Using Graphs and Visualising Data

2002 2007 2002 2007 2002 2007

Brazil Indonesia Russia India U.S.A. China

35

30

25
investment share

20

15

2002 2007 2002 2007 2002 2007

year

Oliver Kirchkamp
 © Oliver Kirchkamp
2 Using Graphs and Visualising Data — Contents

Contents
1 Introduction 4
1.1 Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Properties of good graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 How to present . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4.1 Axes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4.2 Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4.3 Points, lines and bars . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.4.4 Error Bars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5 Legends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.6 Clutter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.7 Unnecessary 3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.8 Aspect ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.9 What to present . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.9.1 Structuring content . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.9.2 Don’t discard parts of your data . . . . . . . . . . . . . . . . . . . . . 28
1.9.3 Projecting data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.9.4 Differences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2 ggplot 33
2.1 Elements of ggplot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2 Labels and legends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3 Scatterplots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

3 ggplot, more advanced features 45

3.1 Segment plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 Densityplots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.3 Histograms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.4 Empirical cumulative distribution . . . . . . . . . . . . . . . . . . . . . . . . 48
3.5 Q-Q plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.6 Sample Q-Q plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.7 Boxplots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.8 Barcharts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.9 Coplots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.10 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.10.1 Types of lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.10.2 Axes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.11 Zooming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.12 Themes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

4 Nominal data 62
4.1 Nominal univariate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
—3

4.2 Nominal bivariate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

4.3 Nominal multivariate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

5 Continuous data – distributions 71

5.1 Diagnostic plots for continuous variables . . . . . . . . . . . . . . . . . . . . 71
5.2 One continuous plus one nominal . . . . . . . . . . . . . . . . . . . . . . . . 72
5.2.1 Histograms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.2.2 Densities and conditional densities . . . . . . . . . . . . . . . . . . . 73
5.2.3 Barplot of means . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.2.4 Means and standard deviation . . . . . . . . . . . . . . . . . . . . . . 76
5.2.5 Empirical cumulative distributions . . . . . . . . . . . . . . . . . . . 77
5.3 More on Dot-plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.5 Two continuous variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.5.1 Scatterplot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.5.2 Scatterplot with data ellipses . . . . . . . . . . . . . . . . . . . . . . 80
5.5.3 Bagplot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.5.4 Kernel densities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

6 Continuous data, causal relations, other problems 87

6.1 Causal relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.1.1 Smooth lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
6.1.2 GAM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6.1.3 Visually weighted Regression . . . . . . . . . . . . . . . . . . . . . . 93
6.1.4 Summary: Two continuous variables . . . . . . . . . . . . . . . . . . 94
6.2 Other problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
6.2.1 Paired data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
6.2.2 Three-dimensional simplex . . . . . . . . . . . . . . . . . . . . . . . 96
6.2.3 Stars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

7 Lattice 98
7.1 Multiway xyplots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
7.2 Syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
7.3 Multiway continued . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
7.4 Densityplots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7.5 Histograms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.6 Empirical cumulative densities . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.7 Q-Q plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
7.8 Sample Q-Q plots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
7.9 Boxplots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.10 Barcharts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
7.11 Coplots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
7.12 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.12.1 Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
 © Oliver Kirchkamp
4 Using Graphs and Visualising Data — 1 INTRODUCTION

7.12.2 Axes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112

1 Introduction
1.1 Literature
• The Elements of Graphing Data (Revised Edition). W. S. Cleveland (1994). Hobart
Press, Summit, New Jersey, U.S.A.

• The Visual Display of Quantitative Information. Edward Tufte (2001). Bertrams.

1.2 Examples
The following four datasets all have the same correlation ρ = 0.8730379.
They all have the same regression coefficients β0 =0.15 and β1 =0.5.

x1 y1 x2 y2 x3 y3 x4 y4
0.00 0.16 0.00 0.17 0.00 0.19 0.00 0.00
0.10 0.21 0.00 0.01 0.10 0.07 0.10 0.14
0.20 0.25 0.00 0.10 0.20 0.21 0.20 0.26
0.30 0.29 0.00 0.25 0.30 0.42 0.30 0.36
0.40 0.33 0.00 0.10 0.40 0.29 0.40 0.44
0.50 0.37 0.00 0.25 0.50 0.49 0.50 0.50
0.60 0.41 0.00 0.22 0.60 0.51 0.60 0.54
0.70 0.45 0.00 0.17 0.70 0.50 0.70 0.56
0.80 0.82 0.00 0.22 0.80 0.59 0.80 0.56
0.90 0.54 0.00 0.01 0.90 0.42 0.90 0.54
1.00 0.58 1.00 0.65 1.00 0.71 1.00 0.50

1 2 3 4

0.8

0.6

0.4
y

0.2

0.0
0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
x
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
—5

William Playfair’s trade-balance time-series chart, 1786:

Charles Minard’s 1869 chart of Napoleon’s 1812 Russian campaign:

 © Oliver Kirchkamp
6 Using Graphs and Visualising Data — 1 INTRODUCTION

John Snow, 1854 Broad Street cholera outbreak:

Data to Ink ratio — Daily Max/Min-Temperature in Erfurt/Weimar 1957 – 2021

46996 numbers:

40
Temperature / [◦ C]

20
Temp. Range
1957-2021

0 2021

-20

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
—7

Climate change in Köln/Bonn from 1951-2021?

tmin tmax

40
0
Temperature / [◦ C]

-4 35

-8
30

-12

1960 1980 2000 2020 1960 1980 2000 2020

year

Aims:

• Note: good graphs are self-explanatory!

→ The key to understand a graphs should not be hidden somewhere in the text!

• Often, the optimal presentation of data is not “standard”.

• There are no “recipes” how to present data.

• We have to use our own imagination.

• Still, some examples might help.

What can be achieved with a good graph?

A graph can…

• …make the reader familiar with the structure of the data (create trust),

• …motivate a research question,

• …summarise conclusions of the paper.

A graph must be excellent:

• Some readers look only or mainly at the figures and the graphs.

• Each graph should tell a story and should be self-explanatory.

• Among the many ways to present our data and our results, we have to chose the best
way.

Presenting only aggregate statistics, tests and estimates might discard relevant structure:
 © Oliver Kirchkamp
8 Using Graphs and Visualising Data — 1 INTRODUCTION

1.3 Properties of good graphs

• Essential items are shown clearly.

• Superfluous items are not shown.

• All elements of the graph are explained in the figure (not only in the text).

1.4 How to present

1.4.1 Axes
Frames: The following two graphs show the same data. However, in the graph on the
left the axes and the data points are superimposed. In the graph on the right axes and data
points are separate items. A frame around the plot region makes it more clear where the
reader should expect data.

set.seed(131)
N<-50
x<-rnorm(n=N,mean=1.5)
y<-x+rnorm(N)
x<-pmax(0,x)
#
z<-rbinom(N,2,.5)
Treatment <- factor(z)
labels<-c("Baseline","A","B")
levels(Treatment)<-labels
gData <- data.frame(x,y,z,Treatment)

par(mfrow=c(1,2),mar=c(4,3,0,1),mex=.5)
plot(y ~ x,axes=FALSE,cex=.25)
axis(side=1,pos=0)
axis(side=2,pos=0)
plot(y ~ x,cex=.25)
axis(3,labels=FALSE)
axis(4,labels=FALSE)
4

4
2

2
y

y
0

0 1 2 3 4
-2

-2

0 1 2 3 4
x x

• Labels and tick marks are separated from the data.

• Ticks on the opposite axis can help.

 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
—9

Ranges: In the following example both graphs show the same data. The only difference is
the range of the axes.
p1 <- ggplot() + geom_point(aes(x,y))
p2 <- ggplot() + geom_point(aes(x,y)) +
scale_x_continuous(expand=expansion(mult=.25)) +
scale_y_continuous(expand=expansion(mult=.25))
grid.arrange(p1,p2,nrow=1)

5.0
6

4
2.5
2
y

y
0
0.0

-2

-2.5
0 1 2 3 4 0 1 2 3 4
x x

Ranges are chosen such that…

• …all data is included,
• …space is used in an efficient way.

Ranges and correlations We tend to perceive more correlation if the data occupies less
space.
• The amount of white space around the data should be similar in all graphs.

Ranges that include zeroes: Mauna Loa Atmospheric CO2 Concentration:

data.frame(lowess(co2)) %>%
ggplot() + geom_line(aes(x,y)) +
labs(y="$CO_2$",x="year") -> p1
p1 + expand_limits(y=0) -> p2
grid.arrange(p1,p2,nrow=1)

360
300
350

200
CO2

CO2

340

330
100

320
0
1960 1970 1980 1990 1960 1970 1980 1990
year year
 © Oliver Kirchkamp
10 Using Graphs and Visualising Data — 1 INTRODUCTION

It can be helpful to include zero, but it can also waste space.

Comparable scales: In the following example we use different scales for the two dia-
grams. This makes them difficult to compare (although space is used efficiently).1

library(pwt10)
data(pwt10.0)
pwt10.0 %>%
filter(country %in% c("Norway","Haiti")) %>%
select(c("cgdpo","pop","country","year")) %>%
mutate(gdp = cgdpo/pop) -> pwt

ggplot(pwt) + geom_line(aes(x=year,y=cgdpo)) +
labs(y="GDPo/head (US \\$)") +
facet_wrap(vars(country),scales="free")

Haiti Norway

20000
400000
GDPo/head (US $)

15000 300000

200000
10000
100000

5000
1960 1980 2000 2020 1960 1980 2000 2020
year

The figure shows real gross domestic product (GDPo) per capita (US dollars in 2017 prices).
Data is taken from Penn World Table Version 10.0.
In the next figure we use the same scale for the two diagrams. Now we see immediately
that GDPo is larger in Brazil and smaller in Indonesia. Of course, presenting both lines in
one diagram might be preferable, here.

ggplot(pwt) + geom_line(aes(x=year,y=cgdpo)) +
labs(y="GDPo/head (US \\$)") +
facet_wrap(vars(country),scales="fixed")

1
Data from Alan Heston, Robert Summers and Bettina Aten, Penn World Table Version 10.0, Center for Inter-
national Comparisons of Production, Income and Prices at the University of Pennsylvania, August 2009.
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 11

Haiti Norway

400000
GDPo/head (US $)
300000

200000

100000

0
1960 1980 2000 2020 1960 1980 2000 2020
year

The figure shows real gross domestic product (GDPo) per capita (US dollars in 2017 prices). Data is taken from
Penn World Table Version 10.0.

ggplot(pwt) + geom_line(aes(year,y=cgdpo)) +
labs(y="GDPo/head (US \\$)") +
facet_wrap(vars(country),scales="fixed") +
scale_y_log10()

Haiti Norway

300000
GDPo/head (US $)

100000

30000

10000

1960 1980 2000 2020 1960 1980 2000 2020

year

A logarithmic scale facilitates comparing relative growth. Also with logs, comparability
does not require to use the same axes for several diagrams. Sometimes it might be better to
use the same scale with a different origin.
Another possibility are sliced scales, i.e. the same scale, but different origins:

xyplot(cgdpo ~ year | country,data=pwt,type="l",

scales=list(y=list(log=10,relation="sliced")),
between=list(x=2),
par.settings = list(layout.widths = list(right.padding = 5)),
yscale.components=yscale.components.log10.3,
ylab="GDPo/head (US\\$)")
 © Oliver Kirchkamp
12 Using Graphs and Visualising Data — 1 INTRODUCTION

1960 1980 2000 2020

Haiti Norway

30000

30000

300000

300000
GDPo/head (US$)
10000

10000

100000

100000
3000

3000

30000

30000
1960 1980 2000 2020
year

A sliced logarithmic scale facilitates comparing relative growth even more.

ggplot(pwt) + geom_line(aes(year,y=cgdpo)) +
labs(y="GDPo/head (US \\$)") +
facet_wrap(vars(country),scales="free") +
scale_y_log10()

Haiti Norway

20000

300000
GDPo/head (US $)

10000 100000

7000

30000
1960 1980 2000 2020 1960 1980 2000 2020
year

The figure shows real gross domestic product (GDPo) per capita (dollars in 2017 prices). Data is taken from
Penn World Table Version 10.0.

Breaks:

data.frame(x=1:8,y=c(1,3,4,52,51,5,4,3)) %>%
mutate(out=y > mean(c(max(y),min(y))),
shrink=min(y[out])-max(y[!out])-2,
yShrink=y - ifelse(out,shrink,0)) -> dBreak

p1<-ggplot(dBreak,aes(x=x,y=y)) + geom_line()
p2<-ggplot(dBreak,aes(x=x,y=y)) + geom_line() + scale_y_log10()
p3<-ggplot(dBreak,aes(x=x,y=y)) + geom_line() + geom_point() +
facet_grid(out ~ .,scale="free_y",space="free_y",as.table=FALSE) +
theme(strip.text.y = element_blank())
grid.arrange(p1,p2,p3,nrow=1)
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 13

52.00
51.75
50 51.50
51.25
30 51.00
40 5

30 10 4
y

y
20 3
3
10 2

0 1 1
2 4 6 8 2 4 6 8 2 4 6 8
x x x

All three graphs try to show the same data. The first graph uses a linear scale. Here the
outlier is clearly visible.
The graph in the middle uses a logarithmic scale. This can be reasonable if ratios of the
variable are interesting.
The last graph tries to save space by “breaking” the axis. If gaps can not be avoided,
dividing the graph into different panels might be preferable (the shingle function might
help to find divisions).
ggplot can “slice” axes, however I find the result not entirely convincing:
GGplot:

ggplot(dBreak,aes(x=x,y=y)) + geom_line() +
geom_point() +
facet_grid(out ~ .,scale="free_y",
space="free_y",as.table=FALSE) +
theme(strip.text.y = element_blank())

52.00
51.75
51.50
51.25
51.00
5

4
y

1
2 4 6 8
x

Lattice:
 © Oliver Kirchkamp
14 Using Graphs and Visualising Data — 1 INTRODUCTION

xyplot(y ~ x|out,data=dBreak,layout=c(1,2),
strip=FALSE,scales=list(y="sliced"),
between=list(y=.5),type="o")

50 51 52 53
y
5
4
3
2
1

2 4 6 8
x

Logarithmic scales: The first graph shows GDPo on a linear scale, the second one uses
a logarithmic scale. On a linear scale countries like Congo and Ethiopia seem to be quite
similar, Germany and U.S.A. look distinct.

The log scale makes it easier to compare ratios. We see that, in relative terms, Germany is
perhaps closer to the United States of America than Congo is to Ethiopia.

N<-20
pwt10.0 %>% filter(year==2019 & country!="China Version 2") %>%
filter(pop >= -sort(-pop)[N]) %>% ## only the top N countries
mutate(country=substr(country,1,10)) %>%
mutate(country=reorder(country,-cgdpo/pop),
gdp=cgdpo/pop) -> pwtG20
ggplot(pwtG20,aes(x=gdp,y=country)) + geom_point() + labs(x="GDP/head")
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 15

Congo, Dem
Ethiopia
Bangladesh
Nigeria
Pakistan
India
Viet Nam
Philippine
Indonesia

country
Egypt
China
Iran (Isla
Brazil
Thailand
Mexico
Turkey
Russian Fe
Japan
Germany
United Sta
0 20000 40000 60000
GDP/head

ggplot(pwtG20,aes(x=gdp,y=country)) + geom_point() + scale_x_log10() + labs(x="GDP/head")

Congo, Dem
Ethiopia
Bangladesh
Nigeria
Pakistan
India
Viet Nam
Philippine
Indonesia
country

Egypt
China
Iran (Isla
Brazil
Thailand
Mexico
Turkey
Russian Fe
Japan
Germany
United Sta
1000 3000 10000 30000
GDP/head

Which logarithm?
When the ratio of the largest to the smallest value is really large, then a logarithmic scale
is easy to grasp.
par(mar=c(4,0,0,0),mex=.5)
plot(NULL,xlim=c(.5,20000),ylim=c(0,1),axes=FALSE,log="x",ylab="",xlab="")
axis(1)

1 10 100 1000 10000

As an alternative we could also write the powers of 10:

 © Oliver Kirchkamp
16 Using Graphs and Visualising Data — 1 INTRODUCTION

labs<-10^(0:4)
plot(NULL,xlim=c(.5,20000),ylim=c(0,1),axes=FALSE,log="x",ylab="",xlab="")
axis(1,at=labs,labels=paste("$10^",log10(labs),"$"))

100 101 102 103 104

The scale is harder to grasp when the ratio is less extreme:

labs<-c(200,300,500,1000,1500)
plot(NULL,xlim=c(150,1800),ylim=c(0,1),axes=FALSE,log="x",ylab="",xlab="")
axis(1,at=labs,labels=labs)

200 300 500 1000 1500

Here using a logarithm with base two can help:

labs<-100*(2^(1:4))
plot(NULL,xlim=c(150,1800),ylim=c(0,1),axes=FALSE,log="x",ylab="",xlab="")
axis(1,at=labs,labels=labs)

200 400 800 1600

1.4.2 Points
Size In the following example the graph on the left uses fairly small points, the one on the
right uses larger points to display the data.
plot(y ~ x,cex=.25)
plot(y ~ x)
5

5
4

4
3

3
2

2
y

y
1

1
0

0
-1

-1
-2

-2

0 1 2 3 4 0 1 2 3 4
x x
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 17

Shape Plotting symbols should be easy to distinguish. Compare the graph on the left with
the graph on the right:

plot(y ~ x,pch=c(15,16,17)[z+1])
legend("bottomright",c("Group I","Group II","Baseline"),pch=c(15,16,17),bg="white")
plot(y ~ x,pch=c(1,3,5)[z+1])
legend("bottomright",c("Group I","Group II","Baseline"),pch=c(1,3,5),bg="white")
5

5
4

4
3

3
2

2
y

y
1

1
Group I Group I
0

0
Group II Group II
Baseline Baseline
-2

-2
0 1 2 3 4 0 1 2 3 4

x x

• When points do not tend to overlap we can use both heavy (•) and light (◦) plotting
symbols. Their contrasts helps us to distinguish different types of points.

• When points tend to overlap, heavy symbols are difficult to disentangle. In this situa-
tion we should use only light symbols.

Kröse’s experiment
Participants see patterns of symbols for 80 ms. They have to identify presence or absence
of a given symbol.

symbols % recognised
+◦ 100.0
+□ 88.1
L+ 68.6
∆↓ 52.3
+T 37.6
+X 30.3
TL 30.6

(B. J. A. Kröse, Local structure analyzers as determinants of preattentive pattern discrimi-

nation. Biological Cybernetics, 1987, Volume 55, Number 5, 289-298.)
 © Oliver Kirchkamp
18 Using Graphs and Visualising Data — 1 INTRODUCTION

Nominal data and points Sometimes, in particular with nominal data, we want to show
the same obversation several times. In the diagram on the left multiple dots are simple printed
on top of each other. One can not see how frequent an observation is. In the middle we add
jitter to each observation, small noise, which allows us to distinguish the single observa-
tions. The left graph shows frequencies as size of the symbol.

set.seed(1)
nomData <- data.frame(x=sample(1:4,size=200,replace=TRUE),y=sample(1:4,size=200,replace=TRUE))
p1 <- ggplot(nomData,aes(x,y)) + geom_point(shape=1)
p2 <- ggplot(nomData,aes(x,y)) + geom_jitter(width=.1,height=.1,shape=1)
nomData %>% group_by(x,y) %>% summarise(size=n()) %>% mutate(y=factor(y)) -> nomData2
p3 <- ggplot(nomData2,aes(x,y,size=size)) + geom_point(shape=1)
grid.arrange(p1,p2,p3,nrow=1)

4
4
4

size
3 3
3 8
10
y

y
12

14
2
2 2
16

1
1
1

1 2 3 4 1 2 3 4 1 2 3 4
x x x

If we have a small number of categories (at least in one dimension), a dotplot might be
better:

ggplot(nomData2,aes(x=x,y=size,color=y,lty=y,shape=y)) +
geom_line() + geom_point() +
theme(legend.key.width = unit(2,"cm"))
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 19

16

14
y
1
size

12 2
3
4
10

1 2 3 4
x

1.4.3 Points, lines and bars

plot(sunspot.year[1:40],ylab="",xlab="")
plot(sunspot.year[1:40],t="l",ylab="",xlab="")
plot(sunspot.year[1:40],t="b",ylab="",xlab="")
plot(sunspot.year[1:40],t="h",ylab="",xlab="")
120

120

120

120
100

100

100

100
80

80

80

80
60

60

60

60
40

40

40

40
20

20

20

20
0

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

• The development over time is easier to see with lines.

• Lines alone make it impossible to find out when the measurements were taken.
 © Oliver Kirchkamp
20 Using Graphs and Visualising Data — 1 INTRODUCTION

1.4.4 Error Bars

set.seed(6)
eBars<-data.frame(y=rnorm(200),i=rep(1:20,each=10))
eBars %>% group_by(i) %>%
summarise(ym = median(y)+mean(y)) %>%
mutate(x = factor(rank(ym))) %>% right_join(eBars) -> eBars2
eBars2 %>% group_by(x) %>%
summarise(s=sd(y),y=mean(y)) -> eBarsS

ggplot(eBarsS,aes(x=x,y=y,min=y-s,max=y+s)) + geom_point(shape=1) + geom_errorbar()

Error bars can refer to serveral quantities:

1

• Standard deviation of the sample

• Standard deviation of the estimated 0

mean y

• 95% confidence intervals of the es- -1

timated mean
.
• .. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
x

We have to explain clearly in the figure which quantity (standard deviation of the sample,
standard deviation of the estimated mean, confidence interval,…) is shown.

Boxplots Boxplots are often more informative than error bars.

ggplot(eBars2,aes(x=x,y=y)) + geom_boxplot(outlier.shape=1)

1
y

-1

-2

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
x

Elements of the boxplot:

 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 21

• The median (thick line in the middle)

• Interquartile range (25% and 75% quantile) (the box)
• Whiskers to the most extreme data point which is no more than 1.5 times the interquar-
tile range from the box.
• Observations outside the whiskers are shown as dots (outliers).

1.5 Legends
p1 <- ggplot(gData,aes(x=x,y=y,shape=Treatment,color=Treatment)) + geom_point() +
scale_shape_manual(values=1:4) +
theme(legend.position=c(.8,.25),legend.box.background = element_rect(colour = "black"))
p2 <- ggplot(gData,aes(x=x,y=y,shape=Treatment,color=Treatment)) +
geom_point() + scale_shape_manual(values=1:4)
grid.arrange(p1,p2,nrow=1)

5.0 5.0

2.5 2.5
Treatment
Baseline
y

A
B
Treatment
0.0 0.0
Baseline
A
B

-2.5 -2.5
0 1 2 3 4 0 1 2 3 4
x x

When we use more than one type of points, we need a legend. Putting the legend inside
the graph saves space. However, a legend outside the graph produces less clutter (see also
1.6).
With lines, we need a legend too. It helps if labels follow the same order as lines (first
graph). Often the graph is easier to understand if we label the curves directly (second graph).

library(directlabels)
gData %>% mutate(Treatment=reorder(Treatment,-y,max)) %>%
ggplot(aes(x=x,y=y,lty=Treatment,color=Treatment)) + geom_smooth(se=FALSE) -> p
grid.arrange(p,
direct.label(p,list("far.from.others.borders","calc.boxes",
"enlarge.box","draw.rects")),nrow=1)
 © Oliver Kirchkamp
22 Using Graphs and Visualising Data — 1 INTRODUCTION

3 3
A

2 Treatment 2
B
y

y
A
1 Baseline 1

B
0 0

e
lin
se
Ba
0 1 2 3 4 0 1 2 3 4
x x

1.6 Clutter
The following graph presents too many things in one graph.

set.seed(123)
N<-24
data.frame(y=rnorm(N),
s=sqrt(abs(rnorm(N))),
group=factor(rep(1:2,length.out=N),label=c("Baseline","Treatment")),
x=rep(1:(N/2),each=2)) %>%
mutate(ymin=y-s,
ymax=y+s) -> dataCl

ggplot(dataCl,aes(x=x,y=y,lty=group)) + geom_line() + geom_errorbar(aes(ymin=ymin,ymax=ymax))

1 group
Baseline
y

0
Treatment
-1

-2

0.0 2.5 5.0 7.5 10.0 12.5

x

Perhaps splitting the information into several graphs can help:

 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 23

ggplot(dataCl,aes(x=x,y=y,lty=group)) + geom_line() + theme(legend.position="top") -> p1

dataCl %>%
ggplot(aes(x=x,y=y,ymin=ymin,ymax=ymax)) + geom_point() + geom_errorbar() +
facet_grid(. ~ group) -> p2
grid.arrange(p1,p2,nrow=1)

Baseline Treatment
group Baseline Treatment
3

1
1

y
0 0
y

-1
-1

-2

-2
2.5 5.0 7.5 10.0 12.5 0.0 2.5 5.0 7.5 10.0 12.50.0 2.5 5.0 7.5 10.0 12.5
x x

1.7 Unnecessary 3D
1.0

1.0
0.0

0.0

Unnecessary 3D effects distract from the content of your graph. Is the bar in the graph on
the left larger or smaller than 1.0? Of course, one can work it out, but a simple dot without
3D (on the right) is much easier to understand.

1.8 Aspect ratio

Less can be more — 45◦
 © Oliver Kirchkamp
24 Using Graphs and Visualising Data — 1 INTRODUCTION

sunsp <- data.frame(count=as.numeric(sunspot.year),year=time(sunspot.year))

p1 <- ggplot(sunsp,aes(x=year,y=count)) + geom_line()
p2 <- ggplot(sunsp,aes(x=year,y=count)) + geom_line() + coord_fixed(.055)
grid.arrange(p1,p2,nrow=1)

150
count

count
100 150
100
500
1700 1800 1900 2000
year
50

1700 1800 1900 2000

year

The graph on the right might be more informative than the one on the left. We can see
that activity increases more quickly than it decreases. This is less obvious in the left graph.
If we feel that the right graph is too flat then we can ‘cut-and-stack’ it as follows:

library(latticeExtra)
xyplot(sunspot.year,aspect="xy",strip=FALSE,strip.left=TRUE,
cut=list(number=3,overlap=0.05))

150
time

100
50
0

1700 1720 1740 1760 1780 1800

150
time

100
50
0

1800 1820 1840 1860 1880

150
time

100
50
0

1900 1920 1940 1960 1980

Time
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 25

In the following example the graph on the left has a slope of about 45◦ . This makes it easier
to see the convexity of the curve.

lco2<-data.frame(lowess(co2))
plot(xyplot(y~x,data=lco2,type="l",aspect="xy"),position=c(0,0,.5,1),more=TRUE)
plot(xyplot(y~x,data=lco2,type="l",aspect=.1),position=c(.5,0,1,1))

360

350

360
340 340
y

y
320

330 1960 1970 1980 1990

x
320

1960 1970 1980 1990

x

• Assume different parts of a graph have slopes s and s · (1 + ϵ).

• We are interested in the differences of the slopes.

∆ = arctan s − arctan(s · (1 + ϵ))

∂∆ s ϵ→0 s
= =
∂ϵ 1 + (1 + ϵ ) · s
2 2 1 + s2
0.4
1 + s2
s

0.2
0.0

0 15 30 45 60 75 90
arctan s
 © Oliver Kirchkamp
26 Using Graphs and Visualising Data — 1 INTRODUCTION

Hence, if we want to see differences in slopes, we should scale the graph such that slopes
are close to 1.
Also in the following graph lines have a slope of about 45◦ . This makes it easier to compare
the different slopes. ‘<

library(pwt10)
N<-6
data(pwt10.0)
pwt10.0 %>%
semi_join(pwt10.0 %>% ## find N most populous countries:
group_by(country) %>%
summarise(popM=median(pop,na.rm=TRUE)) %>%
arrange(-popM) %>%
top_n(N)) %>%
mutate(country=reorder(factor(substr(country,1,10)),-cgdpo/pop,median,na.rm=TRUE),
gdp = cgdpo/pop) -> pwt6

xyplot(gdp ~ year,group=country,
scales=list(y=list(log=10)),
yscale.components=yscale.components.log10.3,
aspect="xy",
type="l",data=pwt6,
auto.key=list(space="right",points=FALSE,lines=TRUE),
ylab="real GDPo/head (US\\$)",xlab="year")
real GDPo/head (US$)

30000 United Sta

Russian Fe
10000 Brazil
Indonesia
3000 China
India
1000

1960 2000
year
45◦ with Lattice

xyplot(gdp ~ year,group=country,
scales=list(y=list(log=10)),
yscale.components=yscale.components.log10.3,
type="l",data=pwt6,
ylab="real GDPo/head (US\\$)",xlab="year")
##
##
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 27

real GDPo/head (US$)

30000

10000

3000

1000

1960 1980 2000 2020

year

When we stretch the graph to fill the entire page, convexities and concavities are harder
to see:

1.9 What to present

1.9.1 Structuring content

layout(matrix(data=c(1,1,2,3),2,2),heights=c(.7,.3))
plot(co2, ylab = "Atmospheric concentration of CO$_2$",las = 1)
lco2 <- lowess(co2)
plot(lco2,t="l",ylab="lowess",xlab="")
plot(co2-lco2$y,t='l',ylab='seasonal')

Mauna Loa Atmospheric CO2 Concentration

360
Atmospheric concentration of CO2
360

350
lowess

340
350

330
340

320
330

1960 1970 1980 1990

320

seasonal

2
-4
1960 1970 1980 1990
1960 1970 1980 1990
Time

Sometimes it helps to show residuals in a separate graph. In the following example only
the graph on the right shows that noise increases for large values of x. The graph on the left
does not reveal this structure.
 © Oliver Kirchkamp
28 Using Graphs and Visualising Data — 1 INTRODUCTION

N <- 200
x <- sort(runif(N))
y0 <- 20 * exp(x)^3
y <- y0 +4*rnorm(N,sd=.5+x)
residuals <- y0-y
plot(lattice::xyplot(y~x,panel=function(...) {
panel.xyplot(...);
panel.xyplot(x=x,y=y0,type="l")}),
position=c(0,0,.5,1),more=TRUE)
plot(xyplot(residuals ~ x),position=c(.5,0,1,1))

400

10

300
5

residuals
200
y

100 -5

-10

0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
x x

1.9.2 Don’t discard parts of your data

Since temperature is measured as integers, we first have to aggregate, so that we can plot
superimposed points from two different flights in a different way.

library(vcd)
data(SpaceShuttle)
par(mfrow=c(1,2),mar=c(4,4.5,4,0),mex=.5)
xx<-with(SpaceShuttle,aggregate(Temperature,list(Temperature=Temperature,
Failures=nFailures),length))
plot(Failures ~ Temperature,data=xx,cex=sqrt(x),
yaxt="n",xlab="Temperature/[\\degree F]",ylab="Failures",
main="all previous flights")
axis(2,at=0:2)
plot(Failures ~ Temperature,data=xx,subset=Failures>0,cex=sqrt(x),
yaxt="n",xlab="Temperature/[\\degree F]",ylab="Failures",
main="only flights with failures")
axis(2,at=1:2)
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 29

all previous flights only flights with failures

2

2
Failures

Failures
1
0

1
55 60 65 70 75 80 55 60 65 70 75
Temperature/[◦ F] Temperature/[◦ F]

Data from S. Dalal, E. B. Fowlkes, B. Hoadly (1989).

This example illustrates the dire consequences of discarding “irrelevant” data: Previous
to the crash of the space shuttle Challenger on 28 January 1986 engineers noticed that the
temperature was much lower (31◦ F) than at other launches before (53◦ F to 81◦ F). NASA
managers considered only the failures of O-rings from previous flights (diagram on the right),
discarding the non-failures. They did not see any pattern in the failures and continued the
countdown.2
Sizes of the symbols are proptional to the number of observations.
An alternative way to present this data is a conditional density plot:
cdplot(as.factor(nFailures) ~ Temperature,ylab="Failures",
xlab="Temperature/[\\degree F]",data=SpaceShuttle)

1.0
0

0.8
0.6
Failures
1

0.4
0.2
2

0.0

55 60 65 70 75 80
◦
Temperature/[ F]

2
Data from S. Dalal, E. B. Fowlkes, B. Hoadly (1989), Risk analysis of the space shuttle: Pre-Challenger pre-
diction of failure, Journal of the American Statistical Association, *84*, 945-957.
 © Oliver Kirchkamp
30 Using Graphs and Visualising Data — 1 INTRODUCTION

1.9.3 Projecting data

Carl Sagan3 argues that intelligence has something to do with the weight of the brain and
the weight of the body. We are supposed to see this from a graph which is similar to the
following:

library(MASS)
data(Animals)
plot(brain ~ body,data=Animals,log="xy")
with(Animals,thigmophobe.labels(body,brain,rownames(Animals),cex=.5))
5000.0

African elephant
Asian elephant
Human
Giraffe
Chimpanzee
500.0

Donkey Gorilla
Horse
Cow
Rhesus monkey Sheep Jaguar Brachiosaurus
Potar monkey Grey wolf
Goat
Pig
brain

Triceratops
50.0

Cat Dipliodocus
Kangaroo

Mountain beaver
Rabbit
5.0

Mole Guinea pig

Rat
0.5

Golden hamster
Mouse

0.1 1.0 10.0 100.0 1000.0 10000.0 100000.0

body

This is too complicated. Stephen Jay Gould4 :

• brain size should be proportional to the surface of the body

• surface grows quadratically with height volume (and weight) grows cubically

→ weight of the brain ∼ weight of the body2/3

2
excess brain = log(brain mass) −
log(body mass)
3
To make it easier to interpret this difference of logs we use logarithms with base 10.
The left graph is ordered by the quantity of “excess brain”, the right one is ordered alpha-
betically. Often dotplots are easier to understand when they are sorted by the quantity.

3
The Dragons of Eden: Speculations on the Evolution of Human Intelligence. Random House, New York, 1977.
4
Ever Since Darwin: Reflections in Natural History. Norton, New York, 1977.
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 31

excess <- with(Animals,log10(brain)-2/3*log10(body))

xx<-data.frame(list(Animals=reorder(rownames(Animals),excess,median),excess=excess))
library(lattice)
plot(dotplot(Animals ~ excess,data=xx),more=TRUE,position=c(0,0,.5,1))
xx<-data.frame(list(Animals=reorder(rownames(Animals),
-as.numeric(factor(rownames(Animals))),median),excess=excess))
plot(dotplot(Animals ~ excess,data=xx),position=c(.5,0,1,1))

Human African elephant

Rhesus monkey Asian elephant
Chimpanzee Brachiosaurus
Potar monkey Cat
Asian elephant Chimpanzee
African elephant Cow
Donkey Dipliodocus
Goat Donkey
Mole Giraffe
Sheep Goat
Gorilla Golden hamster
Cat Gorilla
Grey wolf Grey wolf
Giraffe Guinea pig
Horse Horse
Jaguar Human
Cow Jaguar
Mountain beaver Kangaroo
Rabbit Mole
Pig Mountain beaver
Guinea pig Mouse
Kangaroo Pig
Mouse Potar monkey
Rat Rabbit
Golden hamster Rat
Triceratops Rhesus monkey
Dipliodocus Sheep
Brachiosaurus Triceratops

-1.0 -0.5 0.0 0.5 1.0 1.5 2.0 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
excess excess

Also for a multiway dotplot ordering by quantity helps. In the following example we use
medians of the different categories.

data(pwt10.0)
N <- 12
pwt10.0 %>%
semi_join(pwt10.0 %>% ## find N most populous countries:
group_by(country) %>%
summarise(popM=median(pop,na.rm=TRUE)) %>%
arrange(-popM) %>%
top_n(N)) %>%
filter(year>max(year)-6) %>%
mutate(gdp = cgdpo/pop,
country = reorder(factor(substr(country,1,10)),-gdp,median,na.rm=TRUE)) -> pwt12

ggplot(pwt12,aes(y=country,x=gdp)) + geom_point() + facet_wrap(vars(year),nrow=1) +

scale_x_log10() + labs(x="GDPo/head (US\\$)")
ggplot(pwt12,aes(y=gdp,x=year,color=country,lty=country)) + geom_line() +
scale_y_log10() + labs(y="GDPo/head (US\\$)")
 © Oliver Kirchkamp
32 Using Graphs and Visualising Data — 1 INTRODUCTION

2014 2015 2016 2017 2018 2019

Bangladesh

Pakistan

Nigeria

India

Indonesia
country

China

Brazil

Mexico

Russian Fe

Japan

Germany

United Sta

500010000 30000
50000 500010000 30000
50000 500010000 30000
50000 500010000 30000
50000 500010000 30000
50000 500010000 30000
50000
GDPo/head (US$)

50000
country
United Sta
Germany
30000
Japan
GDPo/head (US$)

Russian Fe
Mexico
Brazil
China
10000 Indonesia
India
Nigeria
5000 Pakistan
Bangladesh

2014 2015 2016 2017 2018 2019

year

1.9.4 Differences

par(mfrow=c(2,1),mex=.5)
x <- seq(-1,5,.01)
y <- sin(x)^6
dy <- abs(6*cos(x)^2*sin(x)^6)
plot(y ~ x,t="l",ylim=c(0,1.2))
lines(x,y+.1*dy+.1,lty=2)
legend("topleft",c("y'","y"),lty=c(2,1),cex=.5)
plot(x,+.05*dy+.1,t="l",ylab="y' - y",xlab="x")
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 33

1.2
y’
y

0.8
y
0.4
0.0
-1 0 1 2 3 4 5
x
0.120
y’ - y
0.100

-1 0 1 2 3 4 5
x

In the top graph it is difficult to assess the difference between the two curves.
If it is the difference that is interesting, then then graph should show the difference (bottom
graph).

2 ggplot
R provides a number of ways to create graphs. The most basic is perhaps the built in plot.
More powerful ones are lattice and ggplot2. Here we use ggplot2 as a starting point. In
this chapter we want to explain how some standard graphs can be created with ggplot2.

2.1 Elements of ggplot

The iris data For our examples we need some data. One standard data set is the iris data.

iris

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

1 5.1 3.5 1.4 0.2 setosa
2 4.9 3.0 1.4 0.2 setosa
3 4.7 3.2 1.3 0.2 setosa
4 4.6 3.1 1.5 0.2 setosa
5 5.0 3.6 1.4 0.2 setosa
6 5.4 3.9 1.7 0.4 setosa
7 4.6 3.4 1.4 0.3 setosa
8 5.0 3.4 1.5 0.2 setosa
[ reached 'max' / getOption("max.print") -- omitted 142 rows ]

Anderson, Edgar (1935). The irises of the Gaspe Peninsula, Bulletin of the American Iris
Society, 59, 2-5.
 © Oliver Kirchkamp
34 Using Graphs and Visualising Data — 2 GGPLOT

aes and geom:

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width)) +
geom_point()

4.5

4.0
Sepal.Width

3.5

3.0

2.5

2.0
5 6 7 8
Sepal.Length

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width)) +
geom_smooth()

3.50
Sepal.Width

3.25

3.00

2.75

2.50
5 6 7 8
Sepal.Length

ggplot(iris) +
geom_point(aes(x=Sepal.Length,y=Sepal.Width))
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 35

4.5

4.0
Sepal.Width

3.5

3.0

2.5

2.0
5 6 7 8
Sepal.Length

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width)) +
geom_point() + geom_smooth()

4.5

4.0
Sepal.Width

3.5

3.0

2.5

2.0
5 6 7 8
Sepal.Length

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width)) +
geom_point(aes(color=Species)) + geom_smooth()

4.5

4.0

Species
Sepal.Width

3.5
setosa

3.0 versicolor
virginica

2.5

2.0
5 6 7 8
Sepal.Length
 © Oliver Kirchkamp
36 Using Graphs and Visualising Data — 2 GGPLOT

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width)) +
geom_point() + geom_smooth(aes(color=Species))

4.5

4.0

Species
Sepal.Width

3.5
setosa

3.0 versicolor
virginica
2.5

2.0

5 6 7 8
Sepal.Length

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width)) +
stat_identity(geom="point")

4.5

4.0
Sepal.Width

3.5

3.0

2.5

2.0
5 6 7 8
Sepal.Length

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width)) +
geom_point(stat="identity")
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 37

4.5

4.0
Sepal.Width

3.5

3.0

2.5

2.0
5 6 7 8
Sepal.Length

ggplot(iris,aes(x=Sepal.Width,color=Species)) +
stat_density()

Species
2
density

setosa
versicolor

1 virginica

0
2.0 2.5 3.0 3.5 4.0 4.5
Sepal.Width

ggplot(iris,aes(x=Sepal.Width,color=Species)) +
stat_density(geom="line")

Species
2
density

setosa
versicolor

1 virginica

0
2.0 2.5 3.0 3.5 4.0 4.5
Sepal.Width
 © Oliver Kirchkamp
38 Using Graphs and Visualising Data — 2 GGPLOT

2.2 Labels and legends

mtcars

mpg cyl disp hp drat wt qsec vs am gear carb

Mazda RX4 21.0 6 160 110 3.90 2.620 16.46 0 1 4 4
Mazda RX4 Wag 21.0 6 160 110 3.90 2.875 17.02 0 1 4 4
Datsun 710 22.8 4 108 93 3.85 2.320 18.61 1 1 4 1
[ reached 'max' / getOption("max.print") -- omitted 29 rows ]

Data from Motor Trend. 1974.

ggplot(mtcars,aes(x=hp,y=wt,color=factor(cyl))) +
geom_point()

factor(cyl)
4
4
wt

6
3
8

100 200 300

hp

ggplot(mtcars,aes(x=hp,y=wt,color=factor(cyl))) +
geom_point() + theme(legend.position="top")

factor(cyl) 4 6 8

4
wt

100 200 300

hp

ggplot(mtcars,aes(x=hp,y=wt,color=factor(cyl))) +
geom_point() +
labs(color="Cyl.",x="Power (hp)",
y="Weight (1000 lbs)")
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 39

5
Weight (1000 lbs)

Cyl.
4
4
6
3
8

100 200 300

Power (hp)

ggplot(mtcars,aes(x=hp,y=wt,color=factor(cyl))) +
geom_point() +
labs(color=NULL,x=NULL,y=NULL)

4
4
6
3 8

100 200 300

ggplot(mtcars,aes(x=hp,y=wt,color=factor(cyl))) +
geom_point() +
theme(legend.position="none")

4
wt

100 200 300

hp
 © Oliver Kirchkamp
40 Using Graphs and Visualising Data — 2 GGPLOT

ggplot(mtcars,aes(x=hp,y=wt,color=factor(cyl))) +
geom_point() +
theme(legend.background=element_rect(
fill="gray",color="black"))

factor(cyl)
4
4
wt

6
3
8

100 200 300

hp

ggplot(mtcars,aes(x=hp,y=wt,color=factor(cyl),
shape=factor(gear))) +
geom_point()

factor(gear)
5
3
4
4 5
wt

3 factor(cyl)
4
2 6
8

100 200 300

hp

ggplot(mtcars,aes(x=hp,y=wt,color=factor(cyl),
shape=factor(gear))) +
geom_point() +
guides(color=guide_legend(order=1))
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 41

factor(cyl)
5
4
6
4 8
wt

3 factor(gear)
3
2 4
5

100 200 300

hp

ggplot(mtcars,aes(x=hp,y=wt,color=factor(cyl),
shape=factor(gear))) +
geom_point() + guides(color="none")

factor(gear)
4
3
wt

4
3
5

100 200 300

hp

ggplot(mtcars,aes(x=hp,y=wt,color=factor(cyl),
shape=factor(gear))) +
geom_point() +
labs(shape="Gears",color="Cyl.")

Gears
5
3
4
4 5
wt

3 Cyl.
4
2 6
8

100 200 300

hp
 © Oliver Kirchkamp
42 Using Graphs and Visualising Data — 2 GGPLOT

2.3 Scatterplots
library(pwt10)
data(pwt10.0)

data(pwt10.0)
pwtYC <- function(years,countries) {
pwt10.0 %>%
semi_join(pwt10.0 %>% ## find N most populous countries:
group_by(country) %>%
summarise(popM=median(pop,na.rm=TRUE)) %>%
arrange(-popM) %>%
top_n(countries)) %>% ## only the ... largest countries
filter(year>max(year)-years) %>% ## only the last ... years
mutate(gdp = cgdpo/pop,
country = substr(country,1,10)) %>%
select(c("country","gdp","year","csh_c","csh_i","csh_g"))
}

Feenstra RC, Inklaar R, Timmer MP (2015). The Next Generation of the Penn World Table,
American Economic Review, 105(10). pp. 3150-82.
pwtYC(years=6, countries=6)

country gdp year csh_c csh_i csh_g

BRA-2014 Brazil 16099.69 2014 0.6338230 0.2311709 0.1671210
BRA-2015 Brazil 15005.83 2015 0.6415980 0.1949314 0.1758238
BRA-2016 Brazil 14154.57 2016 0.6427402 0.1694520 0.1856983
BRA-2017 Brazil 14279.43 2017 0.6361969 0.1696125 0.1872893
BRA-2018 Brazil 14514.13 2018 0.6432136 0.1722161 0.1847648
BRA-2019 Brazil 14570.64 2019 0.6462484 0.1765482 0.1824011
[ reached 'max' / getOption("max.print") -- omitted 30 rows ]

ggplot(data=pwtYC(years=6, countries=6), aes(x=year,y=csh_i)) +

geom_line() + geom_point() +
facet_wrap( ~ reorder(country,csh_i)) +
labs(y="Share investment")

Russian Fe Brazil United Sta

0.4

0.3
Share investment

0.2

India Indonesia China

0.4

0.3

0.2

2014 2015 2016 2017 2018 2019 2014 2015 2016 2017 2018 2019 2014 2015 2016 2017 2018 2019
year
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 43

ggplot(data=pwtYC(years=6, countries=6), aes(y=reorder(country,csh_i),x=csh_i)) +

geom_point() +
facet_wrap( vars(year) ) +
labs(x="Share investment",y=NULL)

2014 2015 2016

China
Indonesia
India
United Sta
Brazil
Russian Fe

2017 2018 2019

China
Indonesia
India
United Sta
Brazil
Russian Fe
0.2 0.3 0.4 0.2 0.3 0.4 0.2 0.3 0.4
Share investment

We can simply draw (within ggplot) two lines with two geoms: …as several geoms:

pwtYC(99,6) %>% ggplot(aes(x=year,y=csh_i)) +

geom_line(aes(color="I")) +
geom_line(aes(y=csh_c,color="C")) +
facet_wrap(~reorder(country,csh_c)) +
labs(y="Share investment",color="type")

Brazil United Sta India

0.75
0.50
Share investment

0.25
type
China Indonesia Russian Fe C
I
0.75
0.50
0.25

1960 1980 2000 2020 1960 1980 2000 2020 1960 1980 2000 2020
year

Alternatively we reshape the data before ggplot.

pwtYC(99,6) %>% head(n=3)

country gdp year csh_c csh_i csh_g

BRA-1950 Brazil 1606.117 1950 0.6536519 0.1972710 0.1327842
BRA-1951 Brazil 1602.992 1951 0.6593236 0.2439667 0.1399094
BRA-1952 Brazil 1739.890 1952 0.6463201 0.2511176 0.1322049
 © Oliver Kirchkamp
44 Using Graphs and Visualising Data — 2 GGPLOT

pwtYC(99,6) %>% tidyr::pivot_longer(cols=starts_with("csh"),

names_to="type",
names_prefix="csh_",
values_to="share") -> pwtLong

pwtLong %>% slice_head(n=3)

# A tibble: 3 x 5
country gdp year type share
<chr> <dbl> <int> <chr> <dbl>
1 Brazil 1606. 1950 c 0.654
2 Brazil 1606. 1950 i 0.197
3 Brazil 1606. 1950 g 0.133

pwtLong %>% ggplot(aes(x=year,y=share,color=type)) +

facet_wrap(~reorder(country,share,max)) +
geom_line()

United Sta Brazil India

0.75
0.50
0.25 type
share

c
China Indonesia Russian Fe
g
0.75 i
0.50
0.25

1960 1980 2000 2020 1960 1980 2000 2020 1960 1980 2000 2020
year

pwtLong %>% ggplot(aes(x=year,y=share,lty=country)) +

geom_line() + facet_wrap(~type) + labs(lty="Country") +
theme(axis.text.x = element_text(angle=45,vjust=.5))

c g i

Country
0.75
Brazil
China
share

0.50
India
Indonesia
0.25 Russian Fe
United Sta
60

80

00

20

60

80

00

20

60

80

00

20
19

19

20

20

19

19

20

20

19

19

20

20

year
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 45

pwtLong %>% ggplot(aes(x=year,y=share,lty=country,color=country)) +

geom_line() + facet_wrap(~type) + labs(color="Country",lty="Country") +
theme(axis.text.x = element_text(angle=45,vjust=.5))

c g i

Country
0.75
Brazil
China
share

0.50
India
Indonesia
0.25 Russian Fe
United Sta
60

80

00

20

60

80

00

20

60

80

00

20
19

19

20

20

19

19

20

20

19

19

20

20
year

3 ggplot, more advanced features

3.1 Segment plots
Sometimes we plot segments. Here we plot a range of the minimum investment share to the
maximum investment share.

pwtYC(99,6) %>% group_by(country) %>%

summarise(min=min(csh_i,na.rm=TRUE),
max=max(csh_i,na.rm=TRUE),
mean=mean(csh_i,na.rm=TRUE)) %>%
ggplot(aes(y=country,xmin=min,xmax=max,x=mean)) +
geom_errorbar() + geom_point() +
labs(x="investment share",y=NULL)

United Sta

Russian Fe

Indonesia

India

China

Brazil

0.1 0.2 0.3 0.4

investment share
 © Oliver Kirchkamp
46 Using Graphs and Visualising Data — 3 GGPLOT, MORE ADVANCED FEATURES

ggplot(pwtYC(99,6),aes(y=country,x=csh_i)) +
stat_summary(fun.min=min,fun.max=max,fun=mean) +
labs(x="investment share",y=NULL)

United Sta

Russian Fe

Indonesia

India

China

Brazil

0.1 0.2 0.3 0.4

investment share

ggplot(pwtYC(99,6),aes(y=country,x=csh_i)) +
stat_summary(fun.min=min,fun.max=max,fun=mean,
geom="crossbar") +
labs(x="investment share",y=NULL)

United Sta

Russian Fe

Indonesia

India

China

Brazil

0.1 0.2 0.3 0.4

investment share

Please don’t do the following:

ggplot(pwtYC(99,6),aes(y=country,x=csh_i)) +
stat_summary(fun.min=min,fun.max=max,
geom="errorbar",width=.2) +
stat_summary(fun=mean,geom="bar",alpha=.3) +
labs(x="investment share",y=NULL)

The “bar” suggests that the elements of the bar have a meaning. This might sometimes
make sense, for example if the bar stands for something you can count. Most of the time
bars make no sense.
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 47

United Sta

Russian Fe

Indonesia

India

China

Brazil

0.0 0.1 0.2 0.3 0.4

investment share

Segment plots and regression results We can also use segment plots to show regression
results. In the following example we use the pwt6.3 dataset to study the relation between
openc and gdpc per country:

reg <- lm(log(gdp) ~ csh_i:country - 1,

data=pwtYC(99,6))
reg.ci<-data.frame(cbind(coef(reg),confint(reg)))
names(reg.ci)<-c("coef","lower","upper")
reg.ci[["country"]] <-
factor(sub("csh_i:country","",rownames(reg.ci)))
reg.ci<-within(reg.ci,
country<-reorder(country,coef))
ggplot(data=reg.ci,aes(y=country,x=coef)) +
geom_point() +
geom_errorbar(aes(xmin=lower,xmax=upper)) +
labs(x="95\\% CI for $\\beta_1$",y=NULL)

Russian Fe

United Sta

Brazil

Indonesia

India

China

25 30 35 40 45 50
95% CI for β1

3.2 Densityplots
 © Oliver Kirchkamp
48 Using Graphs and Visualising Data — 3 GGPLOT, MORE ADVANCED FEATURES

ggplot(iris) +
geom_density(aes(x=Sepal.Length,color="Sepal.Length")) +
geom_density(aes(x=Sepal.Width,color="Sepal.Width")) +
facet_wrap(~ Species) + labs(x="Length, Width")

setosa versicolor virginica

1.0
colour
density

Sepal.Length
0.5 Sepal.Width

0.0
2 4 6 8 2 4 6 8 2 4 6 8
Length, Width

3.3 Histograms
ggplot(iris) +
geom_histogram(aes(x=Sepal.Length,fill="Sepal.Length")) +
geom_histogram(aes(x=Sepal.Width,fill="Sepal.Width")) +
facet_wrap(~ Species) + labs(x="Length, Width")

setosa versicolor virginica

15

10 fill
count

Sepal.Length
Sepal.Width
5

0
2 4 6 8 2 4 6 8 2 4 6 8
Length, Width

3.4 Empirical cumulative distribution

ggplot(iris) +
stat_ecdf(aes(x=Sepal.Length,color="Sepal.Length")) +
stat_ecdf(aes(x=Sepal.Width,color="Sepal.Width")) +
facet_wrap(~ Species) + labs(x="Length, Width", y="ECDF")
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 49

setosa versicolor virginica

1.00

0.75
colour
ECDF

0.50 Sepal.Length
Sepal.Width

0.25

0.00
2 4 6 8 2 4 6 8 2 4 6 8
Length, Width

iris %>% slice_head(n=3)

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

1 5.1 3.5 1.4 0.2 setosa
2 4.9 3.0 1.4 0.2 setosa
3 4.7 3.2 1.3 0.2 setosa

iris %>% tidyr::pivot_longer(cols=starts_with("Sepal.")) %>%

slice_head(n=3)

# A tibble: 3 x 5
Petal.Length Petal.Width Species name value
<dbl> <dbl> <fct> <chr> <dbl>
1 1.4 0.2 setosa Sepal.Length 5.1
2 1.4 0.2 setosa Sepal.Width 3.5
3 1.4 0.2 setosa Sepal.Length 4.9

tidyr::pivot_longer(iris,cols=starts_with("Sepal.")) %>%
ggplot(aes(x=value,color=Species,lty=name)) +
stat_ecdf() + labs(lty="Measure",y="ECDF")

1.00

Measure
0.75 Sepal.Length
Sepal.Width
ECDF

0.50
Species
setosa
0.25 versicolor
virginica

0.00

2 4 6 8
value
 © Oliver Kirchkamp
50 Using Graphs and Visualising Data — 3 GGPLOT, MORE ADVANCED FEATURES

3.5 Q-Q plots

tidyr::pivot_longer(iris,cols=starts_with("Sepal.")) %>%
ggplot(aes(sample=value,color=Species,lty=name)) +
stat_qq() + stat_qq_line() + labs(y="Length, Width", x="Quantile")

Species
setosa
6 versicolor
Length, Width

virginica

4 name
Sepal.Length
Sepal.Width

-2 -1 0 1 2
Quantile

3.6 Sample Q-Q plots

An example:

data.frame(qqplot(0:100,1:5))

x y
1 0 1
2 25 2
3 50 3
4 75 4
5 100 5
5
4
1:5
3
2
1

0 20 40 60 80 100
0:100

We use the qqplot function to prepare data for ggplot:

 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 51

data(Wages,package="Ecdat")
Wages %>% mutate(edG = cut_number(ed,n=3)) %>%
group_by(edG) %>%
summarise(data.frame(qqplot(plot.it=FALSE,lwage[sex!="male"],lwage[sex=="male"]))) %>%
ggplot(aes(x=x,y=y)) + geom_line() + labs(x="female",y="male") +
geom_abline(slope=1,intercept=0) + facet_wrap(~edG)

[4,12] (12,14] (14,17]

7
male

5
4.5 5.0 5.5 6.0 6.5 7.0 4.5 5.0 5.5 6.0 6.5 7.0 4.5 5.0 5.5 6.0 6.5 7.0
female

3.7 Boxplots
Wages %>% mutate(edG = cut_number(ed,n=3)) %>%
ggplot(aes(y=lwage,x=sex)) + geom_boxplot() + facet_wrap(~edG)

[4,12] (12,14] (14,17]

7
lwage

female male female male female male

sex

3.8 Barcharts
ggplot can also do bar charts:
 © Oliver Kirchkamp
52 Using Graphs and Visualising Data — 3 GGPLOT, MORE ADVANCED FEATURES

pwtLong %>% filter(year > max(year)-6) %>%

ggplot(aes(y=country, x=share, fill=type)) +
geom_bar(stat="identity", color="black", position=position_dodge())+
theme_minimal() + facet_wrap(~year)

2014 2015 2016

United Sta
Russian Fe
Indonesia
India
type
China
Brazil
c
country

2017 2018 2019

g
United Sta
Russian Fe i
Indonesia
India
China
Brazil
0.0 0.2 0.4 0.6 0.0 0.2 0.4 0.6 0.0 0.2 0.4 0.6
share

We should note that often a dotplot or xyplot presents the same data in a better way.

pwtLong %>% filter(year > max(year)-6) %>%

ggplot(aes(x=reorder(country,share,max), y=share, group=type, color=type)) +
geom_point() + geom_line() + facet_wrap(~year) + coord_flip() + labs(x=NULL)

2014 2015 2016

United Sta
Brazil
India
Indonesia
Russian Fe
type
China
c
2017 2018 2019
g
United Sta i
Brazil
India
Indonesia
Russian Fe
China
0.2 0.4 0.6 0.2 0.4 0.6 0.2 0.4 0.6
share

(coord_flip, so that lines are drawn properly).

pwtLong %>% ggplot(aes(x=year,y=share,color=type)) +

facet_wrap(~reorder(country,share,max)) +
geom_line()
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 53

United Sta Brazil India

0.75

0.50

0.25
type
share

c
China Indonesia Russian Fe
g

0.75 i

0.50

0.25

1960 1980 2000 2020 1960 1980 2000 2020 1960 1980 2000 2020
year

3.9 Coplots

pwtLong %>% filter(year > max(year)-6) %>%

ggplot(aes(x=reorder(country,share,max), y=share, group=1)) +
geom_point() + geom_line() + facet_grid(type~year) + coord_flip() + labs(x=NULL)

2014 2015 2016 2017 2018 2019

United Sta
Brazil
India

c
Indonesia
Russian Fe
China
United Sta
Brazil
India

g
Indonesia
Russian Fe
China
United Sta
Brazil
India
i

Indonesia
Russian Fe
China
0.2 0.4 0.6 0.2 0.4 0.6 0.2 0.4 0.6 0.2 0.4 0.6 0.2 0.4 0.6 0.2 0.4 0.6
share

3.10 Parameters
3.10.1 Types of lines
With lattice we would choose between different types of lines with type. With ggplot
we use different geoms. In the following graph we use aes(color=...) to create a legend
for the different geoms.
 © Oliver Kirchkamp
54 Using Graphs and Visualising Data — 3 GGPLOT, MORE ADVANCED FEATURES

data(Caschool,package="Ecdat")
ggplot(data=Caschool,aes(x=avginc,y=testscr))+geom_point()+
geom_smooth(aes(color="loess",fill="loess",lty="loess"))+
geom_smooth(aes(color="lm",fill="lm",lty="lm"),method="lm") +
labs(fill="type",color="type",lty="type")

700
type
testscr

lm
loess
650

600
10 20 30 40 50
avginc

3.10.2 Axes
Different scales for different panels As with lattice, also ggplot chooses the same
scale for all panels in a plot. This can be changed with the help of the parameter scales in
facet_wrap.
Same scale (the default):

ggplot(pwtYC(99,6), aes(x=year,y=csh_g)) +
geom_line() +
facet_wrap(~reorder(country,csh_g)) +
labs(y="G")

United Sta India Brazil

0.3

0.2

0.1
G

China Indonesia Russian Fe

0.3

0.2

0.1

1960 1980 2000 2020 1960 1980 2000 2020 1960 1980 2000 2020
year

Free scale (scales=list(x="same",y="free")):

 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 55

ggplot(pwtYC(99,6), aes(x=year,y=csh_g)) +
geom_line() +
facet_wrap(~reorder(country,csh_g),scales="free_y") +
labs(y="G")

United Sta India Brazil

0.14 0.20 0.30

0.25
0.12 0.15 0.20
0.15
0.10 0.10 0.10
G

China Indonesia Russian Fe

0.20 0.35
0.225
0.200 0.15 0.30
0.175 0.25
0.10
0.150
0.20
0.125
1960 1980 2000 2020 1960 1980 2000 2020 1960 1980 2000 2020
year

Sliced scale (facet_grid(...,space='free'), scales have the same scale, but different
origin (this is different than in lattice):

ggplot(pwtYC(99,6), aes(x=year,y=csh_g)) +
geom_line() + facet_grid(.~reorder(country,csh_g),scales="free",space="free") +
labs(y="G") + coord_flip()

United Sta India Brazil China Indonesia Russian Fe

2020

2000
year

1980

1960

0.10
0.12
0.14 0.10 0.15 0.20 0.10 0.15 0.20 0.25 0.30 0.125
0.150
0.175
0.200
0.225 0.10 0.15 0.20 0.20 0.25 0.30 0.35
G

Individual axes
We can influence where an axis is labelled as follows:

ggplot(data=pwtYC(99,6), aes(x=year,y=csh_g)) +
geom_line() + facet_grid(.~reorder(country,csh_g)) + labs(y="G") +
 © Oliver Kirchkamp
56 Using Graphs and Visualising Data — 3 GGPLOT, MORE ADVANCED FEATURES

scale_y_log10(breaks=c(.07,.08,.09,.1,.12,.15,.2,.25,.3)) +
scale_x_continuous(breaks=c(1950,2000))

United Sta India Brazil China Indonesia Russian Fe

0.30
0.25

0.20
G

0.15

0.12
0.10
0.09
0.08
0.07

1950 2000 1950 2000 1950 2000 1950 2000 1950 2000 1950 2000
year

pwtYC(10,30) %>% group_by(country) %>%

summarise(i=median(csh_i,na.rm=TRUE),g=median(csh_g,na.rm=TRUE)) %>%
ggplot(aes(x=i,y=g,label=country)) + geom_label(size=3)

0.30 Ukraine

0.25 Russian Fe
Argentina
Poland
France
Myanmar
0.20 South Afri Japan Kin Iran (Isla
United Turkey
Thailand
Brazil
Mexico Spain
g

Colombia
Germany Italy
0.15 Republic o China
Egypt

United Sta Ethiopia Indonesia

0.10 Pakistan Philippine
Viet Nam
Nigeria India

Bangladesh
0.05
0.1 0.2 0.3 0.4
i

3.11 Zooming
Sometimes we want to show only part of the data. No problem if the graph shows nothing
but the data. If, however, the graph only shows statistics, e.g., a smooth line, the shape of
the line depends on the data that is included.
ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width)) +
geom_smooth() + geom_point() +
annotate("rect",xmin=4.5,xmax=5,ymin=2.9,
ymax=3.4,alpha=.3,fill="red")
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 57

4.5

4.0
Sepal.Width

3.5

3.0

2.5

2.0
5 6 7 8
Sepal.Length

The following graph uses only a subset of the data to calculate the smooth line.

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width)) +
geom_smooth() + geom_point() +
xlim(c(4.5,5)) + ylim(c(2.9,3.4))
#

3.4

3.3
Sepal.Width

3.2

3.1

3.0

2.9
4.5 4.6 4.7 4.8 4.9 5.0
Sepal.Length

The following graph uses the entire data to calculate the smooth line.

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width)) +
geom_smooth() + geom_point() +
coord_cartesian(xlim=c(4.5,5),ylim=c(2.9,3.4))
#
 © Oliver Kirchkamp
58 Using Graphs and Visualising Data — 3 GGPLOT, MORE ADVANCED FEATURES

3.4

3.3
Sepal.Width

3.2

3.1

3.0

2.9
4.5 4.6 4.7 4.8 4.9 5.0
Sepal.Length

3.12 Themes

pwtLong %>% filter(year > max(year)-6) %>%

ggplot(aes(x=reorder(country,share,max), y=share, group=type, shape=type, color=type)) +
labs(x=NULL) +
geom_point() + geom_line() + facet_wrap(~year) + coord_flip() -> p
p

2014 2015 2016

United Sta
Brazil
India
Indonesia
Russian Fe type
China
c
2017 2018 2019
g
United Sta
Brazil i
India
Indonesia
Russian Fe
China
0.2 0.4 0.6 0.2 0.4 0.6 0.2 0.4 0.6
share

p + theme_gray()
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 59

2014 2015 2016

United Sta
Brazil
India
Indonesia
type
Russian Fe
China
c

2017 2018 2019

g
United Sta
Brazil i
India
Indonesia
Russian Fe
China
0.2 0.4 0.6 0.2 0.4 0.6 0.2 0.4 0.6
share

p + theme_bw()

2014 2015 2016

United Sta
Brazil
India
Indonesia
type
Russian Fe
China
c

2017 2018 2019

g
United Sta
Brazil i
India
Indonesia
Russian Fe
China
0.2 0.4 0.6 0.2 0.4 0.6 0.2 0.4 0.6
share

p + theme_light()
 © Oliver Kirchkamp
60 Using Graphs and Visualising Data — 3 GGPLOT, MORE ADVANCED FEATURES

2014 2015 2016

United Sta
Brazil
India
Indonesia
type
Russian Fe
China
c

2017 2018 2019

g
United Sta
Brazil i
India
Indonesia
Russian Fe
China
0.2 0.4 0.6 0.2 0.4 0.6 0.2 0.4 0.6
share

ggthemes The ggthemes library offers a number of additional themes.

library(ggthemes)
p + theme_economist() + scale_colour_economist()

type c g i

2014 2015 2016

United Sta
Brazil
India
Indonesia
Russian Fe
China
2017 2018 2019
United Sta
Brazil
India
Indonesia
Russian Fe
China

0.2 0.4 0.6 0.2 0.4 0.6 0.2 0.4 0.6

share

library(ggthemes)
p + theme_solarized() + scale_colour_solarized("blue")
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 61

2014 2015 2016

United Sta
Brazil
India
Indonesia type
Russian Fe
China
c
2017 2018 2019
g
United Sta
Brazil i
India
Indonesia
Russian Fe
China
0.2 0.4 0.6 0.2 0.4 0.6 0.2 0.4 0.6
share

Setting the strip to a specific color

p + theme(strip.background=element_rect(fill="#8FC9C7"))

2014 2015 2016

United Sta
Brazil
India
Indonesia
Russian Fe
type
China
c
2017 2018 2019
g
United Sta i
Brazil
India
Indonesia
Russian Fe
China
0.2 0.4 0.6 0.2 0.4 0.6 0.2 0.4 0.6
share

Setting colors Parts of the plot which do not represent data can be influenced with theme_set().
If we also want to change the presentation of the data once and for all, we can redefine the
ggplot function:

ggplot <- function(...) ggplot2::ggplot(...) +

scale_fill_brewer(palette="Dark2") +
scale_color_brewer(palette="Dark2") +
scale_shape_manual(values=c(1,4,2,3,0,5:10))
 © Oliver Kirchkamp
62 Using Graphs and Visualising Data — 4 NOMINAL DATA

2014 2015 2016

United Sta
Brazil
India
Indonesia
Russian Fe type
China
c
2017 2018 2019
g
United Sta
Brazil i
India
Indonesia
Russian Fe
China
0.2 0.4 0.6 0.2 0.4 0.6 0.2 0.4 0.6
share

Lines will be drawn in a different color. Points will have a different shape.
If no colors are desired, then use scale_color_grey(end=0) and scale_fill_grey(end=0)

ggplot <- function(...) ggplot2::ggplot(...) +

scale_fill_grey(end=0) +
scale_color_grey(end=0) +
scale_shape_manual(values=c(1,4,2,3,0,5:10))

2014 2015 2016

United Sta
Brazil
India
Indonesia
Russian Fe type
China
c
2017 2018 2019
g
United Sta
Brazil i
India
Indonesia
Russian Fe
China
0.2 0.4 0.6 0.2 0.4 0.6 0.2 0.4 0.6
share

4 Nominal data
The case of purely nominal data is rare, although we might want present a simplified version
(where only nominal categories matter) of the data in the description.

4.1 Nominal univariate

set.seed(123)
nomD <- data.frame(n = rbinom(3,10,.3),
group = c("type A","type B","type C"))
nomD
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 63

n group
1 2 type A
2 4 type B
3 3 type C

ggplot(nomD,aes(x=group,y=n)) +
geom_point() + expand_limits(y=0)
#

2
n

0
type A type B type C
group

ggplot(nomD,aes(x=group,y=n,fill=group)) +
geom_bar(stat="identity") +
theme(legend.position="none")

2
n

0
type A type B type C
group

waste of ink
ggplot(nomD,aes(x="",y=n,fill=group)) +
geom_bar(stat="identity")
#

7.5
group
5.0 type A
n

type B
2.5 type C

0.0

x
 © Oliver Kirchkamp
64 Using Graphs and Visualising Data — 4 NOMINAL DATA

waste of ink, categories hard to compare

ggplot(nomD,aes(x=n,y="",fill=group)) +
geom_bar(stat="identity") +
coord_polar()

0.0

7.5 group
type A

y
2.5 type B
type C

5.0

waste of ink, categories very hard to compare

ggplot(nomD,aes(x=n,y=group)) +
geom_point() +
geom_segment(aes(x=0,xend=n,yend=group)) +
expand_limits(x=0)

type C
group

type B

type A

0 1 2 3 4
n

Juxtaposed bar charts

- waste of ink

Stacked bar charts

- waste of ink

- harder to compare values

Pie chart
- - The eye is not good at comparing angles (except 90◦ and 180◦ ).
Avoid pie charts (unless 90◦ and 180◦ are of special significance).
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 65

4.2 Nominal bivariate

• Barplots

• Bubbleplots

• Mosaicplots (Hartigan and Kleiner, 1984)

• Dot-plots

data(HairEyeColor)
reshape2::melt(HairEyeColor)

Hair Eye Sex value

1 Black Brown Male 32
2 Brown Brown Male 53
3 Red Brown Male 10
4 Blond Brown Male 3
5 Black Blue Male 11
6 Brown Blue Male 50
7 Red Blue Male 10
8 Blond Blue Male 30
9 Black Hazel Male 10
10 Brown Hazel Male 25
[ reached 'max' / getOption("max.print") -- omitted 22 rows ]

myColor <- c("black","#855700","red","#f4ebb3")

myFillSc <- scale_fill_manual(values=myColor)
myColSc <- scale_color_manual(values=myColor)

reshape2::melt(HairEyeColor) %>%
filter(Sex=="Female") %>%
ggplot(aes(x=Eye,y=value,fill=Hair)) +
geom_bar(stat='identity') + myFillSc

125

100
Hair
75 Black
value

Brown
50
Red
Blond
25

0
Brown Blue Hazel Green
Eye

reshape2::melt(HairEyeColor) %>%
filter(Sex=="Female") %>%
ggplot(aes(x=Eye,y=value,fill=Hair)) +
 © Oliver Kirchkamp
66 Using Graphs and Visualising Data — 4 NOMINAL DATA

geom_bar(stat='identity',position="dodge2") +
myFillSc

60

Hair
40 Black

value
Brown
Red
20
Blond

0
Brown Blue Hazel Green
Eye

library(ggmosaic)
reshape2::melt(HairEyeColor) %>%
filter(Sex=="Female") %>% ggplot() +
geom_mosaic(aes(x=product(Hair,Eye),
weight=value,fill=Hair)) +
myFillSc

Blond
Red

Hair
Black
Brown
Hair

Brown
Red
Blond
Black

Brown Blue HazelGreen

Eye

reshape2::melt(HairEyeColor) %>%
filter(Sex=="Female") %>% ggplot() +
geom_mosaic(aes(x=product(Eye,Hair),
weight=value,fill=Eye)) +
scale_fill_manual(values=c("brown","blue",
"#ffdd88","green"))
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 67

Green
Hazel

Blue Eye
Brown

Eye
Blue
Hazel
Brown
Green

Black Brown Red Blond

Hair

reshape2::melt(HairEyeColor) %>%
filter(Sex=="Female") %>%
ggplot(aes(x=Eye,y=Hair,size=value)) +
geom_point()

Blond

value
Red
20
Hair

40
Brown
60

Black

Brown Blue Hazel Green

Eye

reshape2::melt(HairEyeColor) %>%
filter(Sex=="Female") %>%
ggplot(aes(y=Hair,x=value)) +
geom_point() + facet_wrap(vars(Eye),nrow=1)

Brown Blue Hazel Green

Blond

Red
Hair

Brown

Black

0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60
value

reshape2::melt(HairEyeColor) %>%
filter(Sex=="Female") %>%
ggplot(aes(x=Eye,y=value,color=Hair)) +
geom_point() + myColSc
 © Oliver Kirchkamp
68 Using Graphs and Visualising Data — 4 NOMINAL DATA

60

Hair
40 Black

value
Brown
Red
20 Blond

0
Brown Blue Hazel Green
Eye

(Spineplots are very similar to mosaicplots)

Multiway dot-plots Multiway dot-plots are another possibility to present two way count
data:

HairEyeMale <- reshape2::melt(HairEyeColor[,,"Male"])

colEye<-c("brown","blue","sandybrown","green")
myTheme<-within(lTheme,dot.line$col<-colEye)
(d1<-dotplot(Eye ~ value | Hair,data=HairEyeMale,par.settings=myTheme))

10 20 30 40 50

Red Blond
Green

Hazel

Blue

Brown

Black Brown
Green

Hazel

Blue

Brown

10 20 30 40 50
value

colHair<-c("black","gold","brown","red")
myTheme<-within(lTheme,{
dot.line$col<-colEye
superpose.symbol$col<-colHair
superpose.line$col<-colHair})
keys<-list(space="top",columns=2,lines=TRUE)
(d2<-dotplot(Eye ~ value ,group=Hair,data=HairEyeMale,t=c("p","a"),auto.key=keys,par.settings=m
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 69

Black Red
Brown Blond

Green

Hazel

Blue

Brown

10 20 30 40 50
value

Juxtaposed barplot – hard to see a structure in the sub categories

Stacked barplot + easy to assess sum of observations in each category

Mosaicplot + easy to compare relative frequencies

+ easy to assess indepence of categories

Bubbleplot - hard to compare relative frequencies

+ good if the number of categories is large (in particular for numeric categories)

Dot-plot + easy to look up and to compare absolute frequencies

+ easy to see a pattern

4.3 Nominal multivariate

reshape2::melt(HairEyeColor) %>%
ggplot(aes(y=Hair,x=value)) +
geom_point() +
facet_grid(cols=vars(Eye),rows=vars(Sex))
 © Oliver Kirchkamp
70 Using Graphs and Visualising Data — 4 NOMINAL DATA

Brown Blue Hazel Green

Blond
Red

Male
Brown
Black

Hair
Blond

Female
Red
Brown
Black

0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60
value

reshape2::melt(HairEyeColor) %>% ggplot() +

geom_mosaic(aes(x=product(Sex,Hair,Eye),
weight=value,fill=Hair),
divider=mosaic("v"),offset=.05,
show.legend=FALSE) +
myFillSc

Female:Green
Male:Green
Female:Hazel
Male:Hazel

Female:Blue
Sex:Eye

Male:Blue

Female:Brown

Male:Brown

Black Brown Red Blond

Hair

reshape2::melt(HairEyeColor) %>% ggplot() +

geom_mosaic(aes(x=product(Hair,Eye),
weight=value,fill=Hair),
show.legend=FALSE,
divider=mosaic("v")) +
myFillSc + facet_grid(vars(Sex))

Green
Hazel
Male

Blue

Brown
Eye

Green
Hazel
Female

Blue

Brown

Black Brown RedBlond

Hair
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 71

5 Continuous data – distributions

5.1 Diagnostic plots for continuous variables

set.seed(123)
data.frame(x = rnorm(100,mean=12,sd=4)) %>%
ggplot(aes(sample=x)) + stat_qq() +
stat_qq_line() +
labs(x="Theoretical quantiles",
y="Sample quantiles")

20
Sample quantiles

15

10

-2 -1 0 1 2
Theoretical quantiles

mtcars %>%
ggplot(aes(sample=mpg)) +
stat_qq() + stat_qq_line()

35

30

25
y

20

15

10

-2 -1 0 1 2
x

mtcars %>%
ggplot(aes(sample=mpg,color=factor(cyl))) +
stat_qq() + stat_qq_line()
 © Oliver Kirchkamp
72 Using Graphs and Visualising Data — 5 CONTINUOUS DATA – DISTRIBUTIONS

30
factor(cyl)
4

y
6
20
8

10
-1 0 1
x

qqnorm compares with a given (theoretical) distribution. qqplot compares with a given
empirical distribution.

5.2 One continuous plus one nominal

5.2.1 Histograms

Histograms don’t let you see a difference:

iris %>% ggplot(aes(Sepal.Length)) +

geom_histogram(bins=11)

30

20
count

10

0
4 5 6 7 8
Sepal.Length

ggplot(iris,aes(Sepal.Length)) +
geom_histogram(bins=12,fill="gray",color="black")
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 73

20

15

count
10

0
4 5 6 7 8
Sepal.Length

ggplot(iris,aes(x=Sepal.Length,fill=Species)) +
geom_histogram()

12.5

10.0

Species
7.5
count

setosa

5.0 versicolor
virginica

2.5

0.0
5 6 7 8
Sepal.Length

ggplot(iris,aes(x=Sepal.Length,fill=Species)) +
geom_histogram(position="dodge")

7.5

Species
count

5.0 setosa
versicolor
virginica
2.5

0.0
5 6 7 8
Sepal.Length

5.2.2 Densities and conditional densities

 © Oliver Kirchkamp
74 Using Graphs and Visualising Data — 5 CONTINUOUS DATA – DISTRIBUTIONS

ggplot(iris,aes(x=Sepal.Length,color=Species)) +
geom_density()

1.2

0.8
Species

density
setosa
versicolor
0.4 virginica

0.0
5 6 7 8
Sepal.Length

ggplot(iris, aes(Sepal.Length,fill = Species))+

geom_density(position = "fill")

1.00

0.75
Species
density

setosa
0.50
versicolor
virginica
0.25

0.00
5 6 7 8
Sepal.Length

ggplot(iris, aes(Sepal.Length, y=Species, fill=Species)) +

geom_boxplot() + theme(legend.position="none")

virginica
Species

versicolor

setosa

5 6 7 8
Sepal.Length
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 75

Elements of the boxplot

• The median: usually a thick line in the middle
• The “box”, usually from the 25% to the 75% quantile
• The “whiskers”, usually to the most extreme data points which are not more than 1.5×
the interquartile range.
If the data is normally distributed, then the interquartile range covers 1.35 standard
deviations. Hence, 1.5× the interquartile range are 2.02 standard deviations. Outside
the whiskers we should, hence, observe 4.3% (or about 5%) of all observations.

5.2.3 Barplot of means

On the right you see a barplot of means. I show this type only for completeness. Avoid,
under all circumstances! You can show much more information in this space. On the right
you see, for comparison, a boxplot.
ggplot(iris,aes(x=Sepal.Length,y=Species)) +
stat_summary(fun=mean,geom="bar")

virginica
Species

versicolor

setosa

0 2 4 6
Sepal.Length

ggplot(iris, aes(Sepal.Length, y=Species, fill=Species)) +

geom_boxplot() + theme(legend.position="none")

virginica
Species

versicolor

setosa

5 6 7 8
Sepal.Length
 © Oliver Kirchkamp
76 Using Graphs and Visualising Data — 5 CONTINUOUS DATA – DISTRIBUTIONS

5.2.4 Means and standard deviation

Means and standard deviation are much less informative than boxplots. The following three
graphs all show the same four distributions which all have identical sample means and stan-
dard deviations (right diagram). Still, scattergrams (left) and boxplots (middle) reveal that
the four samples are quite different.

set.seed(123)
xx<-as.data.frame(cbind(seq(0,2,length=20),c(seq(-8,2,length=10),seq(4,6,length=10)),
c(seq(0,1.8,length=18),4,4.1),rnorm(20)))
for(i in 1:4) {xx[,i]<-(xx[,i]-mean(xx[,i]))/sd(xx[,i])}
xx<-reshape(xx,direction="long",v.names="x",varying=list(1:4))
xx<-within(xx,{time<-as.factor(time);levels(time)<-letters[1:4]})

ylim=range(c(xx$x))
par(mfrow=c(1,3))
with(xx,plot(x ~ as.integer(time),xaxt="n",ylim=ylim,xlab="",main="sample"))
axis(1,at=1:4,labels=letters[1:4])
boxplot(x ~ time,data=xx,ylim=ylim,main="boxplot")
library(plotrix)
dispData<-aggregate(x~time,
FUN=function(x) c(mean=mean(x),sd=sd(x)),
data=within(xx,time<-as.numeric(time)))
with(dispData,{
plot(x[,"mean"] ~ time,dispData,xaxt="n",ylim=ylim,xlab="",ylab="",
main="means and sample standard deviations")
dispersion(1:4,x[,"mean"],x[,"sd"])
})
axis(1,at=1:4,labels=letters[1:4])

sample boxplot means and sample standard deviations

2

2
1

1
x

x
0

0
-1

-1

-1
-2

-2

-2

a b c d a b c d a b c d

time

• Means and sample standard deviation may be misleading

• Boxplots provide more information, make fewer assumptions

 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 77

5.2.5 Empirical cumulative distributions

ggplot(iris,aes(Sepal.Length,color=Species)) +
stat_ecdf() +
labs(y="Empirical CDF")

1.00

0.75
Empirical CDF

Species
setosa
0.50
versicolor
virginica
0.25

0.00
5 6 7 8
Sepal.Length

5.3 More on Dot-plots

We use dot-plots if we have a small number of non-anonymous categories:

data(pwt10.0)
N <- 12
pwt10.0 %>%
semi_join(pwt10.0 %>% ## find N most populous countries:
group_by(country) %>%
summarise(popM=median(pop,
na.rm=TRUE)) %>%
arrange(-popM) %>%
top_n(N)) %>%
filter(year>max(year)-6) %>%
mutate(gdp = cgdpo/pop) %>%
select(c("country","gdp","year")) -> pwt12

pwt12

country gdp year

BGD-2014 Bangladesh 3478.825 2014
BGD-2015 Bangladesh 3736.502 2015
BGD-2016 Bangladesh 3847.664 2016
BGD-2017 Bangladesh 4113.227 2017
BGD-2018 Bangladesh 4421.308 2018
BGD-2019 Bangladesh 4652.617 2019
BRA-2014 Brazil 16099.694 2014
BRA-2015 Brazil 15005.833 2015
BRA-2016 Brazil 14154.573 2016
BRA-2017 Brazil 14279.429 2017
 © Oliver Kirchkamp
78 Using Graphs and Visualising Data — 5 CONTINUOUS DATA – DISTRIBUTIONS

BRA-2018 Brazil 14514.132 2018

BRA-2019 Brazil 14570.642 2019
CHN-2014 China 11733.252 2014
[ reached 'max' / getOption("max.print") -- omitted 59 rows ]

ggplot(pwt12,aes(y=country,x=gdp)) + geom_point() +
scale_x_log10() + labs(x="GDPo/head (US\\$)")

United States of America

Russian Federation
Pakistan
Nigeria
Mexico
country

Japan
India
Indonesia
Germany
China
Brazil
Bangladesh

5000 10000 30000 50000

GDPo/head (US$)

pwt12 %>%
mutate(country = reorder(factor(substr(country,1,10)),-gdp)) %>%
ggplot(aes(y=country,x=gdp)) + geom_point() +
scale_x_log10() + labs(x="GDPo/head (US\\$)")

Bangladesh
Pakistan
Nigeria
India
Indonesia
country

China
Brazil
Mexico
Russian Fe
Japan
Germany
United Sta

5000 10000 30000 50000

GDPo/head (US$)

Multiway dot-plots

pwt12 %>%
mutate(country = reorder(factor(substr(country,1,10)),-gdp)) %>%
ggplot(aes(y=country,x=gdp)) + geom_point() +
scale_x_log10() + labs(x="GDPo/head (US\\$)") +
facet_wrap(vars(year))
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 79

2014 2015 2016

Bangladesh
Pakistan
Nigeria
India
Indonesia
China
Brazil
Mexico
Russian Fe
Japan
Germany
United Sta
country

2017 2018 2019

Bangladesh
Pakistan
Nigeria
India
Indonesia
China
Brazil
Mexico
Russian Fe
Japan
Germany
United Sta
5000 10000 30000 50000 5000 10000 30000 50000 5000 10000 30000 50000
GDPo/head (US$)

5.4 Summary
Histograms
+ Everybody understands them
– Don’t reveal small differences
– Depend on breaks

Densities
+ Easy to understand
– Need assumptions (must be estimated)
– Depend on bandwidth

Conditional density plots

+ Easy to understand
+ Reveals even small differences between distributions
– Needs assumptions (must be estimated)

Boxplot
+ Shows summary statistics
– Aggregates data

Barplot of means
– Uses a lot of space to show a small amount of information.

ECDF
+ Provides a lot of information
+ Doesn’t depend much on parameters
 © Oliver Kirchkamp
80 Using Graphs and Visualising Data — 5 CONTINUOUS DATA – DISTRIBUTIONS

+ Reveals even small differences between distributions

– Not so easy to understand

Q-Q Plot
+ Provides a lot of information
+ Doesn’t depend much on parameters
+ Reveals even small differences between distributions
– Only compares two variables

Dot-Plot
+ Provides detailed information
+ Doesn’t depend much on parameters
– Requires a small number of observations

5.5 Two continuous variables

5.5.1 Scatterplot

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width,color=Species,shape=Species)) +
geom_point()

4.5

4.0

3.5 Species
Sepal.Width

setosa
versicolor
3.0
virginica

2.5

2.0

5 6 7 8
Sepal.Length

With larger data frames scatterplots might provide too much information:

5.5.2 Scatterplot with data ellipses

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width,color=Species,shape=Species)) +
geom_point() + ggpubr::stat_conf_ellipse(bary=FALSE)
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 81

4.5

4.0

Species
Sepal.Width

3.5
setosa

3.0 versicolor
virginica

2.5

2.0
4 5 6 7 8
Sepal.Length

library(car)
attach(iris)
xhist <- hist(Sepal.Width, breaks=10,plot=FALSE)
yhist <- hist(Sepal.Length, breaks=10,plot=FALSE)
xrange <- range(xhist$breaks)
yrange <- range(yhist$breaks)
layout(rbind(c(2,0),c(1,3)),
widths=c(4,1), heights=c(1,4))
par(mar=c(4,4,0,0))
plot(Sepal.Width, Sepal.Length,
xlim=xrange, ylim=yrange)
dataEllipse(Sepal.Width,Sepal.Length,
levels=c(.5,.95),plot.points=FALSE)
par(mar=c(0,4,0,0))
barplot(xhist$counts, axes=FALSE)
par(mar=c(4,0,0,0))
barplot(yhist$counts, axes=FALSE,
horiz=TRUE)
detach(iris)
 © Oliver Kirchkamp
82 Using Graphs and Visualising Data — 5 CONTINUOUS DATA – DISTRIBUTIONS

8
7
Sepal.Length

6
5
4

2.0 2.5 3.0 3.5 4.0

Sepal.Width

We use hist to calculate the range of the plot and to prepare the barplot:

attach(iris)
xhist <- hist(Sepal.Width, breaks=10,plot=FALSE)
yhist <- hist(Sepal.Length, breaks=10,plot=FALSE)
xhist$breaks

[1] 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 4.0 4.2 4.4

xhist$counts

[1] 4 7 13 23 36 24 18 10 9 3 2 1

detach(iris)

library(car)
attach(iris)
dataEllipse(Sepal.Length,Sepal.Width,
groups=Species,levels=c(.5,.95))
detach(iris)
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 83

setosa

setosa

4.0
virginica
versicolor

Sepal.Width
virginica

3.5
versicolor

3.0
2.5
2.0

4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0

Sepal.Length

library(car)
attach(iris)
dataEllipse(Sepal.Length,Sepal.Width,
groups=Species,levels=c(.5,.95),
draw=TRUE,plot.points=FALSE,add=FALSE)
detach(iris)

setosa

setosa
4.0

virginica
versicolor
Sepal.Width

virginica
3.5

versicolor
3.0
2.5
2.0

4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0

Sepal.Length

5.5.3 Bagplot

P. J. Rousseeuw, I. Ruts, J. W. Tukey (1999)

• The dark-blue area: The “bag”. This area contains 50% of all observations.

• The light-blue area: Contains all points which are in the bag 3 times expanded.

• Points outside the light-blue area are considered outliers.

 © Oliver Kirchkamp
84 Using Graphs and Visualising Data — 5 CONTINUOUS DATA – DISTRIBUTIONS

-10

y
-20

0 1
g

library(aplpack)
with(iris,bagplot(Sepal.Length,Sepal.Width))

4.0
3.5
y
3.0
2.5
2.0

4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0

x

5.5.4 Kernel densities

library(ks)
iris %>%
select(Sepal.Length,Sepal.Width,Species) ->
data2
dlply(data2,.(Species),
function(d) {
kde(d[,1:2])
}) -> kdeList
with(data2,
plot(Sepal.Length,Sepal.Width,cex=.2,
col="gray",pch=as.numeric(Species)))
for(i in 1:length(kdeList))
plot(kdeList[[i]],add=TRUE,lty=i,
col.fun=function(n){rainbow(n)})
legend("topright",
lty=1:length(kdeList),
pch=1:length(kdeList),
names(kdeList),cex=.5)
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 85

setosa
versicolor
virginica

4.0
50
25

3.5
Sepal.Width
75
50
75 25
50
3.0
75 25
2.5
2.0

4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0

Sepal.Length

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width,
color=Species,shape=Species)) +
geom_density_2d() +
geom_point()

4.5

4.0

Species
Sepal.Width

3.5
setosa

3.0 versicolor
virginica

2.5

2.0
5 6 7 8
Sepal.Length

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width)) +
stat_density_2d(aes(fill = ..level..),
geom = "polygon",
colour="white") +
scale_fill_distiller(palette= "Spectral")
 © Oliver Kirchkamp
86 Using Graphs and Visualising Data — 5 CONTINUOUS DATA – DISTRIBUTIONS

4.0
level
3.5

Sepal.Width
0.4

0.3
3.0
0.2

0.1
2.5

2.0
5 6 7 8
Sepal.Length

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width)) +
geom_bin2d(bins=10) +
scale_fill_continuous(type = "viridis")

4 count
Sepal.Width

10.0

7.5
3 5.0

2.5

4 5 6 7 8
Sepal.Length

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width)) +
geom_hex(bins=10) +
scale_fill_continuous(type = "viridis")

4.5

4.0
count
Sepal.Width

3.5 7.5

3.0 5.0

2.5
2.5

2.0

4 5 6 7
Sepal.Length
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 87

6 Continuous data, causal relations, other problems

6.1 Causal relations
6.1.1 Smooth lines

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width,
color=Species,shape=Species)) +
geom_point() + geom_smooth()

4.5

4.0

Species
Sepal.Width

3.5
setosa

3.0 versicolor
virginica
2.5

2.0

5 6 7 8
Sepal.Length

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width,
color=Species,shape=Species)) +
geom_point() + geom_smooth(method="lm")

4.5

4.0

Species
Sepal.Width

3.5
setosa

3.0 versicolor
virginica

2.5

2.0
5 6 7 8
Sepal.Length

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width,
color=Species,shape=Species)) +
geom_smooth(span=1)
 Using Graphs and Visualising Data — 6 CONTINUOUS DATA, CAUSAL RELATIONS,

© Oliver Kirchkamp
88 OTHER PROBLEMS

4.5

4.0

Species

Sepal.Width
3.5
setosa
3.0 versicolor
virginica
2.5

2.0

5 6 7 8
Sepal.Length
How smooth?

ggplot(iris,aes(x=Sepal.Length,y=Sepal.Width,
color=Species,shape=Species)) +
geom_smooth(span=.5)

Species
Sepal.Width

setosa
3 versicolor
virginica

5 6 7 8
Sepal.Length

6.1.2 GAM
Loess (locally estimated scatterplot smoothing) only relates one variable to a smooth function
of one other variables. What if there are more variables?
For more complex relationships (and as an extension of the linear model) we can use GAM
(generalised additive models).
Linear Regression:

Y = β0 + β1 X1 + β2 X2 + . . . + u
GAM (Generalised additive model):

Y = β0 + s1 (X1 ) + s2 (X2 ) + . . . + βk Xk . . . + u

est.ols <- lm(testscr ~ elpct + avginc + str,data=Caschool)

library(mgcv)
est.gam <- gam(testscr ~ s(elpct) + s(avginc) + str,data=Caschool)
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 89

Here is the output for the standard OLS model:

summary(est.ols)

Call:
lm(formula = testscr ~ elpct + avginc + str, data = Caschool)

Residuals:
Min 1Q Median 3Q Max
-42.800 -6.862 0.275 6.586 31.199

Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 640.31550 5.77489 110.879 <2e-16 ***
elpct -0.48827 0.02928 -16.674 <2e-16 ***
avginc 1.49452 0.07483 19.971 <2e-16 ***
str -0.06878 0.27691 -0.248 0.804
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Residual standard error: 10.35 on 416 degrees of freedom

Multiple R-squared: 0.7072,Adjusted R-squared: 0.7051
F-statistic: 334.9 on 3 and 416 DF, p-value: < 2.2e-16

The output for a GAM is similar to the output for OLS. Of course, the splines (here for
elpct and avginc) are not shown. The output provides only the result of an F-test and the
estimated degrees of freedom (edf).

summary(est.gam)

Family: gaussian
Link function: identity

Formula:
testscr ~ s(elpct) + s(avginc) + str

Parametric coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 656.9103 5.3040 123.852 <2e-16 ***
str -0.1402 0.2689 -0.521 0.602
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Approximate significance of smooth terms:

edf Ref.df F p-value
s(elpct) 2.416 3.023 87.53 <2e-16 ***
s(avginc) 3.171 3.983 116.57 <2e-16 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
 Using Graphs and Visualising Data — 6 CONTINUOUS DATA, CAUSAL RELATIONS,

© Oliver Kirchkamp
90 OTHER PROBLEMS

R-sq.(adj) = 0.731 Deviance explained = 73.5%

GCV = 99.459 Scale est. = 97.662 n = 420

Here are the functions s1 and s2 :

plot(est.gam,pages=0)

s(elpct,2.42)

20
-20

0 20 40 60 80

elpct
s(avginc,3.17)

20
-20

10 20 30 40 50

avginc

GAM permits interactions of splines:

library(mgcv)
est2.gam <- gam(testscr ~ s(elpct,avginc) + str,data=Caschool)

summary(est2.gam)

Family: gaussian
Link function: identity

Formula:
testscr ~ s(elpct, avginc) + str

Parametric coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 658.0740 5.3777 122.371 <2e-16 ***
str -0.1995 0.2728 -0.731 0.465
---
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 91

Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Approximate significance of smooth terms:

edf Ref.df F p-value
s(elpct,avginc) 18.96 23.79 47.72 <2e-16 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

R-sq.(adj) = 0.743 Deviance explained = 75.6%

GCV = 98.084 Scale est. = 93.19 n = 420

Now the spline is a smooth surface:

plot(est2.gam,pages=1,pers=TRUE,theta=0)
s(elpc
t,avgi

nc
nc,18

avgi
.96)

elpct

plot(est2.gam,pages=1,pers=TRUE,theta=40)
 Using Graphs and Visualising Data — 6 CONTINUOUS DATA, CAUSAL RELATIONS,

© Oliver Kirchkamp
92 OTHER PROBLEMS

s(elpc
t,a
vginc
,18.96
)
elp

c
gin
ct

av
plot(est2.gam,pages=1,pers=TRUE,theta=75,phi=5)
s(elpct,avginc,18.96)
elpct

avginc

The surface can also be shown as contours:

plot(est2.gam,pages=1)
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 93

-1se s(elpct,avginc,18.96) +1se

50

40
40

40
avginc

20
30
30

30
30
-10 -20
-30

-20
20 20
20
20

10 10
10
-10
-10
0
00
10

-20
-10
-20
-30
-30 -30 -40

0 20 40 60 80

elpct

6.1.3 Visually weighted Regression

source("~/R/vwreg.R")
set.seed(123)
CaPart<-Caschool[sample(1:nrow(Caschool),50),]

• Solomon Hsiang (2012). Visually weighted Regression.

• Felix Schonbrodt (2012): Implementation in R.

vwReg(testscr~avginc,data=CaPart,B=100,
spag=TRUE,slices=50)

690
testscr

660

630

600
10 20 30 40
avginc

Same situation with loess+standard deviation:

ggplot(CaPart,aes(x=avginc,y=testscr)) +
geom_smooth() +
geom_point(shape=1)
 Using Graphs and Visualising Data — 6 CONTINUOUS DATA, CAUSAL RELATIONS,

© Oliver Kirchkamp
94 OTHER PROBLEMS

725

700

675

testscr
650

625

600

10 20 30 40
avginc

6.1.4 Summary: Two continuous variables

Scatterplot
+ makes no assumptions
– with a large dataset the graph might be cluttered
– with categorical data points might superimpose

Data ellipses
– assumes a linear relationship

Bagplot
– not very well known

Kernel densities
+ easy to understand
– relies on assumptions (must be estimated, depend on bandwidth)

Regression line
– Assumes a linear causal relationship

Loess/GAM/VWReg
– Assume causal (not necessarily linear) relationship

6.2 Other problems

6.2.1 Paired data
Sometimes two-dimensional data comes in pairs where both elements can be compared with
each other. One value might be recorded before, the other after a treatment.
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 95

If the data are highly correlated then a standard scatterplot (left diagram) wastes a lot of
space top left and bottom right from the 45◦ -line.
The Tukey mean-difference plot (also known as Bland-Altman plot) basically rotates the
diagram by 45◦ and, thus, can save space. (This plot aims at showing agreement of the two
elements of the pairs and, hence, also shows the mean of the differences ± two standard
deviations.)
The bumpchart presents essentially the same information, but with a focus on the identity
of the observations. We would usually not do this for anonymous observations, but, e.g. if
observations are for countries or for cities.
Create some paired data:

set.seed(1)
N<-12
x<-runif(N)
y<-x+.1*rnorm(N)
pairD <- data.frame(x=x,y=y,g=letters[1:N])

pairD %>% ggplot(aes(x=x,y=y)) + geom_point() +

directlabels::geom_dl(aes(label=g),
method="smart.grid")

f g
0.75 d
c i
0.50
y

e b h
l a
0.25

j k
0.25 0.50 0.75
x

Bland-Altman / Tukey mean-difference:

pairD %>%
mutate(xyMean = (x+y)/2, yxDiff = (y-x)) %>%
ggplot(aes(x=xyMean,y=yxDiff)) + geom_point() +
labs(x="(x+y)/2",y="y-x") +
directlabels::geom_dl(aes(label=g),
method="smart.grid")
 Using Graphs and Visualising Data — 6 CONTINUOUS DATA, CAUSAL RELATIONS,

© Oliver Kirchkamp
96 OTHER PROBLEMS

0.1 e b i
l a c f
0.0
j k d
y-x

-0.1
g
-0.2 h
0.25 0.50 0.75
(x+y)/2

Bump-plot:
pairD %>% tidyr::pivot_longer(cols=c("x","y")) %>%
mutate(time=ifelse(name=="x",0,1)) %>%
ggplot(aes(x=time,y=value,color=g)) +
ggbump::geom_bump() +
geom_text(data=pairD,aes(x=-.01,y=x,label=g)) +
geom_text(data=pairD,aes(x=1.01,y=y,label=g)) +
theme(legend.position="none")

g
d f
f g
d
0.75 i
hi c
c
value

0.50
b
h
b
ae
0.25 a l
kel k
j j
0.00 0.25 0.50 0.75 1.00
time

6.2.2 Three-dimensional simplex

Three dimensional variables are notoriously difficult to present. However, quantities like
prices and probabilities can often be conveniently represented in a simplex:
set.seed(123)
data3 <-matrix(runif(150),ncol=3)
type<-data3[,1]>.5 # <- add groups
data3<-data3/rowSums(data3) # <- normalise
colnames(data3)<-c("x","y","z")
data3

x y z
[1,] 0.30809754 0.0491014380 0.64280102
[2,] 0.50424783 0.2828580196 0.21289415
[3,] 0.24106888 0.4709212352 0.28800989
[4,] 0.45065923 0.0622128465 0.48712793
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 97

[5,] 0.47394996 0.2826906163 0.24335942

[6,] 0.03987656 0.1807812499 0.77934219
[7,] 0.33635678 0.0812264534 0.58241676
[8,] 0.39584570 0.3341408730 0.27001343
[9,] 0.29692218 0.4819404184 0.22113740
[10,] 0.46680404 0.3828188688 0.15037709
[11,] 0.37416520 0.2600901850 0.36574461
[12,] 0.53370870 0.1116555756 0.35463572
[13,] 0.60375503 0.3421393873 0.05410558
[ reached getOption("max.print") -- omitted 37 rows ]

triax.plot(data3,show.grid=TRUE,pch=16*type+1,
cex.ticks=.7)
legend(1,1,c("Treatment A","Treatment B"),
pch=c(1,17))

Treatment A
Treatment B
0 .1

0.9
0
0.3 .2

0.8
0.7
0 .4

0.6
0
y
z

0 .5

0.5
0.7 .6

0.4
0.3
0.9 .8

0.2
0

0.1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1

6.2.3 Stars

stars(mtcars[, 1:7], len = 0.8, key.loc = c(12, 1.5),draw.segments = TRUE,cex=.5)

 © Oliver Kirchkamp
98 Using Graphs and Visualising Data — 7 LATTICE

Mazda RX4 Wag Hornet 4 Drive Valiant

Mazda RX4 Datsun 710 Hornet Sportabout

Merc 240D Merc 280 Merc 450SE

Duster 360 Merc 230 Merc 280C

Merc 450SLC Lincoln Continental Fiat 128

Merc 450SL Cadillac Fleetwood Chrysler Imperial

Toyota Corolla Dodge Challenger Camaro Z28

Honda Civic Toyota Corona AMC Javelin

Fiat X1-9 Lotus Europa Ferrari Dino

Pontiac Firebird Porsche 914-2 Ford Pantera L

disp cyl
mpg
Volvo 142E hp
Maserati Bora qsec
drat wt

7 Lattice
7.1 Multiway xyplots
Sometimes we want to display one type of diagram separately for different levels of a factor.
Here is an example:
Example: development of investment share (ci) over time (year), separately for each
country.
Get a subset of the data (six largest countries, later than 2001) from the Penn World Table:

library(pwt)
lattice.options(default.args=list(as.table=TRUE))
data(pwt6.3)

Add average population to data:

xx<-with(pwt6.3,aggregate(pop,list(country=country),mean))
xx<-subset(xx,country!="China Version 2")
N<-6
xx<-subset(xx,x>=-sort(-xx[["x"]])[N])
xx<-merge(xx,pwt6.3)
xx<-subset(xx,year>2001)

Give two countries a shorter name:

levels(xx$country)[grep("United States",levels(xx$country))]<-"U.S.A."
levels(xx$country)[grep("China",levels(xx$country))]<-"China"

reorder the countries. The order of the factor is used later in the plots. Here we order
according to the median of ci:
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 99

xx<-within(xx,country<-reorder(factor(country),xx$ci,
function(x) -median(x,na.rm=TRUE)))

Sorting the data along year and country makes it easier to draw connect lines lateron:

xx.pw<-xx[with(xx,order(country,year)),]

xx.pw[,c("country","year","pop","ci")]

country year pop ci

89 China 2002 1284275.9 28.54705
90 China 2003 1291496.0 31.11946
77 China 2004 1298847.6 32.79388
78 China 2005 1306313.8 32.71553
79 China 2006 1313973.7 33.04857
80 China 2007 1321851.9 33.25413
296 U.S.A. 2002 287501.5 24.76686
297 U.S.A. 2003 289985.8 25.41723
298 U.S.A. 2004 292805.6 26.96360
299 U.S.A. 2005 295583.4 27.82229
[ reached 'max' / getOption("max.print") -- omitted 26 rows ]

xyplot(ci ~ year| country,data=xx.pw,

ylab="investment share",type="b")

2002 2003 2004 2005 2006 2007

China U.S.A. India

30
25
investment share

20
15

Russia Indonesia Brazil

30
25
20
15

2002 2003 2004 2005 2006 2007 2002 2003 2004 2005 2006 2007
year

xyplot(ci ~ year,group=country,data=xx.pw,ylab="investment share",type="b",

auto.key=list(space="right",title="country"))
 © Oliver Kirchkamp
100 Using Graphs and Visualising Data — 7 LATTICE

30
country
investment share

25 China
U.S.A.
India
20 Russia
Indonesia
Brazil
15

2002 2003 2004 2005 2006 2007

year

The legend of the previous plot does not show the lines.

xyplot(ci ~ year,group=country,data=xx.pw,ylab="investment share",type="b",

auto.key=list(space="right",title="country",lines=TRUE))

30
country
investment share

25 China
U.S.A.
India
20 Russia
Indonesia
Brazil
15

2002 2003 2004 2005 2006 2007

year

7.2 Syntax
The data we want to display in our lattice is described with the help of a formula:

Graphs with variables on the vertical and horizontal axis:

• vertical ∼ horizontal creates only one graph

• vertical ∼ horizonal | conditioning variable creates for each level of the conditioning
variable one panel with one graph.
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 101

• vertical ∼ horizontal,group=grouping variable creates only one panel and superim-

poses within this panel graphs for each level of the grouping variable.

• vertical ∼ horizontal | conditioning variable ,group=grouping variable creates for each

level of the conditioning variable one panel. Within these panels graphs for each level
of the grouping variable are superimposed.

Graphs with variables only on the horizontal axis (examples would be density plots,
histograms, etc.):
…when R creates values for vertical axis (e.g. histogram, densityplot, ecdfplot):

• ∼ horizontal creates only one graph

• ∼ horizonal | conditioning variable creates for each level of the conditioning vari-
able one panel with one graph.

• ∼ horizontal,group=grouping variable creates only one panel and superimposes within

this panel graphs for each level of the grouping variable.

• ∼ horizontal | conditioning variable ,group=grouping variable creates for each level

of the conditioning variable one panel. Within these panels graphs for each level of
the grouping variable are superimposed.

Several variables on the horizontal axis

• …∼ h1 + h2 … shows two variables h1 and h2

Types of lines The parameter types determines how points are displayed: type="b" or
type=c("b","smooth","g")
”p” ”l” ”b” ”r” ”g”
horizontal=TRUEhorizontal=FALSE

”s” ”S” ”h” ”a” ”smooth”

”s” ”S” ”h” ”a” ”smooth”

type
 © Oliver Kirchkamp
102 Using Graphs and Visualising Data — 7 LATTICE

7.3 Multiway continued

Instead of having different panels for different countries, we could also have different panels
for different years:
In the next plot we swap variables. We apply factor to year, so that it appears as a text
in the shingles.

xyplot(country ~ ci| factor(year),data=xx.pw,xlab="investment share")

15 20 25 30

2002 2003 2004

Brazil
Indonesia
Russia
India
U.S.A.
China
country

2005 2006 2007

Brazil
Indonesia
Russia
India
U.S.A.
China

15 20 25 30 15 20 25 30
investment share

If the vertical variable (country in this case) is a factor, then dotplot generates even nicer
graphs:

dotplot(country ~ ci| factor(year),data=xx.pw,

xlab="investment share",horizonal=TRUE)

15 20 25 30

2002 2003 2004

Brazil
Indonesia
Russia
India
U.S.A.
China
2005 2006 2007
Brazil
Indonesia
Russia
India
U.S.A.
China

15 20 25 30 15 20 25 30
investment share

We can, of course, show more than one variable on the horizontal axis:
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 103

keys<-list(text=c("private","gov."),space="right",lines=TRUE,size=2,between=.5)
dotplot(country ~ ci+cg| factor(year),data=xx.pw,xlab="investment share",
horizonal=TRUE,auto.key=keys,t="b")

10 15 20 25 30

2002 2003 2004

Brazil
Indonesia
Russia
India
U.S.A.
China private
2005 2006 2007 gov.
Brazil
Indonesia
Russia
India
U.S.A.
China

10 15 20 25 30 10 15 20 25 30
investment share

Certainly, we can also have more than one variable on the vertical axis:
xyplot(ci+cg ~ year| country,layout=c(6,1),data=xx.pw,
ylab="investment share",auto.key=keys,t="b")

2002 2004 2006 2002 2004 2006 2002 2004 2006

China U.S.A. India Russia Indonesia Brazil

30
investment share

25
private
20
gov.

15

10

2002 2004 2006 2002 2004 2006 2002 2004 2006

year

Segment plots Sometimes we have to plot segments. Here we plot a range of the minimum
investment share to the maximum investment share.
library(latticeExtra)
xx2<-as.data.frame(t(sapply(by(xx.pw,list(xx.pw$country),function(x)
c(min=min(x$ci),mean=mean(x$ci),max=max(x$ci))),c)))
xx2
 © Oliver Kirchkamp
104 Using Graphs and Visualising Data — 7 LATTICE

min mean max

China 28.54705 31.91310 33.25413
U.S.A. 24.76686 26.63036 28.11900
India 15.71790 20.26952 25.16494
Russia 17.58459 19.97048 23.60284
Indonesia 12.70857 14.43829 15.58585
Brazil 12.30157 13.19624 14.82930

xx2<-within(xx2,{country<-factor(rownames(xx2))})
segplot(country ~ min+max,centers=mean,draw.bands=FALSE,xlab="investment share",data=xx2)

U.S.A.

Russia

Indonesia

India

China

Brazil

15 20 25 30
investment share

segplot(reorder(factor(country),mean) ~ min+max,centers=mean,
xlab="investment share",draw.bands=FALSE,data=xx2)

China

U.S.A.

India

Russia

Indonesia

Brazil

15 20 25 30
investment share

Segment plots and regression results We can also use segment plots to show regression
results. In the following example we use the pwt6.3 dataset to study the relation between
openc and cgdp per country:
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 105

reg<-with(xx.pw,lm(log(cgdp) ~ openc:country - 1))

reg.ci<-data.frame(cbind(coef(reg),confint(reg)))
names(reg.ci)<-c("coef","lower","upper")
reg.ci[["country"]]<-factor(sub("openc:country","",rownames(reg.ci)))
reg.ci

coef lower upper country

openc:countryChina 0.1371746 0.1250925 0.1492567 China
openc:countryU.S.A. 0.4036432 0.3744708 0.4328155 U.S.A.
openc:countryIndia 0.1931830 0.1746383 0.2117278 India
openc:countryRussia 0.1634039 0.1499233 0.1768846 Russia
openc:countryIndonesia 0.1499098 0.1363253 0.1634943 Indonesia
openc:countryBrazil 0.3370612 0.3086532 0.3654693 Brazil

segplot(reorder(country,coef)~lower+upper,
centers=coef,data=reg.ci,
draw.bands=FALSE,
segments.fun = panel.arrows,
ends = "both",angle = 90,
length = 1, unit = "mm")

U.S.A.

Brazil

India

Russia

Indonesia

China

0.2 0.3 0.4

7.4 Densityplots

data(pwt5.6)
pwt5.6<-within(pwt5.6,continent<-sub(" & ","+",continent))
keys<-list(text=c("private","gov."),space="top",columns=2,lines=TRUE)
densityplot(~i+g | continent,data=pwt5.6,plot.points=FALSE,xlab="investment share",
auto.key=keys)
 © Oliver Kirchkamp
106 Using Graphs and Visualising Data — 7 LATTICE

private gov.
0 20 40 60 80 0 20 40 60 80 0 20 40 60 80

Africa Asia Central+North America Europe Oceania South America

0.10
0.08
Density

0.06
0.04
0.02
0.00

0 20 40 60 80 0 20 40 60 80 0 20 40 60 80
investment share

7.5 Histograms

histogram(~c | continent,data=pwt5.6,plot.points=
FALSE,xlab="consumption share")

0 20 40 60 80 100

Africa Asia Central+North America

30
20
Percent of Total

10
0
Europe Oceania South America

30
20
10
0

0 20 40 60 80 100 0 20 40 60 80 100
consumption share

7.6 Empirical cumulative densities

library(latticeExtra)
ecdfplot(~c | continent,data=pwt5.6,xlab="consumption share")
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 107

20 40 60 80 100

Africa Asia Central+North America

1.0
0.8
0.6
0.4
Empirical CDF

0.2
0.0
Europe Oceania South America
1.0
0.8
0.6
0.4
0.2
0.0

20 40 60 80 100 20 40 60 80 100
consumption share

key<-list(x=0,y=1,corner=c(0,1),background="white",border=TRUE)
ecdfplot(~c ,groups= continent,data=pwt5.6,
auto.key=key,xlab="consumption share")

1.0 Africa
Asia
0.8 Central+North America
Europe
Empirical CDF

Oceania
0.6
South America
0.4

0.2

0.0

20 40 60 80 100
consumption share

7.7 Q-Q plots

qqmath(~c | continent,data=pwt5.6,ylab="consumption share",type="l",

panel = function(x, ...) {
panel.qqmathline(x, ...)
panel.qqmath(x, ...)
})
 © Oliver Kirchkamp
108 Using Graphs and Visualising Data — 7 LATTICE

-2 0 2

Africa Asia Central+North America

100
80
60
consumption share

40
20

Europe Oceania South America

100
80
60
40
20

-2 0 2 -2 0 2
qnorm

qqmath(~c ,groups= continent,aspect="xy",data=pwt5.6,

auto.key=list(space="top",
lines=TRUE,points=FALSE),
ylab="consumption share",type="l")

Africa
Asia
Central+North America
Europe
Oceania
South America

100
consumption share

80

60

40

20

-2 0 2
qnorm

7.8 Sample Q-Q plots

Here we have to factor ed to show the values of ed in the shingles.

library(Ecdat)
data(Wages)
qq(sex ~ lwage | factor(ed),data=subset(Wages,ed>=7),type="l")
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 109

5 6 7 8 5 6 7 8 5 6 7 8

7 8 9 10 11 12
8
7
6
5
male

13 14 15 16 17
8
7
6
5

5 6 7 8 5 6 7 8 5 6 7 8
female

7.9 Boxplots
Here we have to factor ed to make clear whether we want boxplots over ed or over lwage.

bwplot(lwage ~ factor(ed) | sex ,data=subset(Wages,ed>=7))

female male

7
lwage

7 8 9 10 11 12 13 14 15 16 17 7 8 9 10 11 12 13 14 15 16 17

7.10 Barcharts
lattice can also do bar charts:

keys<-list(text=c("private","gov."),space="top",columns=2)
barchart(country ~ ci+cg| as.factor(year),data=xx.pw,xlab="investment share",horizonal=TRUE,
auto.key=keys)
 © Oliver Kirchkamp
110 Using Graphs and Visualising Data — 7 LATTICE

private gov.
10 15 20 25 30

2002 2003 2004

Brazil
Indonesia
Russia
India
U.S.A.
China
2005 2006 2007
Brazil
Indonesia
Russia
India
U.S.A.
China

10 15 20 25 30 10 15 20 25 30
investment share

We should note that often a dotplot or xyplot presents the same data in a better way.

keys<-list(text=c("private","gov."),space="right",lines=TRUE,size=2,between=.5)
dotplot(country ~ ci+cg| factor(year),data=xx.pw,xlab="investment share",
horizonal=TRUE,auto.key=keys,t="b")

10 15 20 25 30

2002 2003 2004

Brazil
Indonesia
Russia
India
U.S.A.
China private
2005 2006 2007 gov.
Brazil
Indonesia
Russia
India
U.S.A.
China

10 15 20 25 30 10 15 20 25 30
investment share

7.11 Coplots

data(warpbreaks) ## given two factors

coplot(breaks ~ 1:length(breaks) | tension*wool, data = warpbreaks,
xlab="index")
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 111

Given : tension

L M H

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

70

Given : wool
40

B
breaks

10
70
40

A
10

0 10 20 30 40 50

index

7.12 Parameters

7.12.1 Types

Usually lattice renders data as points. The argument type=(...) modifies this behaviour.
Some useful values are the following:

• type='p': points

• type='l': lines (in the order of the dataset)

• type='b': lines and points

• type='g': a grid

• type='r': a regression line

• type='smooth': a loess smooth line

data(Caschool,package="Ecdat")
xyplot(testscr ~ str,data=Caschool,type=c("p","g","r","smooth"))
 © Oliver Kirchkamp
112 Using Graphs and Visualising Data — 7 LATTICE

700

680
testscr

660

640

620

14 16 18 20 22 24 26
str

7.12.2 Axes

Different scales for different panels Usually, lattice chooses the same scale for all
panels in a plot. This can be changed with the help of the parameter scales.
Same scale (the default):

xyplot(ci ~ year| as.factor(country),data=xx.pw,ylab="investment share",t="b")

2002 2003 2004 2005 2006 2007

China U.S.A. India

30
25
investment share

20
15

Russia Indonesia Brazil

30
25
20
15

2002 2003 2004 2005 2006 2007 2002 2003 2004 2005 2006 2007
year

Free scale (scales=list(x="same",y="free")):

xyplot(ci ~ year| as.factor(country),data=xx.pw,ylab="investment share",t="b",

scales=list(x="same",y="free"))
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 113

2002 2003 2004 2005 2006 2007

China U.S.A. India

25 26 27 28
29 30 31 32 33

16 18 20 22 24
investment share

Russia Indonesia Brazil

12.5 13.5 14.5

13.0 14.0 15.0
22
20
18

2002 2003 2004 2005 2006 2007 2002 2003 2004 2005 2006 2007
year

Sliced scale (scales=list(x="same",y="sliced"), scales have the same scale, but dif-
ferent origin):

xyplot(ci ~ year| as.factor(country),data=xx.pw,ylab="investment share",t="b",

scales=list(x="same",y="sliced"))

2002 2003 2004 2005 2006 2007

China U.S.A. India

26 28 30 32 34 36

22 24 26 28 30

16 18 20 22 24
investment share

Russia Indonesia Brazil

10 12 14 16 18
10 12 14 16 18
16 18 20 22 24

2002 2003 2004 2005 2006 2007 2002 2003 2004 2005 2006 2007
year

Individual axes
We can influence where an axis is labelled as follows:

myscale<-list(x=list(at=2002:2007,labels=c(2002,"","","","",2007)),
y=list(log=TRUE,at=c(15,20,25,30,35)))
xyplot(ci ~ year| as.factor(country),layout=c(6,1),
scales=myscale,data=xx.pw,ylab="investment share",t="b")
 © Oliver Kirchkamp
114 Using Graphs and Visualising Data — 7 LATTICE

2002 2007 2002 2007 2002 2007

China U.S.A. India Russia Indonesia Brazil

35
30
investment share

25

20

15

2002 2007 2002 2007 2002 2007

year

More complex plots Let us start with some simple data:

xyplot(testscr ~ avginc,data=Caschool)

700

680
testscr

660

640

620

10 20 30 40 50
avginc

xyplot provides a loess smoother, but how can we provide more detail, e.g. confidence
bands for the smoother?
Let us first calculate the necessary data:
data(Caschool,package="Ecdat")
cal.lo<-loess(testscr ~ avginc,data=Caschool)
newx <- with(Caschool,seq(min(avginc),max(avginc),length.out=50))
cal.pred <- predict(cal.lo,newdata=newx,se=TRUE)
cal.df<-with(cal.pred,{data.frame(testscr=fit,
avginc=newx,
upper=fit+qnorm(.975)*se.fit,
lower=fit+qnorm(.025)*se.fit)})
head(cal.df)
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 115

testscr avginc upper lower

1 612.5926 5.335000 621.0852 604.1000
2 621.0255 6.355265 626.8337 615.2173
3 628.4496 7.375531 632.2356 624.6637
4 634.8632 8.395796 637.3966 632.3298
5 640.2646 9.416061 642.3314 638.1978
6 644.6218 10.436326 646.6465 642.5971

xyplot(testscr ~ avginc,data=cal.df,type="l")

700

680
testscr

660

640

620

10 20 30 40 50
avginc

xyplot(testscr ~ avginc,data=cal.df,type="l",
panel=panel.xyplot)

Error in xyplot.formula(panel = function(...) {: formal argument "panel" matched by multiple

actual arguments

xyplot(testscr ~ avginc,data=cal.df,type="l",
panel=function(...) panel.xyplot(...))

Error in xyplot.formula(panel = function(...) {: formal argument "panel" matched by multiple

actual arguments

xyplot(testscr ~ avginc,data=cal.df,type="l",
panel=function(...) {
panel.xyplot(...);
panel.refline(h=660)
})

Error in xyplot.formula(panel = function(...) {: formal argument "panel" matched by multiple

actual arguments
 © Oliver Kirchkamp
116 Using Graphs and Visualising Data — 7 LATTICE

with(cal.df,xyplot(testscr ~ avginc,type="l",
panel=function(...) {
panel.xyplot(...);
panel.xyplot(avginc,upper,type="l",lty=2)
panel.xyplot(avginc,lower,type="l",lty=2)
}))

Error in xyplot.formula(panel = function(...) {: formal argument "panel" matched by multiple

actual arguments

All this could be done with the help of the built in panel.smoother function:
xyplot(testscr ~ avginc,data=Caschool,
panel=function(...) {
panel.smoother(...)
panel.xyplot(...)
})

Error in xyplot.formula(panel = function(...) {: formal argument "panel" matched by multiple

actual arguments

Themes
keys<-list(text=c("consume","private invest.","gov."),lines=TRUE,
space="top",columns=3)
mTheme1<-custom.theme(symbol = brewer.pal(3, "Set1"),
bg = "grey90", fg = "black", pch = 16,lty=1:3,lwd=3)
mTheme2<-custom.theme(symbol = brewer.pal(3, "Pastel1"),
fg = "black", lty=1:3,lwd=3)
mTheme3<-custom.theme(symbol = brewer.pal(3, "Paired"),
fg = "black")
mTheme3$strip.background$col=brewer.pal(3, "Pastel2")
xx<-xyplot(cc+ ci + cg ~ year| as.factor(country),layout=c(6,1),
data=xx.pw,ylab="",t="b",auto.key=keys)

xx
consume private invest. gov.
2002 2004 2006 2002 2004 2006 2002 2004 2006

China U.S.A. India Russia Indonesia Brazil

60

40

20

2002 2004 2006 2002 2004 2006 2002 2004 2006

year
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 117

update(xx,par.settings=mTheme1)

consume private invest. gov.

2002 2004 2006 2002 2004 2006 2002 2004 2006

China U.S.A. India Russia Indonesia Brazil

60

40

20

2002 2004 2006 2002 2004 2006 2002 2004 2006

year

update(xx,par.settings=mTheme2)

consume private invest. gov.

2002 2004 2006 2002 2004 2006 2002 2004 2006

China U.S.A. India Russia Indonesia Brazil

60

40

20

2002 2004 2006 2002 2004 2006 2002 2004 2006

year

update(xx,par.settings=mTheme3)
 © Oliver Kirchkamp
118 Using Graphs and Visualising Data — 7 LATTICE

consume private invest. gov.

2002 2004 2006 2002 2004 2006 2002 2004 2006

China U.S.A. India Russia Indonesia Brazil

60

40

20

2002 2004 2006 2002 2004 2006 2002 2004 2006

year

update(xx,par.settings=standard.theme("pdf", color=FALSE))

consume private invest. gov.

2002 2004 2006 2002 2004 2006 2002 2004 2006

China U.S.A. India Russia Indonesia Brazil

60

40

20

2002 2004 2006 2002 2004 2006 2002 2004 2006

year

Vector graphs versus raster images — don’t rasterize!:

update(xx,par.settings=standard.theme("pdf", color=FALSE))
 [ 4 July 2022 19:50:51]
© Oliver Kirchkamp
— 119

Vector graphs

• tikz (for LATEX)

• eps (sometimes)

• pdf (sometimes)

• svg

• wmf

• ...

Raster graphs

• jpeg

• png

• gif

• tiff

• pdf (sometimes)

• ...