium-sized
datasets and the resulting graphics are still difficult to decode); jittering where small
random values are added to the coordinates of the points to separate them (this can
work, again only for datasets that are not really large); alpha-blending (where the
shading of the points depends on how many there are); and hexagonal binning with
shading. Zooming in to areas of particular interest can be valuable too. No approach
is perfect and it is advisable to look at a range of plots and consider incorporating
density estimates.
266 Graphical Data Analysis with R

Printing graphics
The way graphics look on screen and the way they look printed from a saved file
may be rather different, so you need to check. This can apply to colour, fonts, and
spacing.

Multiple windows
Unlike some other systems, R requires you to open a new graphics window every
time you draw a new plot, if you want to still keep existing plots. Its default be-
haviour is to draw a new graphic in the currently active graphics window. This can
be irritating, and you have to get used to it. It is usually safest to always draw a
new window, as you can lose track of which is the current active window. The code
dev.new() has been taken for granted in much of the book.
This works differently for the RStudio interface, where you can use x11() in-
stead, if you have access to an X Server, but that does take you out of RStudio.
You then have some plots in X11 and some in RStudio. Within RStudio you do not
overwrite old plots when you draw new ones, which is good, but it does not encour-
age you to see more than one plot at once, which is not so good. You can always flip
through previously drawn graphics or, with sufficient advance planning, draw several
in the same window.

Drawing several independent plots in one window

To draw a table of plots in base graphics, use par(mfrow=c(n,m)), which splits
up the window into n rows and m columns. More complex layouts can be produced
using layout. If you use a package based on grid, producing grobs, as ggplot2
does, you can use the function grid.arrange from the gridExtra package to get
a row and column layout. For more flexibility, you can combine arrangeGrob
with grid.arrange, as was shown in Figure 12.2.
Figure 13.1 gives a simple example with two smaller plots above and one bigger
plot below. First the plots are prepared as individual grobs and then the window to
display the plots is set up. Observe how the grobs are nested and how the bigger plot
is allocated more space.
If you want to use some of the useful graphics for categorical data in vcd you
face the additional problem that while vcd is based on grid (like ggplot2 and lat-
tice), it uses an earlier version and does not produce the appropriate graphics objects
(grobs). Even combining only vcd plots is hard work.
The two graphics systems in R, graphics and grid, do not mix well. There is
a package gridBase, which in the words of its author, Paul Murrell, “can be used,
in some situations, with a little care, to overcome this inherent incompatibility”
[Murrell, 2011].
Some Notes on Graphics with R 267

12000

30000
9000

20000
6000

10000 3000

0 0
0.0 2.5 5.0 0 1 2 3 4 5
carat carat

2000

1000

0
0 1 2 3 4 5
carat

FIGURE 13.1: Three histograms of the variable carat from the diamonds dataset
drawn together with different binwidths. The first plot suggests a coarse skew distri-
bution. The second plot suggests there may be some favoured values. The third plot
reveals the curious pattern of a number of skew distributions aligned together one
after the other. The explanation is probably that rounded values of carat like 0.3, 0.5,
1, 1.2 are favoured with no diamonds of slightly smaller sizes being offered for sale
and progressively fewer larger ones till the next rounded value is reached.
p0 <- ggplot(diamonds, aes(x=carat)) + ylab("")
p1 <- p0 + geom_bar(binwidth=1)
p2 <- p0 + geom_bar(binwidth=0.1)
p3 <- p0 + geom_bar(binwidth=0.01)
grid.arrange(arrangeGrob(p1, p2, ncol=2), p3, nrow=2)
268 Graphical Data Analysis with R

Naming objects
• Give new objects new names
If you drop, combine, or reorder levels in a categorical variable, it is best to cre-
ate new variables with new names. Otherwise simple mistakes can lead to nasty
errors. This applies to new objects of all kinds, but it is particularly easy to run
into trouble with levels.

• Naming new variables

Keep new variables associated with the data frame they are derived from.

iris <- within(iris, area <- Petal.Width*Petal.Length)

will modify the existing data frame, iris, by adding an additional column, area,
yielding a new version, also called iris, in the global environment. (Note that you
must use the <- operator for assignment here; you cannot use =.) Whereas

area <- with(iris, area <- Petal.Width*Petal.Length)

will generate a separate new variable, area, in the global environment using the
columns of the data frame iris.

• Give plot objects names

Functions like hist(x) plot the histogram in the current active window and
h1 <- hist(x); plot(h1) does the same, but now you can access some
of the components, e.g., h1$counts. Unfortunately, you cannot always access
all you want. hist does not give you the binwidth directly, you have to calculate
it from with(h1, breaks[2]-breaks[1]). This assumes that you know
the bins are of equal size (highly recommended, only mathematical statisticians
think of non-equal binsizes and they don’t analyse real data very often) or that
you have checked if h1$equidist is true. At a pinch you could use

binw <- with(h1, if(equidist) breaks[2]-breaks[1])

Naming density estimates is particularly useful for getting at the bandwidth used.

Reordering categories for a barchart (and ordering in general)

The default ordering of categories is alphabetical. By redefining the levels attribute
of a categorical variable you can change the order. To avoid confusion (and the pos-
sibility of mistakes) it is best to define a new variable. Here is a reordering of the car
type variable by mean weight for the Cars93 dataset in the MASS package:

data(Cars93, package="MASS")
Cars93 <- within(Cars93, TypeWt <- reorder(Type, Weight, mean))
Some Notes on Graphics with R 269

You could achieve the same ordering by specifying the new variable’s levels directly

Cars93 <- within(Cars93,

Type1 <- factor(Type, levels=c("Small", "Sporty",
"Compact", "Midsize", "Large", "Van")))

Compare the level orderings with

with(Cars93, table(TypeWt, Type1))

The value of choosing a good ordering can be seen in plots like Figure 13.2.

ggplot(Cars93, aes(TypeWt,100/MPG.city)) + geom_boxplot() +

ylab("Gallons per 100 miles") + xlab("Car type")

6
Gallons per 100 miles

2
Small Sporty Compact Midsize Large Van
Car type

FIGURE 13.2: Fuel consumption by type of car ordered by average car weight. (It
would look even better if the Sporty and Compact groups were switched.)

The replacement function levels can also be used to combine categories by re-
peating a value.

Cars93 <- within(Cars93, {

Type3 <- Type
levels(Type3) <- c("Small", "Large", "Midsize",
"Small", "Sporty", "Large")
})
270 Graphical Data Analysis with R

Ordering can be useful for sorting cases (which can be important in overlaying one
plot on another as in Figures 6.3 and 6.11), for sorting categories (for instance in the
bottom two plots in Figure 4.1 or in the doubledecker plot in Figure 7.9), for sorting
variables (as in Figures 6.11 and 6.15, cf. §6.7), and in drawing conditional plots like
trellis displays.
The important functions are order for producing an ordered vector of case in-
dices and reorder for ordering the levels of a factor. The functions sort and
rank apply to single vectors. With sort you can order the vector and with rank
get the ranks.

Reshaping datasets and graphics

Datasets frequently have to be restructured for particular graphics. As usual with R
there are often several ways to achieve the same effects. In this book the packages
reshape2, tidyr, and dplyr have been used, and they do require a certain amount
of study to understand how they work. Both reshape2 with its melting and casting
functions and tidyr with its related gathering and spreading functions seem to need
particular effort, but are well worth it. It must be pointed out, however, that it seems
likely that the newer package tidyr will supersede reshape2. One big advantage of
this group of packages is the ability to chain operations together (as was done, for
instance, in the preparatory code for Figure 9.6).
Graphics themselves can be manipulated in several ways and facetting is a valu-
able general option for constructing small multiples of ggplot objects. The option
coord_flip() is a handy way of transposing a ggplot display.

Missing values
Functions do not all deal with missing values the same way. For instance, compare
what you get with mean(X1) to what you get for the mean with summary(X1),
when X1 is a variable with at least one missing value. If a plot does not work as
you expect, then missing values could be the reason and you may have to set a plot
function option or restructure your dataset accordingly.

Using the code and finding out about function options

This book is about the why of graphics not the how. There are a number of fine
publications on how to draw graphics with R, so the emphasis here is different and
there are occasionally fewer explanations of code details than some readers might
like. If you need more information, R’s help pages for functions are a good first place
to look, although some are much better than others. Sometimes you find explanations
and examples of a function’s use covering the options available in comprehensive
detail, sometimes not. Some packages have excellent vignettes, others not. If you
need additional assistance, then the next step would be to type your query into your
favourite search engine and that may well supply a solution to your problem.
Some Notes on Graphics with R 271

13.7 Other graphics

This book has discussed analysing data graphically with some standard types of
graphic. Perhaps it could have been called “The Seven Basic Plots”, except that ti-
tle has already been taken for discussing different kinds of plots by another book
[Booker, 2004]. It is important to know the graphics you use well, so that you have
experience in interpreting and understanding what they show. That is why the book
has concentrated on only a few graphic types. The range of different patterns that can
arise with them is surprising, and graphics can be very informative when you know
what patterns to look for and what features might be revealed.
Of course, there are many other graphics, which you can consider and which
might be especially useful for revealing particular information in certain datasets.
Each of us may favour specific graphics forms over others but we should all be pre-
pared to acknowledge that it is the information revealed, which is most important,
not the graphic used.
The same principles apply to working with these additional graphics as have been
discussed throughout the book. Amongst other things you should try many different
variations of the plots, both in the options and the formatting, use more than one
plot, ensure you have made informative comparisons—and not expect to find a single
‘best’ display. And if you present your plot to others bear in mind that they may not
be as familiar with that type of graphic as you are.
Most of the graphics can be found in one or more packages in R in one form or an-
other, and the Graphics Task View offers a convenient overview [Lewin-Koh, 2013].
You might draw: added variable plots, association plots, bagplots, Bangdiwala’s
agreement plots, barnest graphics, beanplots, beeswarm plots, bikini charts, biplots,
Bland-Altman plots, bubble plots, bumps charts, centipede plots, coplots, corrgrams,
cotab plots, coxcombs, dendrograms, Dickens’ plots, donut plots, downfall plots, fan
plots, forest plots, fourfold plots, funnel plots, hanging rootograms, hive plots, ici-
cle plots, intersection plots, Kaplan-Meier plots, kite charts, L’Abbe plots, ladder
plots, mountain plots, navel charts, Ord plots, pictograms, polarAnnulus plots, pro-
file plots, pyramid plots, radar plots, ring diagrams, rose diagrams, Sankey diagrams,
scree plots, sieve diagrams, silhouette plots, skyline plots, spaghetti plots, sparklines,
speedometer charts, spider plots, spiecharts, stream graphs, sunflower plots, Taylor
diagrams, ternary diagrams, tetris plots, thickening plots, tile plots, towel plots, trace
plots, treemaps, triplets, violin plots, waffle plots, waterfall plots, ... to name but a
few.
Apologies if your favourite plot is not included, no slight is intended. Apologies
also that a couple of names that are not plots at all have also sneaked into the list, but
now I can’t remember which ones they are...
272 Graphical Data Analysis with R

13.8 Large datasets

‘Big Data’ is a much discussed term nowadays and refers to datasets that are too large
for common or garden software, whatever that implies at the time of the discussion
and depending on what you want to do with the data. Some methods of analysis will
work for pretty well any size of dataset and others are much more limited—just think
of hierarchical clustering for example.
Even large datasets that can quite comfortably be analysed on a modern laptop
bring special complications with them. From a statistical point of view, there are two
problem areas. On the one hand, large numbers of variables lead to large numbers
of tests, so that many are bound to be significant. On the other hand, large numbers
of cases mean that even small effects will be declared significant in individual tests,
although they may not be of practical importance.
There is another issue which lessens the relevance of these two, large datasets
tend to be heterogeneous rather than homogeneous. Preliminary work may have to
be done in selecting variables of interest and determining the subset of cases to use
in an analysis before any testing is done, and the resulting dataset(s) may be much
smaller. And if the dataset is still very large in the number of cases, it is worth taking
advantage of the classical properties of statistical samples: there is no need to analyse
the whole dataset, as its aggregate properties will be just as apparent in a random
sample. To be on the safe side, we can either use fairly large samples or repeat an
analysis with several different samples. This approach is fine for what you might call
global properties of a dataset, but will not work for local ones such as unusual small
groups.
There are some properties of large datasets that raise particular issues for graphics
[Unwin et al., 2006]. There are almost certain to be some outliers, perhaps errors,
on many variables and this distorts the scaling of initial plots, as we have seen for
instance in Figures 3.11 and 9.8. Nominal variables may have so many categories
that they need to be grouped to make displays informative [Eick and Karr, 2002].
In general there will be more of everything, more points, more bars, more lines,
and plots need to be drawn larger. Zooming in becomes a useful tool and should be
accompanied by an overall unzoomed view (a bird’s-eye view) so that context is not
lost.
Large datasets bring advantages too. You can examine ideas in much greater de-
tail and check for more complex structures. If a hypothesis is developed on one part
of the data, it may possibly be checked on another. If that cannot be done, it may
still be feasible to explore the likely consequences of the idea and test it on other
variables from the dataset not yet used in the analysis.
The final point to make about working with large datasets is also a graphical one:
if there is a major effect, you should be able to draw a display that shows it.
Some Notes on Graphics with R 273

13.9 Perfecting graphics

The emphasis in this book is on using graphics for data analysis, for exploring data
rather than for formally presenting data. This means that it is important to be able
to draw graphics quickly and flexibly, not having to worry about dotting the last i or
crossing the last t of any displays. This demands reliable defaults for scales, legends,
colour, aspect ratio, and all the other properties of graphics. In general R is good at
that and even if it does not quite match up to your requirements in all respects, you
can still prepare your own default versions of the graphics types you use most often
to use instead.
If you have to present graphics in a publication, you may have to be much more
concerned with details: the fonts used, the precise colours, the placing of legends,
supplementary annotations and so on. It is possible to specify everything you need
in R, it is just that sometimes it takes rather a lot of code and some juggling with
the options. There is every reason to prepare as perfect a graphic as you can, once
you know which graphic you want and provided you have the time. Getting a display
just right for presentation is a different mode of working to carrying out analyses of
data and it is probably best to keep the two activities separate. Ideally neither design
issues nor computing problems should interfere with the analysis process.
For exploratory work the aim should be to produce flexible and informative
graphics, looking at many useful, if unpolished, graphics, to get an overall picture,
rather than concentrating on one elegant graphic that may not reveal the full story. To
paraphrase John Tukey: Better to draw several approximate graphics saying some-
thing about the right question than to draw one precise graphic relating to the wrong
question.
14
Summary

The single biggest problem in communication is the illusion that it has taken
place.

George Bernard Shaw

14.1 Data analysis and graphics

Good graphics are informative, effective and flexible. Graphics can be attractive and
encourage discussion, they can be more insightful and convincing than text. They are
fairly easy to explain to others, often easier to explain than the results of statistical
analyses. There is every reason to use them more in data analysis.
With the benefit of hindsight, Graphical Data Analysis is simple, you just show
the information contained in the data. In practice that is less straightforward than it
sounds, you first have to find out what information is there. The approach described
in this book recommends using a collection of common graphic displays to help
you uncover information. This is not about looking for a single optimal display and
showing how to draw the plot as attractively and precisely as possible. Rather it is
about drawing many displays and aiming to find an optimal set. Here ‘optimal’ means
that taken as a group the set of graphics show a lot—if not all—of the information
that is in the data.
Different types of graphics highlight different features in datasets. It is worth
drawing a few to check what you can see. A group of graphics of the same type, but
with different formatting and scaling, may also pick out a range of different aspects
of data. It is useful to experiment with different options: growing or shrinking plots,
changing their aspect ratios, reordering categories, and trying out different binwidths
or bandwidths. Of course, it is all too easy to get lost in large numbers of variables
and piles of output. Successful analyses, particularly of big datasets, depend on good
organisation. Ideally you need a manager for your analyses like the housekeeper
played by Helen Mirren in the film “Gosford Park”, who knew in advance what the
guests needed, before they knew it themselves.

275
276 Graphical Data Analysis with R

14.2 Key features of GDA

GDA entails drawing and interpreting graphics, the drawing alone is not enough.
“Every Picture Tells a Story" (Rod Stewart) and it is an essential part of GDA to look
for the story in every picture. A graphic with no message is no use.
GDA requires a strategic approach. It is about using many graphics at once and
being prepared to consider several different lines of thought in parallel. Who can tell
which of the features identified in graphics may turn out to be the most interesting?
It is important to have overall goals and simultaneously to be flexible in pursuing the
general goal of discovering new information.
Graphics may look different because of how they are drawn—the formatting,
scaling, and colouring. Using many graphics avoids possibly misleading effects in
individual displays due to this. On the other hand the same graphic may be interpreted
differently because of the size of the underlying dataset or the subject matter context.
Any such difference is intrinsic and should be emphasised not avoided.
GDA is about the generation of ideas, not just data description, and not the test-
ing of ideas, the contrast Tukey was referring to in writing of detectives and judges.
GDA should be used in close association with statistical modelling, as the two ap-
proaches complement each other well. Testing tells you if a null hypothesis should
be rejected but not necessarily why. Graphics show you what might be ‘wrong’,
but gives no guidance on how strong the evidence is. Statistical tests are useful for
checking whether an interesting graphical feature is significant. Graphics are useful
for asking whether a statistically significant result is really of interest
You can’t prove anything with statistics, but you can disprove some things and
the same goes for graphics. Drawing many graphics quickly and informatively for
exploratory purposes is also different from drawing a few graphics attractively and
precisely for presentation purposes. Tukey might have suggested the contrast of de-
tectives and designers.

14.3 Strengths and weaknesses of GDA

Every approach has its pluses and minuses. In the case of GDA there can be down-
sides due to badly chosen graphics, overloaded graphics, poorly organised groups of
graphics, over-interpretation of graphics, and apophenia (seeing patterns where there
aren’t any). GDA is, of course, also affected if data are inadequate, a problem for all
statistical and data analyses.
The strengths of GDA lie in its flexibility and ease of communication. It is also
rather robust. If you make an error or draw the wrong plot, you can readily see what
has happened and fix the problem. GDA is good for data cleaning, exploring data
structure, detecting outliers and unusual groups, identifying trends and clusters, spot-
Summary 277

ting local patterns, evaluating modelling output, and presenting results. It is about
generating ideas, while statistics is more about evaluating ideas.

14.4 Recommendations for GDA

A number of recommendations recur in this book:
• Use graphics to discover information that is difficult to investigate statistically.
• Draw many graphics and vary the graphics options.
• Gain experience in interpreting graphics and use the graphics types you know
well.
• Consider reformatting datasets before drawing graphics.
• Make appropriate comparisons and choose comparable scales.

• Check any graphical result you find with statistical models, where possible.
Statistics and graphics complement one another.
• Always remember how important context is in interpreting results.
The same statistics and graphics might be interpreted quite differently for dif-
ferent applications. That is why this book has concentrated on using ‘real’ datasets.
Finally, interactive graphics have occasionally been mentioned in this book as a tool
with potential for GDA and that is an important topic for the future.
References

[Aggarwal, 2013] Aggarwal, C. (2013). Outlier Analysis. Springer, New York.

[Agresti, 2007] Agresti, A. (2007). Categorical Data Analysis. Wiley, New York,
2nd edition.
[Albert, 1994] Albert, J. (1994). Exploring baseball hitting data: What about
those breakdown statistics? Journal of the American Statistical Association,
89(427):1066–1074.
[ASA, 2014] ASA (2014). Publications of the American Statistical Association.
http://amstat.tandfonline.com. (accessed 18.10.2014).
[Bache and Lichman, 2013] Bache, K. and Lichman, M. (2013). UCI machine
learning repository. http://archive.ics.uci.edu/ml. (accessed
12.11.2014).
[Barnett, 1994] Barnett, A. (1994). How numbers can trick you. Technology Review,
97(7):38–45.
[Barnett and Lewis, 1994] Barnett, V. and Lewis, T. (1994). Outliers in Statistical
Data. Wiley, Chichester, 3rd edition.
[Becker et al., 1988] Becker, R., Chambers, J., and Wilks, A. (1988). The New S
Language. Wadsworth & Brooks/Cole, Pacific Grove, California.
[Becker et al., 1996] Becker, R. A., Cleveland, W. S., and Shyu, M. J. (1996). The
visual design and control of trellis display. Journal of Computational and Graph-
ical Statistics, 5(2):123–155.
[Bederson and Shneiderman, 2003] Bederson, B. and Shneiderman, B. (2003). The
Craft of Information Visualization: Readings and Reflections. Morgan Kaufmann,
San Francisco, California.
[Ben-Gal, 2005] Ben-Gal, I. (2005). Outlier detection. In Maimon, O. and Rokach,
L., editors, Data Mining and Knowledge Discovery Handbook, pages 131–146.
Springer, New York.
[Benford, 1938] Benford, F. (1938). The law of anomalous numbers. Proceedings
of the American Philosophical Society, 78(4):551–572.

279
280 Graphical Data Analysis with R

[Benjamini, 2010] Benjamini, Y. (2010). Discovering the false discovery rate. Jour-
nal of the Royal Statistical Society: Series B (Statistical Methodology), 72(4):405–
416.
[Bertin, 2010] Bertin, J. (2010). Semiology of Graphics. Esri Press, Redlands, Cal-
ifornia.
[Bissantz, 2009] Bissantz, N. (2009). Me, myself, and BI. http://blog.
bissantz.com/vis-a-vis. (accessed 14.12.2013).
[Booker, 2004] Booker, C. (2004). The Seven Basic Plots. Continuum, London.
[Brain, 2004] Brain, M. (2004). How alligators work. http://animals.
howstuffworks.com/reptiles/alligator3.htm. (accessed
12.10.2014).
[Brant and Nguyen, 2008] Brant, S. and Nguyen, G. (2008). Is there a gender dif-
ference in the prevalence of Crohn’s disease or ulcerative colitis? Inflammatory
Bowel Diseases, 14(2).
[Bretz et al., 2010] Bretz, F., Hothorn, T., and Westfall, P. (2010). Multiple Compar-
isons Using R. Chapman & Hall/CRC Press, Boca Raton, Florida, USA.
[Brewer, 2013] Brewer, C. (2013). ColorBrewer. http://www.colorbrewer.
org. (accessed 12.11.2014).
[Chang, 2012] Chang, W. (2012). R Graphics Cookbook. O’Reilly Media, Beijing.
[Chen et al., 2008] Chen, C., Haerdle, W., and Unwin, A. R. (2008). Handbook of
Data Visualization. Springer, Berlin.
[Clemons and Pagano, 1999] Clemons, T. and Pagano, M. (1999). Are babies nor-
mal? The American Statistician, 53(4):298–302.
[Cleveland, 1994] Cleveland, W. (1994). The Elements of Graphing Data. Hobart
Press, Summit, New Jersey, revised edition.
[Cleveland et al., 1982] Cleveland, W., Diaconis, P., and McGill, R. (1982). Vari-
ables on scatterplots look more highly correlated when the scales are increased.
Science, 216:1138–41.
[Cleveland et al., 1988] Cleveland, W., McGill, M. E., and McGill, R. (1988). The
shape parameter of a two-variable graph. Journal of the American Statistical
Association, 83(402):289–300.
[Cleveland and McGill, 1987] Cleveland, W. and McGill, R. (1987). Graphical per-
ception: The visual decoding of quantitative information on graphical displays of
data. Journal of the Royal Statistical Society A, 150(3):192–229.
[Cleveland, 1993] Cleveland, W. S. (1993). Visualizing Data. Hobart Press, Summit,
New Jersey.
References 281

[Coen et al., 1969] Coen, P., Gomme, D., and Kendall, M. (1969). Lagged rela-
tionships in economic forecasting. Journal of the Royal Statistical Society A,
132(2):133–163.
[Cook and Swayne, 2007] Cook, D. and Swayne, D. (2007). Interactive and Dy-
namic Graphics for Data Analysis. Springer, New York.
[Davison, 2008] Davison, A. (2008). Statistical Models. Cambridge University
Press, Cambridge.
[De Veaux et al., 2011] De Veaux, R., Velleman, P., and Bock, D. (2011). Stats:
Data and Models. Pearson, Boston.

[Dodge, 1996] Dodge, Y. (1996). The guinea pig of multiple regression. In Rieder,
H., editor, Robust Statistics, Data Analysis, and Computer Intensive Methods,
pages 91–117. Springer, New York.
[DuToit et al., 1986] DuToit, S., Steyn, A., and Stumpf, R. (1986). Graphical Ex-
ploratory Data Analysis. Springer, New York.
[Edwards et al., 1963] Edwards, W., Lindman, H., and Savage, L. (1963). Bayesian
statistical inference for psychological research. Psychological Review, 70(3):193–
242.
[Eick and Karr, 2002] Eick, S. and Karr, A. (2002). Visual scalability. Journal of
Computational and Graphical Statistics, 11(1):22–43.
[Fahrmeir et al., 2013] Fahrmeir, L., Kneib, T., and Lang, S. (2013). Regression.
Springer, Berlin.

[Few, 2012] Few, S. (2012). Show Me the Numbers: Designing Tables and Graphs
to Enlighten. Analytics Press, Burlingame, California, 2nd edition.

[Fisher, 1925] Fisher, R. (1925). Statistical Methods for Research Workers. Oliver
and Boyd, Edinburgh.

[Fisher, 1936] Fisher, R. (1936). The use of multiple measurements in taxonomic

problems. Annals of Eugenics, 7(2):179–188.
[FiveThirtyEight, 2014] FiveThirtyEight (2014). Why Pollsters Think They Under-
estimated ‘No’ in Scotland. http://fivethirtyeight.com/datalab/
why-pollsters-think-they-underestimated-no-in-scotland/.
(accessed 30.11.2014).

[Fletcher, 2001] Fletcher, A. (2001). The Art of Looking Sideways. Phaidon, Lon-
don.
[Flury and Riedwyl, 1988] Flury, B. and Riedwyl, H. (1988). Multivariate Statistics
A Practical Approach. Chapman & Hall, London.
282 Graphical Data Analysis with R

[Forina et al., 1983] Forina, M., Armanino, C., Lanteri, S., and Tiscornia, E. (1983).
Classification of olive oils from their fatty acid composition. In Martens, H. and
Russwurm, H. J., editors, Food Research and Data Analysis, pages 189–214. Ap-
plied Science Publishers, London.
[Forina et al., 1988] Forina, M., Leardi, R., Armanino, C., and Lanteri, S. (1988).
Parvus - an extendible package for data exploration, classification and correlation.
[Freedman et al., 2007] Freedman, D., Pisani, R., and Purves, R. (2007). Statistics.
W.W. Norton, New York, 4th edition.
[Frey et al., 2011] Frey, B., Savage, D., and Torgler, B. (2011). Behavior under ex-
treme conditions: The Titanic disaster. Journal of Economic Perspectives, 25:209–
222.
[Friendly, 1994] Friendly, M. (1994). Mosaic displays for multi-way contingency
tables. Journal of the American Statistical Association, 89(1):190–200.
[Friendly, 2000] Friendly, M. (2000). Visualizing Categorical Data. SAS, Cary,
North Carolina.
[Friendly, 2011] Friendly, M. (2011). Gallery of data visualization. http://www.
datavis.ca/gallery/. (accessed 12.11.2014).
[Fung, 2011] Fung, K. (2011). Junk charts. http://junkcharts.typepad.
com/junk_charts/. (accessed 12.11.2014).
[Gannett, 1898] Gannett, H. (1898). Statistical Atlas of the United
States. Washington, D.C.: Government Print Office. (available at
http://www.census.gov/history/www/programs/geography/
statistical_atlases.html, accessed 20.10.2014).
[Gastwirth, 2006] Gastwirth, J. (2006). A 60 million dollar statistical issue arising
in the interpretation and calculation of a measure of relative disparity: Zuni Public
School District 89 v. U.S. Department of Education. Law Probability & Risk,
5(1):33–61.
[Gastwirth, 2008] Gastwirth, J. (2008). The U.S. Supreme Court finds a statute’s de-
scription of a simple statistical measure of relative disparity ‘ambiguous’ allowing
the Secretary of Education to interpret the formula: Zuni Public School District
89 v. U.S. Department of Education II. Law Probability & Risk, 7(3):225–248.
[Geary, 1947] Geary, R. C. (1947). Testing for normality. Biometrika, 34:209–242.
[Gelman, 2011] Gelman, A. (2011). Statistical modeling, causal inference, and
social science. http://www.stat.columbia.edu/~gelman/blog/.
(accessed 12.11.2014).
[Gelman et al., 2013] Gelman, A., Carlin, J., Stern, H., Dunson, D., Vehtari, A., and
Rubin, D. (2013). Bayesian Data Analysis. Chapman & Hall/CRC, Boca Raton,
Florida, 3rd edition.
References 283

[Gelman and Hill, 2006] Gelman, A. and Hill, J. (2006). Data Analysis Using Re-
gression and Multilevel/Hierarchical Models. Cambridge University Press, Cam-
bridge.
[Gilby, 1911] Gilby, W. H. (1911). On the significance of the teacher’s appreciation
of general intelligence. Biometrika, 8:94–108.
[Good and Gaskins, 1980] Good, I. J. and Gaskins, R. A. (1980). Density estimation
and bump-hunting by the penalized likelihood method exemplified by scattering
and meteorite data. Journal of the American Statistical Association, 75(369):42–
56.
[Google, 2010] Google (2010). Ngram viewer. https://books.google.
com/ngrams. (accessed 12.11.2014).
[Gregg, 1994] Gregg, P. (1994). Out for the count: A social scientist’s analysis of
unemployment statistics in the UK. Journal of the Royal Statistical Society A,
157(2):253–270.
[Guardian, 2013] Guardian (2013). University guide 2013. http:
//www.guardian.co.uk/news/datablog/2012/may/22/
university-guide-2013-guardian-data. (accessed 12.11.2014).
[Hanley, 2004] Hanley, J. A. (2004). “Transmuting” women into men: Galton’s fam-
ily data on human stature. The American Statistician, 58(3):237–243.
[Hartigan, 1975] Hartigan, J. A. (1975). Clustering Algorithms. John Wiley & Sons,
Inc., New York.
[Hartigan and Hartigan, 1985] Hartigan, J. A. and Hartigan, P. M. (1985). The dip
test of unimodality. The Annals of Statistics, 13(1):70–84.
[Hartigan and Kleiner, 1981] Hartigan, J. A. and Kleiner, B. (1981). Mosaics for
contingency tables. In Computer Science and Statistics: Proceedings of the 13th
Symposium on the Interface, pages 268–273, New York. Springer.

[Hastie et al., 2001] Hastie, T., Tibshirani, R., and Friedman, J. (2001). The Ele-
ments of Statistical Learning. Springer, New York, 2nd edition.
[Hummel, 1996] Hummel, J. (1996). Linked bar charts: Analysing categorical data
graphically. Computational Statistics, 11:23–33.
[Hurley and Oldford, 2010] Hurley, C. and Oldford, W. (2010). Pairwise display
of high-dimensional information via Eulerian tours and Hamiltonian decomposi-
tions. Journal of Computational and Graphical Statistics, 19(4):861–886.
[Hyndman, 2013] Hyndman, R. J. (2013). Time series task view. http:
//cran.r-project.org/web/views/TimeSeries.html. (accessed
12.11.2014).
284 Graphical Data Analysis with R

[IAAF, 2001] IAAF (2001). Scoring Tables for Combined Events. IAAF, Monaco.
[IBM, 2007] IBM (2007). Many eyes. http://many-eyes.com/. (accessed
12.11.2014).
[Inselberg, 2009] Inselberg, A. (2009). Parallel Coordinates: Visual Multidimen-
sional Geometry and Its Applications. Springer, New York.
[Izenman, 2008] Izenman, A. (2008). Modern Multivariate Statistical Techniques.
Springer, New York.
[Izenman and Sommer, 1988] Izenman, A. and Sommer, C. (1988). Philatelic mix-
tures and multimodal densities. Journal of the American Statistical Association,
83(404):941–953.
[James, 2014] James, S. (2014). US and UK inflation. http://www.sscnet.
ucla.edu/polisci/faculty/james/download/inflation.xls.
(accessed 12.11.2014).
[Johnson and Shneiderman, 1991] Johnson, B. and Shneiderman, B. (1991). Tree-
maps: a space-filling approach to the visualization of hierarchical information
structures. In Proceedings of the IEEE Conference on Visualization ’91, pages
284–291.
[Kleiber and Zeileis, 2008] Kleiber, C. and Zeileis, A. (2008). Applied Economet-
rics with R. Springer, New York.
[Kosara, 2011] Kosara, R. (2011). eagereyes. http://eagereyes.org/.
(accessed 12.11.2014).
[Kosara et al., 2006] Kosara, R., Bendix, F., and Hauser, H. (2006). Parallel sets:
Visual analysis of categorical data. Transactions on Visualization and Computer
Graphics, 12(4):558–568.
[Krause and McConnell, 2012] Krause, A. and McConnell, M. (2012). A Picture is
Worth a Thousand Tables: Graphics in Life Sciences. Springer, New York.
[Lewin-Koh, 2013] Lewin-Koh, N. (2013). Graphics task view. http://cran.
r-project.org/web/views/Graphics.html. (accessed 12.11.2014).
[Lock, 1993] Lock, R. (1993). 1993 new car data. Journal of Statistics Education,
1(1).
[Lumley, 2010] Lumley, T. (2010). Complex Surveys: A Guide to Analysis Using R.
Wiley, Hoboken, New Jersey.
[Maindonald and Braun, 2010] Maindonald, J. and Braun, R. (2010). Data Analysis
and Graphics Using R. Cambridge University Press, Cambridge.
[Malik, 2010] Malik, W. (2010). Data Cleaning of Large Datasets: New Methods
and Techniques. Verlag Dr. Hut, Munich.
References 285

[Mardia et al., 1979] Mardia, K. V., Kent, J. T., and Bibby, J. M. (1979). Multivari-
ate Analysis. Academic Press, London.
[Meyer et al., 2006] Meyer, D., Zeileis, A., and Hornik, K. (2006). The strucplot
framework: Visualizing multi-way contingency tables with vcd. Journal of Statis-
tical Software, 17(3):1–48.
[Mihos, 2005] Mihos, C. (2005). Astronomy 221. http://burro.astr.
cwru.edu/Academics/Astr221/HW/HW5/HW5.html. (accessed
17.11.2014).
[Mittal, 2011] Mittal, H. (2011). R Graph Cookbook. PACKT PUBLISHING, Birm-
ingham.

[Murrell, 2005] Murrell, P. (2005). R Graphics. Chapman & Hall, London.

[Murrell, 2009] Murrell, P. (2009). Data Technologies. Chapman & Hall, London.
[Murrell, 2011] Murrell, P. (2011). R Graphics. Chapman & Hall, London, 2nd
edition.
[New York Times, 2011] New York Times (2011). New York
Times Infographics. http://www.smallmeans.com/
new-york-times-infographics/. (accessed 17.11.2014).
[Newcomb, 1881] Newcomb, S. (1881). Note on the frequency of use of the differ-
ent digits in natural numbers. American Journal of Mathematics, 4(1):39–40.
[Norman, 1988] Norman, D. A. (1988). The Design of Everyday Things. MIT Press,
Cambridge, Mass.
[OCSI, 2009] OCSI (2009). Improving data visualisation for the public
sector. http://www.improving-visualisation.org/. (accessed
12.11.2014).

[Olsen, 1998] Olsen, C. (1998). A comparison of parametric and semiparametric es-

timates of the effect of spousal health insurance coverage on weekly hours worked
by wives. Journal of Applied Econometrics, 13(5):543–565.

[Ord and Fildes, 2013] Ord, K. and Fildes, R. (2013). Principles of Business Fore-
casting. Cengage Learning, Mason, Ohio.
[Pearson and Lee, 1903] Pearson, K. and Lee, A. (1903). On the laws of inheritance
in man: I. Inheritance of physical characters. Biometrika, 2(4):357–462.
[Pilhoefer et al., 2012] Pilhoefer, A., Gribov, A., and Unwin, A. (2012). Comparing
clusterings using Bertin’s idea. IEEE Transactions on Visualization and Computer
Graphics, 18(12):2506–2515.
286 Graphical Data Analysis with R

[Pilhoefer and Unwin, 2013] Pilhoefer, A. and Unwin, A. (2013). New approaches
in visualization of categorical data: R-package extracat. Journal of Statistical
Software, 53(7):1–25.
[Playfair, 2005] Playfair, W. (2005). Playfair’s Commercial and Political Atlas and
Statistical Breviary. Cambridge University Press, Cambridge.
[Rahlf, 2014] Rahlf, T. (2014). Datendesign mit R — 100 Visualisierungsbeispiele.
Open Source Press, Munich.
[rlearnr, 2009] rlearnr (2009). Learning R. http://learnr.wordpress.
com/. (accessed 12.11.2014).
[Robbins, 2005] Robbins, N. (2005). Creating More Effective Graphs. Wiley-
Blackwell, Hoboken, New Jersey.
[Rogers and Hsu, 2001] Rogers, J. and Hsu, J. (2001). Multiple comparisons of bio-
diversity. Biometrical Journal, 43(5):617–625.
[Rosling, 2013] Rosling, H. (2013). Gapminder. http://www.gapminder.
org. (accessed 12.11.2014).
[Rubin, 1987] Rubin, D. (1987). Multiple Imputation for Nonresponse in Surveys.
Wiley, New York.
[Salmistu, 2013] Salmistu, J. (2013). Decathlon2000. http://www.
decathlon2000.com/eng/. (accessed 12.11.2014).
[Sarkar, 2008] Sarkar, D. (2008). Lattice: Multivariate Data Visualization with R.
useR. Springer, New York.
[Schilling et al., 2002] Schilling, M. F., Watkins, A. E., and Watkins, W. (2002). Is
human height bimodal? The American Statistician, 56(3):223–229.
[Scott, 1992] Scott, D. (1992). Multivariate Density Estimation — Theory, Practice,
and Visualization. Wiley, New York.
[Sheehan et al., 1991] Sheehan, P., Fastovsky, D., Hoffmann, R. G., Berghaus, C.,
and Gabrielt, D. (1991). Sudden extinction of the dinosaurs: Latest Cretaceous,
Upper Great Plains, U.S.A. Science, 254(5033):835–839.
[Simmon, 2014] Simmon, R. (2014). Subtleties of color. http:
//earthobservatory.nasa.gov/blogs/elegantfigures/
2013/08/05/subtleties-of-color-part-1-of-6/.
(accessed 13.10.2014).
[Snee, 1974] Snee, R. (1974). Graphical display of two-way contingency tables. The
American Statistician, 28(1):9–12.
[Spence, 2007] Spence, R. (2007). Information Visualization: Design for Interac-
tion. Pearson, Harlow, England.
References 287

[Sunday Independent, 2013] Sunday Independent (2013). FG and FF neck

and neck in poll. http://www.independent.ie/irish-news/
fg-and-ff-neck-and-neck-in-poll-29507501.html.
(accessed 17.11.2014).
[Guardian, 2011] Guardian (2011). Datastore. http://www.guardian.co.
uk/data. (accessed 12.11.2014).
[Theus, 2013] Theus, M. (2013). Statistical graphics and more. http://www.
theusRus.de/blog/. (accessed 12.11.2014).
[Theus and Urbanek, 2007] Theus, M. and Urbanek, S. (2007). Interactive Graphics
for Data Analysis. CRC Press, London.

[Thode Jr., 2002] Thode Jr., H. (2002). Testing for Normality. Marcel Dekker, New
York.
[TimelyPortfolio, 2013] TimelyPortfolio (2013). R financial time series plotting.
http://timelyportfolio.github.io/rCharts_time_series/
history.html. (accessed 12.11.2014).

[Tsay, 2014] Tsay, R. (2014). Multivariate Time Series Analysis: with R and Finan-
cial Applications. Wiley, Hoboken, New Jersey.
[Tufte, 1990] Tufte, E. (1990). Envisioning Information. Graphic Press, Cheshire,
Connecticut.
[Tufte, 2001] Tufte, E. (2001). The Visual Display of Quantitative Information.
Graphic Press, Cheshire, Connecticut, 2nd edition.

[Tufte, 2013] Tufte, E. (2013). ET NOTEBOOKS.

http://www.edwardtufte.com/bboard/. (accessed 12.11.2014).

[Tukey, 1993] Tukey, J. (1993). Graphic comparisons of several linked aspects:

Alternatives and suggested principles. Journal of Computational and Graphical
Statistics, 2(1):1–33.
[Tutz, 2012] Tutz, G. (2012). Regression for Categorical Data. Cambridge Univer-
sity Press, Cambridge.

[Unwin et al., 2013] Unwin, A., Hofmann, H., and Cook, D. (2013). Let graphics
tell the story - datasets in R. R Journal, 5(1):117–129.
[Unwin, 2008] Unwin, A. R. (2008). Good graphics? In Chen, C., Haerdle, W., and
Unwin, A. R., editors, Handbook of Data Visualization, pages 57–78. Springer,
New York.
[Unwin et al., 2006] Unwin, A. R., Theus, M., and Hofmann, H. (2006). Graphics
of Large Datasets. Springer, New York.
288 Graphical Data Analysis with R

[Unwin et al., 2003] Unwin, A. R., Volinsky, C., and Winkler, S. (2003). Parallel
coordinates for exploratory modelling analysis. Computational Statistics & Data
Analysis, 43(4):553–564.
[Urbanek et al., 2014] Urbanek, S., Woodhull, G., et al. (2014). Rcloud. http:
//stats.research.att.com/RCloud/. (accessed 12.11.2014).
[van Belle et al., 2004] van Belle, G., Fisher, L., Heagerty, P., and Lumley, T. (2004).
Biostatistics: A Methodology For the Health Sciences. Wiley, New York, 2nd
edition.
[Vande Moere, 2011] Vande Moere, A. (2011). information aesthetics. http://
infosthetics.com/. (accessed 12.11.2014).

[Venables and Ripley, 2002] Venables, W. N. and Ripley, B. D. (2002). Modern Ap-
plied Statistics with S. Springer, New York, 4th edition.
[Viz, 2011] Viz (2011). Visualizing.org. http://www.visualizing.org/.
(accessed 12.11.2014).
[von Bortkiewicz, 1898] von Bortkiewicz, L. (1898). Das Gesetz der kleinen
Zahlen. Teubner, Leipzig.
[Wainer, 1997] Wainer, H. (1997). Visual Revelations. Springer, New York.
[Wainer, 2004] Wainer, H. (2004). Graphic Discovery: A Trout in the Milk and
Other Visual Adventures. Princeton University Press, Princeton, New Jersey.
[Wainer, 2009] Wainer, H. (2009). Picturing the Uncertain World: How to Un-
derstand, Communicate, and Control Uncertainty through Graphical Display.
Princeton University Press, Princeton, New Jersey.
[Wand, 1997] Wand, M. (1997). Data-based choice of histogram bin width. The
American Statistician, 51(1):59–64.

[Wand and Jones, 1995] Wand, M. and Jones, M. (1995). Kernel Smoothing. Chap-
man & Hall, London.
[Ware, 2008] Ware, C. (2008). Visual Thinking: for Design. Morgan Kaufmann,
San Francisco, California.
[Wattenberg, 2005] Wattenberg, M. (2005). Name voyager. http://www.
babynamewizard.com/voyager. (accessed 11.11.2014).
[Wegman, 1990] Wegman, E. (1990). Hyperdimensional data analysis using parallel
coordinates. Journal of the American Statistical Association, 85:664–675.
[Wickham, 2009] Wickham, H. (2009). ggplot2: Elegant graphics for data analysis.
useR. Springer, New York.
References 289

[Wikipedia, 2013] Wikipedia (2013). Old Faithful. http://en.wikipedia.

org/wiki/Old_Faithful. (accessed 17.11.2014).
[Wikipedia, 2014] Wikipedia (2014). Charge of the Light Brigade. http://en.
wikipedia.org/wiki/Charge_of_the_Light_Brigade. (accessed
21.3.2014).
[Wilkinson, 2005] Wilkinson, L. (2005). The Grammar of Graphics. Springer, New
York, 2nd edition.
[Wills, 2012] Wills, G. (2012). Visualizing Time. Springer, New York.
[Worcester, 1996] Worcester, R. (1996). Political polling: 95% expertise and 5%
luck. Journal of the Royal Statistical Society. Series A (Statistics in Society),
159(1):5–20.
[Wright, 2013] Wright, K. (2013). Revisiting Immer’s barley data. The American
Statistician, 67(3):129–133.
[Yau, 2011] Yau, N. (2011). FlowingData. http://flowingdata.com/.
(accessed 17.11.2014).
[Yule, 1926] Yule, G. (1926). Why do we sometimes get nonsense-correlations be-
tween time-series?—A study in sampling and the nature of time-series. Journal
of the Royal Statistical Society, 89(1):1–63.

[Zeileis et al., 2009] Zeileis, A., Hornik, K., and Murrell, P. (2009). Escaping rgb-
land: Selecting colors for statistical graphics. Computational Statistics & Data
Analysis, 53:3259–3270.

[Zheng and Gastwirth, 2010] Zheng, T. and Gastwirth, J. (2010). On bootstrap tests
of symmetry about an unknown median. Journal of Data Science, 8:397–412.
 Statistics
The R Series

Graphical Data Analysis with R

Graphical Data Analysis with R shows you what information you
can gain from graphical displays. The book focuses on why you draw
graphics to display data and which graphics to draw (and uses R to
do so). All the datasets are available in R or one of its packages and
the R code is available at rosuda.org/GDA.
Graphical Data
Graphical data analysis is useful for data cleaning, exploring data
structure, detecting outliers and unusual groups, identifying trends
and clusters, spotting local patterns, evaluating modelling output,
Analysis with R
and presenting results. This book guides you in choosing graphics
and understanding what information you can glean from them. It
can be used as a primary text in a graphical data analysis course
or as a supplement in a statistics course. Colour graphics are used
throughout.
Features
• Concentrates on why graphics are drawn and what they reveal
• Emphasises the value of drawing a variety of different graphics
• Uses real datasets in R to show how graphical data analysis
works in practice
• Supplies R code for all the graphics on the author’s website
• Includes a set of exercises in each chapter to facilitate hands-on
learning
Antony Unwin is a professor of computer-oriented statistics and data
analysis at the University of Augsburg. He is a fellow of the American
Statistical Society. His research focuses on data visualisation,
especially in interactive graphics.
Unwin

Antony Unwin

K25332

w w w. c rc p r e s s . c o m

K25332_cover.indd 1 3/4/15 3:38 PM