Chapter 6 Graphs

By default, it is possible to make a lot of graphs with R without the need of any external packages. However, in this chapter, we are going to learn how to make graphs using ggplot2 which is a very powerful package that produces amazing graphs. There is an entry cost to ggplot2 as it works in a very different way than what you would expect, especially if you know how to make plots with the basic R functions already. But the resulting graphs are well worth the effort and once you’ll know more about ggplot2 you’ll see that in a lot of situations it is actually faster and easier. Even if you are not interested in drawing plots, I advise you continue reading, as we will dig deeper into functional programming territory and learn how to graph an arbitrary large amount of plots with a few lines of code.

6.1 Resources

Before showing some examples and the general functionality of ggplot2, I list here some online resources that I keep coming back to:

I have a cookbook approach to using ggplot2; I try to find an example online that looks similar to what I have in mind, copy and paste the code and then adapt it to my case. The above resources are the ones I consult the most in these situations (I also go back to past code I’ve written, of course). Don’t hesitate to skim these resources for inspiration and to learn more about some extensions to ggplot2. In the next subsections I am going to show you how to draw the most common plots, as well as show you how to customize your plots with ggthemes.

6.2 Examples

6.2.1 Barplots

To follow the examples below, load the following libraries:

ggplot2 is an implementation of the Grammar of Graphics by Wilkinson (2006), but you don’t need to read the books to start using it. If we go back to the Star Wars data (contained in dplyr), and wish to draw a barplot of the gender, the following lines are enough:

The first argument of the function is the data (called starwars in this example), and then the function aes(). This function is where you list the variables you want to map, and to quote the help file of aes(), describes how the variables are mapped to visual properties (aesthetics) of geoms. You can get different kind of plots by using different geom_ functions.

You can also change the coordinate system in your barplot:

6.2.2 Density

geom_density() is the geom that allows you to get density plots:

## Warning: Removed 6 rows containing non-finite values (stat_density).

Let’s go into more detail now; what if you would like to plot the densities for females and males only (removing the droids from the data first)? This can be done by first filtering the data using dplyr and then separating the dataset by gender:

The above lines do the filtering; only keep gender if gender is in the vector "female", "male". This is much easier than having to write gender == "female" | gender == "male". Then, we pipe this dataset to ggplot:

## Warning: Removed 5 rows containing non-finite values (stat_density).

Let’s take a closer look to the aes() function: I’ve added fill = gender. This means that the there will be one density plot for each gender in the data, and each will be coloured accordingly. This is where ggplot2 might be confusing; there is no need to write explicitly (even if it is possible) that you want the female density to be red and the male density to be blue. You just map the variable gender to this particular aesthetic. You conclude the plot by adding geom_density() which is this case is the plot you want. We will see later how to change the colours of your plot.

6.2.3 Line plots

For the line plots, we are going to use official unemployment data (the same as in the previous chapter, but with all the available years). Get it from here (downloaded from: http://www.statistiques.public.lu/stat/TableViewer/tableView.aspx?ReportId=12950&IF_Language=eng&MainTheme=2&FldrName=3&RFPath=91).

Let’s plot the unemployment for the canton of Luxembourg only:

Because line plots are 2D, you need to specify the y and x axes. There is also another option you need to add, group = 1. This is to tell aes() that the dots have to be connected with a single line. What if you want to plot more than one commune?

This time, I’ve specified group = division which means that there has to be one line per as many communes as in the variable division. I do the same for colours. I think the next example illustrates how ggplot2 is actually brilliant; if you need to add a third commune, there is no need to specify anything else; no need to add anything to the legend, no need to specify a third colour etc:

6.2.4 Facets

In some case you have a factor variable that separates the data you wish to plot into different categories. If you want to have a plot per category you can use the facet_grid() function. Careful though, this function does not take a variable as an argument, but a formula, hence the ~ symbol in the code below:

## Warning: Removed 4 rows containing non-finite values (stat_density).

By changing the formula, you change how the facetting is done:

## Warning: Removed 4 rows containing non-finite values (stat_density).

Recall the categorical variable more_1 that we computed in the previous chapter? Let’s use it as a faceting variable:

## Warning: Removed 4 rows containing non-finite values (stat_density).

6.2.5 Pie Charts

I am not a huge fan of pie charts, but sometimes this is what you have to do. So let’s see how you can create pie charts. First, let’s create a mock dataset with the function tibble::tribble() which allows you to create a dataset line by line:

This data is not in the right format though, which is wide. We need to have it in the long format for it to work with ggplot2. For this, let’s use tidyr::gather() as seen in the previous chapter:

Now, let’s plot this data, first by creating 3 bar plots:

In the code above, I introduce a new option, called stat = "identity". By default, geom_bar() counts the number of observations of each category that is plotted, which is a statistical transformation. By adding stat = "identity", I force the statistical transformation to be the identity function, and thus plot the data as is.

To create the pie chart, first we need to compute the share of each id to var1, var2, etc…

To do this, we first group by id, then compute the total. Then we use a new function ungroup(). After using ungroup() all the computations are done on the whole dataset instead of by group, which is what we need to compute the share:

Let’s take a look to see if this is what we wanted:

## # A tibble: 15 x 5
##    id    variable value total share
##    <chr> <chr>    <dbl> <dbl> <dbl>
##  1 a     var1      26.5  160. 0.165
##  2 b     var1      30    150  0.2  
##  3 c     var1      34    150. 0.226
##  4 a     var2      38    160. 0.237
##  5 b     var2      30    150  0.2  
##  6 c     var2      32    150. 0.213
##  7 a     var3      30    160. 0.187
##  8 b     var3      28    150  0.187
##  9 c     var3      30    150. 0.199
## 10 a     var4      32    160. 0.199
## 11 b     var4      32    150  0.213
## 12 c     var4      28    150. 0.186
## 13 a     var5      34    160. 0.212
## 14 b     var5      30    150  0.2  
## 15 c     var5      26.5  150. 0.176

If you didn’t understand what ungroup() did, rerun the last few lines with it and inspect the output.

To plot the pie chart, we create a barplot again, but specify polar coordinates:

As you can see, this typical pie chart is not very easy to read; compared to the barplots above it is not easy to distinguish a from b from c. It is possible to amend the pie chart a bit to make it clearer, by specifying variable as the x:

But as a general rule, avoid pie charts if possible. Barplots show the same information, much, much clearer.

6.2.6 Adding text to plots

Sometimes you might want to add some text to your plots. This is possible with geom_text():

You can put anything after label =but in general what you’d want are the values, so that’s what I put there. But you can also refine it, imagine the values are actualy, say €:

6.3 Customization

Every plot you’ve seen until now was made with the default look of ggplot2. If you want to change the look, you can apply a theme, and a colour scheme. Let’s take a look at themes first by using the ones found in the package ggthemes. But first, let’s learn how to change the names of the axes and how to title a plot.

6.3.2 Themes

Let’s go back to the unemployment line plots. For now, let’s keep the base ggplot2 theme, but modify it a bit. For example, the legend placement is actually a feature of the theme. This means that if you want to change where the legend is placed you need to modify this feature from the theme. This is done with the function theme():

What I also like to do is remove the title of the legend, because it is often superfluous:

The legend title has to be an element_text object.element_text objects are used with theme to specify how text should be displayed. element_blank() draws nothing and assigns no space (not even blank space).

You could modify every feature of the theme like that, but there are built-in themes that you can use:

For example in the code above, I have used theme_minimal() which I like quite a lot. You can also use themes from the ggthemes package, which even contains a STATA theme, if you like it:

As you can see, theme_stata() has the legend on the bottom by default, because this is how the legend position is defined within the theme. However the legend title is still there. Let’s remove it:

ggthemes even features an Excel 2003 theme (don’t use it though):

6.3.5 Saving plots on disk

There are two ways to save plots on disk; one through the Plots in RStudio and another using the ggsave() function. Using RStudio, navigate to the Plots pane and click on Export. You can then choose where to save the plot and other various options:

This is fine if you only generate one or two plots but if you generate a large number of them, it is less tedious to use the ggsave() function:

There are other options that you can specify such as the width and height, resolution, units, etc…

6.4 Exercises

Exercise 1

This exercise will force you to look for ggplot2 functions that you don’t know yet. But it’s a good exercise, so my advice is to try to do it, even if takes you a long time to find the right functions.

Load the Bwages dataset from the Ecdat package. Your first task is to create a new variable, educ_level, which is a factor variable that equals:

  • “Primary school” if educ == 1
  • “High school” if educ == 2
  • “Some university” if educ == 3
  • “Master’s degree” if educ == 4
  • “Doctoral degree” if educ == 5

Use case_when() for this.

Then, plot a scatter plot of wages on experience, by education level. Add a theme that you like, and remove the title of the legend.

The scatter plot is not very useful, because you cannot make anything out. Instead, use another geom that shows you a non-parametric fit with confidence bands. You do not need to estimate a model or anything like that, you just need to find the right geom.

References

Wilkinson, Leland. 2006. The Grammar of Graphics. Springer Science & Business Media.