Applied Statistics Bo Markussen

Statistical Methods for the Biosciences November, 2021

Graphics with the R-package “ggplot2”

ggplot2 is an R package for making statistical plots. It is created by
Hadley Wickham, and it is based on the Grammar of Graphics (Wilkinson,
2005). A grammar is part of a language, just like Danish and English, with
syntax and semantics. The purpose of a Grammar-of-Graphics is to describe
a statistical plot to the computer. After the computer knows the details of
the wanted plot it can make the plot. But it is also possible to store the
grammatical description of the plot, which then e.g. may be extended later.
Using ggplot2 is not the only way to make graphics in R. The old Base
Graphics system inside R also produce very nice (but not as nice) plots, and
sometimes I prefer to use this. And before ggplot2 became available the
lattice package was very popular (but it is not recommended any more).
For one of my publications (Asare et al., 2017) I wanted to make the following
plot:

The only way I know of making such a plot in a visually appealing layout is
via ggplot2. From a statistical point of view the plot contains two things:

1. Observed data (the points).

2. Predictions from a statistical model (the lines).

1
Applied Statistics Bo Markussen
Statistical Methods for the Biosciences November, 2021

In general I like to include such a figure in my scientific papers. It not only

presents the underlying data to the reader of paper, but it also displays the
applicability of the used statistical model as well as the biological variation.
The figure above has quite a few features:

ˆ The y-axis is transformed. More precisely using the cubic root. Al-
though the cubic root isn’t a standard transformation (like the loga-
rithm), it can still be implemented rather easily (but we will not try
this in this exercise). Note also, that the reader of the figure don’t need
to know the precise transformation in order to perceive the information
contained in the graphics.

ˆ The plot has been split into 3 × 4 = 12 panels arranged in 3 columns

and 2 × 2 = 4 rows. This has been done according to the levels of 1 and
1 + 1 = 2 categorical variables (with 3, 2 and 2 levels, respectively).

ˆ The same x- and y-axes are used in all 12 panels. This makes it possible
to compare data across the panels.

ˆ Within each panel 4 different plotting symbols and 4 different line types
are used to distinguish observed data and model predictions from 4
different regions. These symbols and line types are summarized in the
legend placed to the right of the 12 panels.

ˆ The model predictions (i.e. the lines) are only made in the range of
the observed data. And omitted if there are no observed data in the
corresponding panel.

Now it is time for you to work

Please work on the following questions in small groups of 2 to 4 persons:

1. Discuss the features of the figure on page 1. Do you agree with descrip-
tion I made above?

2. In order to try ggplot2 yourselves open RStudio and install the

ggplot2-package (if you haven’t got the package already).

3. The basic reference on ggplot2 is the book (which should be freely

available to students at KU via the link embedded in this pdf):

Hadley Wickham, “ggplot 2”, Springer, Use R! series, 2009.

2
Applied Statistics Bo Markussen
Statistical Methods for the Biosciences November, 2021

The main data example in Chapter 2 of this book is the diamonds

dataset, which contains information on the price of about 54,000 dia-
monds. We will also be using this dataset for this exercise. Execute the
following R codes in the Console window, and read about the diamonds
dataset:

> library(ggplot2)
> ?diamonds

4. The data frame diamonds is very big with almost 54,000 observations.
Sometimes it can be useful to make a random selection of the obser-
vations in order to avoid having a lot of points plotted on top of each
other. Execute the following R codes, and discuss what they do:

> mypoints <- diamonds[sample(1:nrow(diamonds),1000),]

> mypoints
> head(mypoints)

5. Let’s make our first plots using ggplot2! Try the following R codes
one-by-one (!) and discuss the relations between graphical output and
the R code. Can you see what is plotted?

> ggplot(mypoints,aes(x=carat,y=price)) + geom_point()

> ggplot(mypoints,aes(x=carat,y=price,shape=color)) + geom_point()
> ggplot(mypoints,aes(x=carat,y=price,color=color)) + geom_point()
> ggplot(mypoints,aes(x=carat,y=price,color=color,size=cut)) + geom_point()

6. Basically the ggplot()-function has two arguments:

ˆ The first argument is a data frame. This is somewhat similar to
an Excel sheet, that is, a spreadsheet like organization of data.
In our example this is the data frame that we called mypoints in
item 4 above.
ˆ The second argument is a so-called aesthetic mapping, which tells
R how the variables inside the data frame should be used. This is
defined using another R function, namely aes(x=carat,y=price).
Here we tell R that the carat value should be on the x-axis, and
the price should be on the y-axis. Ok?
The ggplot() only sets up the data/plotting situation. To plot any-
thing you should tell what to plot. In the example above this is done
by adding points. If we instead add lines, then we get something less
useful (do you agree?) in this example:

3
Applied Statistics Bo Markussen
Statistical Methods for the Biosciences November, 2021

> ggplot(mypoints,aes(x=carat,y=price,color=color)) + geom_line()

7. Above we have written (almost) the same ggplot() code 5 times. To

avoid doing this even more times, let’s define it as a variable. Please
try

> myplot <- ggplot(mypoints,aes(x=carat,y=price,color=color))

> myplot + geom_point() + aes(shape=cut)

Note, that after executing the first line the variable “myplot” appears
in the Environment window. This variable now contains the results of
the ggplot-call, and can be used instead of writing this.

8. It’s easy to subdivide the plot into several panels. Let’s try this accord-
ing to the categorical variables cut and clarity. Execute the following
R codes one-by-one and discuss the output:

> myplot + geom_point() + facet_grid(cut~clarity)

> myplot + geom_point() + facet_grid(clarity~cut)

9. To export the most recent figure, such that it can be inserted in a paper
or a report, use the ggsave() function. The following codes saves the
plot to your working directory in png and pdf format1 , respectively:

> ggsave("diamonds.png")
> ggsave("diamonds.pdf")

Try to insert the generated figure in a Word document.

10. As already hinted at several places above the output of a call to ggplot()
is not a graphical output, but a grammatical description of that out-
put. What you see on the screen is a print() of that description. One
implication of this is that you can add more “layers” to the description
before it is printed. The symbol for adding components is “+”, which
in this context shouldn’t be confused with the mathematical operation
of adding numbers. To change the axes to be logarithmic you add this
information. As above; try and think about:

> myplot + geom_point() + scale_x_log10() + scale_y_log10()

1
I recommend png format if you use Word, and pdf if you use PDFLaTeX.

4
Applied Statistics Bo Markussen
Statistical Methods for the Biosciences November, 2021

What happens if you remove either scale_x_log10() or scale_y_log10()?

I guess that you are unhappy with the appearance of the x-axis. There
are too few tick points, right? This can be fixed by hand via

> myplot + geom_point() +

scale_x_log10(breaks=seq(0.5,3,0.5)) + scale_y_log10()

11. You can also add smoothing lines and other statistical output to the
graph:

> myplot + geom_point() + geom_smooth()

Note, that a smoothing line is generated for each of the diamond colors.
The reason for this is that the separation into distinct colors is inherited
from the aes() code inside our variable “myplot”. To make a single
smoothing line for all diamonds we must turn down the inheritance,
and restate the necessary aesthetics. Thus,

> myplot + geom_point() +

geom_smooth(aes(x=carat,y=price),inherit.aes = FALSE)

Note that the smoothing method has change from loess (= local poly-
nomial regression fitting) to gam (= generalized additive model).

ˆ Can you find out why this is the case? Hint: see the help page by
executing ?geom_smooth in the R console.
ˆ Can you change back to loess-smooting? Hint: use the option
method="loess"

12. A linear relation between price and carat appears to be plausible on

the log-log scale:

> myplot + geom_point() + scale_x_log10() + scale_y_log10() +

geom_smooth(aes(x=carat,y=price), inherit.aes = FALSE, method="lm")

Now try to make a figure that contains both a non-linear smoothing

line (either loess or gam) and the linear regression line in the same plot!

13. We have written the two options “aes(x=carat,y=price)” and “inherit.aes = FALSE”
many time above. Discuss whether it would have been more clever to
define “myplot” as

5
Applied Statistics Bo Markussen
Statistical Methods for the Biosciences November, 2021

> myplot <- ggplot(mypoints,aes(x=carat,y=price))

How would this change the solution code for the above questions?

Please note that the lines inserted on the figure on page 1 were not gen-
erated automatically by ggplot2. Instead I used predictions from a linear
mixed effects model fitted via the R-package lme4. In order to do that I
made another data frame with the model predictions, and used this new data
frame together with the geom_line() function. I hope to be able to give an
example of this technique later in the course.
If you want to read more about ggplot2, then you might start at the
homepage:

http://www.r-bloggers.com/basic-introduction-to-ggplot2/

Or perhaps even better read Chapter 3 in the book:

http://r4ds.had.co.nz/

End of exercise