Your first “real” R plot

In this workshop, many of us will create our first real plot. For this purpose we will use the ggplot2 package. This choice of package is based on usage, many people use it and therefore you can easily find help online. ggplot2 is also integrated or highly compatible with other commonly used packages in R.

It is a good idea to write your code in a R script in this session. Be sure to comment your code extensively, this will help you explain to yourself what you are doing and make it easy to reuse parts of your code.

A commented line in R code starts with #:

# This is a comments
a <- c("roses", "are", "red")
# This is another comment

#### Sections can be specified with several number/hash/pound signs ####

# Sections in scripts and code chunks help you structure your work. 
# In R studio, sections can be located from the editor.

It is also a good idea to use the comments to write a statement about the purpose of the script or analysis your are writing. Later we will talk about keeping files in a structured way in projects.

Installing packages

We need to start by installing required packages. From the console we can type

install.packages("tidyverse")

This will install the tidyverse package, a package containing many package. On the tidyverse website you can read:

The tidyverse is an opinionated collection of R packages designed for data science. All packages share an underlying design philosophy, grammar, and data structures.

By using the tidyverse you will adopt a special dialect of R. A dialect that is very efficient and fairly easy to read (as a human).

Tidyverse will install ggplot2 for you. Notice that you only need to install a package once. It is therefore not a good idea to have an unconditional install.packages() in your script.

To get ggplot2 to start working we need to use another command:

library("ggplot2")

Notice that I’ve put ggplot2 inside citation marks ". This is optional!

The library function loads all function contained in the package to your R session. This means that you can access and use them.

For these exercises we will use another package. The exscidata package contains data sets related to exercise physiology. To install it we need a bit more code. Since exscidata is not on CRAN, but on github we can use the remotes package.

# Install package from cran
if(!("remotes" %in% .packages(all.available = TRUE))) {
        install.packages("remotes")
}


# Using remotes, install exscidata from github
if(!("exscidata" %in% .packages(all.available = TRUE))) {
        remotes::install_github("dhammarstrom/exscidata")
}



# Load the exscidata package
library(exscidata)

Note

Above is a if statement. Ordinarily, the statement can be read as: “If the condition is TRUE, then do whatever is in the brackets”. However, we also use a ! around a parentheses containing the %in% operator. The ! negates the test. If the package name is not contained in the vector of all packages created by the .packages(all.available = TRUE), then we want to install the package.

This is a way not having to install packages that are already installed when running your script.

We are now set to load data into our session.

Loading data

The data set we will use in these exercises is called cyclingstudy. You can have a look at the variables by using the help command ?cyclingstudy.

To load data from a package we can use data("cyclingstudy")

Review your understanding

Explain the difference between install.packages() and library()
What is the tidyverse, what will we use ggplot2 for?
What does it mean when a package is not on CRAN?
Identify at least one numeric variable in the help pages for cyclingstudy, identify at least one categorical variable.
What happens in your environment when you type data("cyclingstudy")?

A data set can be accessed in multiple ways. We may want to see the data. We can do this by typing View(cyclingstudy) in the console (notice the capital V). Or we can show a couple of rows and columns in the console by typing cyclingstudy.

Mapping data into `aes()`

The ggplot function (from the ggplot2 package) takes quite a lot of arguments. However, very few are needed to create a graph.

The ggplot2 system uses:

data → The dataset containing variables to plot
aesthetics → Scales where the data are mapped
geometries → Geometric representations of the data
facet → A part of the dataset
statistical transformations → Summaries of data
coordinates → The coordinate space
themes → Plot components not linked to data

We build a graph by mapping variables to different locations and visual characteristics of what is called geoms. This system makes it easy to build different types of graphs using similar syntax.

We will start by mapping to continuous variables to the coordinate system. For this exercise, use the variables weight.T1 and sj.max.

ggplot needs to know were the variables can be found, we therefore have to specify the data argument first. Next we map the variables to the x and y coordinates of the graph. You can copy the code below to your R script.

ggplot(data = cyclingstudy, aes(x = cmj.max, y = sj.max))

Review your understanding

Explain to your friend what we have done so far.
What is mapping in this context?
Define “continuous variables”

Adding geometric representations

We have not yet added graphical representations of the data mapped to coordinates (x and y). These can be added to the plot using the + operator, as we will do below.

Think about a ggplot as a layered construction. Layers can be added (+) to build the graph you want. Layers are added to the graph sequentially, this means it matters in which order you add them.

But what to add?

There are many geoms, for an overview go to Help > Cheat Sheets > Data Visualization with ggplot2

Task 1: Identify a geom suitable for two continuous variables, that will show individual data points on x- and y- coordinates.

Task 2: Call up the help page of the selected geom and find out what you need to add as arguments to the geom

Review your understanding

Explain what the argument inherit.aes = TRUE means.
Explain the sentence (from the help page): “If NULL, the default, the data is inherited from the plot data as specified in the call to ggplot().

By adding, for example, points to the plot, we will be able to see the data. How would you add the geom that creates points to your plot?

Show the code for a possible solution

ggplot(data = cyclingstudy, aes(x = cmj.max, y = sj.max)) + geom_point()

Group work

Task 1: Using the cheat sheet, beside x and y. What other aesthetics (aes) may be added to the plot that will affect the appearance of the points?
Task 2: Using pen and paper, draw a figure of the cyclingstudy data using one categorical variable and one continuous variable and hand the figure to the next group.
Task 3: Code the figure! Create the figure that the other group has drafted for you.

Changing colors and shapes outside mapping

Characteristics such as shapes or colors can also be added to geoms outside the aes(). This means we will override any mapping already given in aes(). As we already have seen, mappings that are inherited from the ggplot function to geoms.

Group work

Task 1: Change characteristics of your plot outside data mapping using color and fill, linetype, size and shape. What geoms are responsive to each change?

Review your understanding

Explain the what the code will produce, without running the code below (google “r shapes” to see what number each shape has.)

# Example 1
ggplot(data = cyclingstudy, aes(x = cmj.max, y = sj.max, color = group)) + geom_point(color = "blue")

# Example 2
ggplot(data = cyclingstudy, aes(x = cmj.max, y = sj.max, color = group)) + geom_point(shape = 8)

# Example 3
ggplot(data = cyclingstudy, aes(x = cmj.max, y = sj.max, color = weight.T1)) + geom_point(shape = 8)

# Example 4
ggplot(data = cyclingstudy, aes(x = cmj.max, y = sj.max, shape = timepoint)) + geom_point(color = "red")

Scales

In ggplot2 there are pre-set palettes for colors, orders of shapes and line types etc. Often you would want to control such settings.

There are many sources for informed selection of colors, one is colorbrewer2. We

To change color scales we use scale_*_* functions that will help you set, e.g., colors manually. In the example below we create a gradient from two colors

ggplot(data = cyclingstudy, 
       aes(x = cmj.max, y = sj.max, color = weight.T1)) + 
        geom_point() +
        scale_colour_gradient(
                low = "#e41a1c",  
                high = "#4daf4a")

Discrete variables can also be set with colors using scale_color_manual

ggplot(data = cyclingstudy, 
       aes(x = cmj.max, y = sj.max, color = timepoint)) + 
        geom_point() +
        scale_color_manual(values = c("#66c2a5","#fc8d62","#8da0cb","#e78ac3"))

Review your understanding

Explain to a friend what scales do

Grouping data

Some aspects of a plot requires that data points are connected together. This essentially means that some a variable needs to group data points. In our example data set, subject gives the identity of each participant. We may use this information to group data points, or connect them with e.g. geom_line(). By adding group = subject to the aes() call in ggplot we will group all geoms that allow grouping.

Review your understanding

Before running the code below. Explain what you expect it will show.

# Example 1
ggplot(data = cyclingstudy, aes(x = cmj.max, 
                                y = sj.max, 
                                color = group,
                                shape = timepoint,
                                group = subject)) + 
        geom_point() +
        geom_line()

Faceting plots

A plot can be quite cluttered, and in this case, give a false impression of a lot of data. The design of this study results in a data set where each participant is tested at multiple time-points. We can therefore create facets based on some aspect of the data, such as time-point.

The facet_wrap() and facet_grid() creates facets.

# Example 1
ggplot(data = cyclingstudy, aes(x = cmj.max, 
                                y = sj.max, 
                                color = group)) + 
        geom_point() +
        facet_wrap(~ timepoint)

As can be seen in the code above, facet_wrap takes a one-handed formula, the ~ (tilde), indicates a formula. We can read this as “wrap by time-point”.

In facet_grid() we will use a two-handed formula, meaning that both sides of the tilde needs information. If we want to group only by rows, we will use a . to indicate that nothing will be used to group by column.

ggplot(data = cyclingstudy, aes(x = cmj.max, 
                                y = sj.max, 
                                color = group)) + 
        geom_point() +
        facet_grid( timepoint ~ .)

In facet_grid above, we could replace the . with another variable. How would you write the code to facet the graph by group in rows and time-point in columns?

Show the code

ggplot(data = cyclingstudy, aes(x = cmj.max, 
                                y = sj.max, 
                                color = group)) + 
        geom_point() +
        facet_grid( group ~ timepoint)

Review your understanding

Explain to a friend, what do the plot produced by the code above show? Include everything that is important to reproduce the plot in your description. Try to describe it without pointing at the plot!

Plot annotations

Annotations can be added to plots, these are often user specified, such as labels and plot titles.

We specify labels with the labs() function and add annotations with the annotate() function.

ggplot(data = cyclingstudy, aes(x = cmj.max, 
                                y = sj.max, 
                                color = group)) + 
        geom_point() +
        labs(x = "Maximal Counter movement jump height (cm)", 
             y = "Maximal squat jump height (cm)", 
             title = "This is the title", 
             subtitle = "This is the subtitle", 
             caption = "This is a caption", 
             color = "This is the group aesthetics") +
        
        annotate(geom = "text", x = 25, y = 35, label = "This is a text annotation")

Notice that labs()makes use of all aesthetic mappings and annotate requires a geom.

Themes

The theme() function is used for non-data layers in the plot.

Combine separate plots

There are two commonly used system for combining individual plots, patchwork, https://patchwork.data-imaginist.com/ and cowplot, https://cran.r-project.org/web/packages/cowplot/vignettes/introduction.html.

The idea here is create figures with multiple individual figures.

patchwork has a very simple syntax.

library(patchwork)


a <- ggplot(data = cyclingstudy, aes(x = cmj.max, 
                                y = sj.max, 
                                color = group)) + 
        geom_point() 

b <- ggplot(data = cyclingstudy, aes(y = sj.max, 
                                x = group)) + 
        geom_boxplot() 


c <- ggplot(data = cyclingstudy, aes(x = timepoint, 
                                y = sj.max, 
                                group = subject)) + 
        geom_line() 



(a | b) / c

cowplot uses plot_grid to arrange plots

library(cowplot)

plot_grid(a, b, c, nrow = 2)

plot_grid can also use a “nested” structure.

plot_grid(plot_grid(a, b, nrow = 1), 
          c, nrow = 2)

In both frameworks, annotations can be added to plots to indicate panels/sub-plots.

Saving output

Output from ggplot2 can be saved from RStudio using the export buttom. However, a more reproducible manner is to save the output using ggsave

Group work

Task 1: Create three separate plots from the cycling data set and save them as objects in your environment.
Task 2: Use cowplot and patchwork to group the plots together.
Task 3: Save the plot using ggsave, explore the help pages to find what arguments are needed!

Exercises

Reproduce figure 1.3 from Spiegelhalter (2019)

# download data
child_heart <- read_csv("https://raw.githubusercontent.com/dspiegel29/ArtofStatistics/master/01-1-2-3-child-heart-survival-times/01-1-child-heart-survival-x.csv")

Reproduce figure 2.2 and 2.3

beans <- read.csv("https://raw.githubusercontent.com/dspiegel29/ArtofStatistics/master/02-2-3-jelly-bean-counts/02-1-bean-data-full-x.csv", header = FALSE)

Your first “real” R plot

Installing packages

Loading data

Mapping data into aes()

Adding geometric representations

Group work

Changing colors and shapes outside mapping

Group work

Scales

Grouping data

Faceting plots

Plot annotations

Themes

Combine separate plots

Saving output

Group work

Exercises

Mapping data into `aes()`