Introduction to [R] ggplot2

R's ggplot2 module, written by Hadley Wickham is a graphics system that is based on the Grammar of Graphics.

Plotting your data with ggplot2 can be as simple as a plot(x,y) command, or as complex as several lines of R commands. However, the result, even in the most basic theme is publication-ready.

I will cover a short introduction to the module, split into two parts. The first part will cover the qplot function, which is the quick-and-dirty way of plotting data with this module. The second part will cover the more low-level plotting configuration and I will try to glance over the various aspect of it.

One important thing to remember is that ggplot2 operates on data frames. Therefore for the purpose of this notebook I will keep using the mpg dataset that is included in the module. Let's load it and see what it holds..

In [1]:
## if you don't have ggplot2 install execute: install.packages("ggplot2")

In [2]:

In [3]:

'data.frame':	234 obs. of  11 variables:
 $ manufacturer: Factor w/ 15 levels "audi","chevrolet",..: 1 1 1 1 1 1 1 1 1 1 ...
 $ model       : Factor w/ 38 levels "4runner 4wd",..: 2 2 2 2 2 2 2 3 3 3 ...
 $ displ       : num  1.8 1.8 2 2 2.8 2.8 3.1 1.8 1.8 2 ...
 $ year        : int  1999 1999 2008 2008 1999 1999 2008 1999 1999 2008 ...
 $ cyl         : int  4 4 4 4 6 6 6 4 4 4 ...
 $ trans       : Factor w/ 10 levels "auto(av)","auto(l3)",..: 4 9 10 1 4 9 1 9 4 10 ...
 $ drv         : Factor w/ 3 levels "4","f","r": 2 2 2 2 2 2 2 1 1 1 ...
 $ cty         : int  18 21 20 21 16 18 18 18 16 20 ...
 $ hwy         : int  29 29 31 30 26 26 27 26 25 28 ...
 $ fl          : Factor w/ 5 levels "c","d","e","p",..: 4 4 4 4 4 4 4 4 4 4 ...
 $ class       : Factor w/ 7 levels "2seater","compact",..: 2 2 2 2 2 2 2 2 2 2 ...

The qplot() function

The qplot() is the most basic function of the module. It works like the plot() function of the base graphics system.

The plots in ggplot2 are made of objects that belong in two categories:

  • aesthetics $\rightarrow$ size, shape, color
  • geoms $\rightarrow$ points, lines, bar-plots

As already mentioned, ggplot2 works on data frames. In fact the qplot() function acts as a "graphical wrapper" around the data frame, extracting all the necessary information for the needs of plotting the dataset. This means that it has access to all numerical and categorical information within the dataset, as well as their labels. In fact, it will use the labels of the columns as axes titles in the final plot. Therefore, one important thing to keep in mind is always have your factor variables labeled.

Let's make our first plot! Using the miles to the galon variable for the consumption of the car in the highway

In [4]:
qplot(hwy, data=mpg)

stat_bin: binwidth defaulted to range/30. Use 'binwidth = x' to adjust this.

Since only one variable was given to plot, the function chose to illustrate a histogram to best describe the input data.

In the function call: qplot(hwy, data=mpg)

the first argument is the name of the variable to be plotted in the dataset, while the second keyword argument is the name of the dataset from which to load the information.

Notice how the name of the variable was used to label the x-axis, while the fact that the histogram includes counts per bins is shown as the y-axis label. In addition, the width of the histogram is chosen automatically to describe all data points, while the default binning is $\frac{\mathrm{range}}{30}$.

Since the function identified that this is one variable and chose to plot it with a histogram, additional optimisation of the plot can be achieved with including additional arguments. For example, changing the default width of the bins:

In [5]:
qplot(hwy, data=mpg, binwidth=4)

To create a two dimensional scatter plot for two variables of the dataframe one can simply do

In [6]:
qplot(displ, hwy, data=mpg)

The order of the arguments correspond to x-axis, y-axis and dataset.

One additional level of information can be achieved by modifying the aesthetics. For example, one could add the information of the factor variable drv that shows if the vehicle is forward, rear of 4-wheel drive. To show such information on the same plot, one could use different colors to specify this categorical feature.

So projecting 3D information on a 2D plot by simply adding one more argument...

In [7]:
qplot(displ, hwy, data=mpg, color=drv)

Notice how the plot points are colorised to indicate different levels of the drv variable. Moreover, the legend for the coloring has been added automatically.

Assume now that we also add statistics to the plot. For example let's try to describe our data using a smooth curve to see the overall trends.

This can be done by adding geoms. In fact, in the previous example we had already invoked the call of a geom; the data points. The qplot() function and ggplot2 in general when seeded only with the information of the variable and the dataset to which it belongs, they have no idea what to do with it. Thus, at that level, no plot has been created. What qplot() does, is to automatically invoke -for example- the geom("points") to plot the scatter plot of the two variables.

Let's add a geom, including a smooth line:

In [8]:
qplot(displ, hwy, data=mpg, geom="smooth")

geom_smooth: method="auto" and size of largest group is <1000, so using loess. Use 'method = x' to change the smoothing method.

This plots a smooth line that corresponds to the dataset, while the gray area defines the $95\%$ confidence interval around it. Notice that the points have dissapeared. This is due to the fact that by specifying the geom that the plot should use, the default configuration is overwritten. If I'd want to also include the points then I have to specify it explicitly.

In [9]:
qplot(displ, hwy, data=mpg, geom=c("smooth","point"))

geom_smooth: method="auto" and size of largest group is <1000, so using loess. Use 'method = x' to change the smoothing method.

Finally, notice that I have removed the condition color=drv. This is due to the fact that when such a categorisation is present, the geoms will be drawn for each category separately. For example..

In [10]:
qplot(displ, hwy, data=mpg, color=drv ,geom=c("smooth","point"))

geom_smooth: method="auto" and size of largest group is <1000, so using loess. Use 'method = x' to change the smoothing method.

One additional way to visualise the separation of the data would be not to separate them by color but rather separate them by shape.

In [11]:
qplot(displ, hwy, data=mpg, shape=drv)

Where we used circles, triangles and squares as markers for the various categories.

Finally, one could use differnet statistical methods for the smoothing. For example if instead of the confidence level one wanted to use a linear model to identify the linear relation, one could specify

In [12]:
qplot(displ, hwy, data=mpg, color=drv, geom=c("smooth","point"), method=lm)


Another interesting feature of ggplot2 is the introduction of facets. These are panel-like plots that separate the distribution of variables based on the level of one categorical variable. This is similar to what color and fill arguments had as a result in the plots we've seen up to now, with the difference that the different categories are not overlayed, but rather split into different plots.

Facets can be used as an argument in the qplot() function. They have a very distinct syntax:

  • The facet can be split by rows or by columns or both;
  • One variable is used to split the panel in rows and one to split it in columns;
  • In the syntax of facets the two variables are defined in the arguments list and are separated by a ~ (tilde) symbol.
  • If only one variable is to be used, then in the place of the non-used variable a . is placed, to keep the syntax rows ~ columns.

For example, to split the plots by the drv variable and organise them in one row with three columns:

In [13]:
qplot(displ, hwy, data=mpg, facets = .~drv)

While to do the same thing but split them in rows in one column:

In [14]:
qplot(displ, hwy, data=mpg, facets = drv~.)

Again, the same logic applies on geoms.

In [15]:
qplot(displ, hwy, data=mpg, facets = drv~., geom=c("point","smooth"))

geom_smooth: method="auto" and size of largest group is <1000, so using loess. Use 'method = x' to change the smoothing method.
geom_smooth: method="auto" and size of largest group is <1000, so using loess. Use 'method = x' to change the smoothing method.
geom_smooth: method="auto" and size of largest group is <1000, so using loess. Use 'method = x' to change the smoothing method.