14 Legend: Shape Size and Alpha

In this chapter, we modify legends when shape, size or alpha is mapped to variables.

Why this matters

Shape, size and alpha encode extra variables without new panels; map size and alpha to continuous variables only.

Common pitfall

Mapping size or alpha to a discrete variable produces a misleading legend.

14.1 Change legend point shapes

We will learn to modify the following using scale_shape_manual when shape is mapped to categorical variables:

title
breaks
limits
labels
values

Plot

Let us start with a scatter plot examining the relationship between displacement and miles per gallon from the mtcars data set. We will map the shape of the points to the cyl variable.

ggplot(mtcars) +
  geom_point(aes(disp, mpg, shape = factor(cyl)))

As you can see, the legend acts as a guide for the shape aesthetic. Now, let us learn to modify the different aspects of the legend.

Change the legend title

The title of the legend (factor(cyl)) is not very intuitive. If the user does not know the underlying data, they will not be able to make any sense out of it. Let us change it to Cylinders using the name argument.

ggplot(mtcars) +
  geom_point(aes(disp, mpg, shape = factor(cyl))) +
  scale_shape_manual(name = "Cylinders", values = c(4, 12, 24))

If you have mapped shape/size to a discrete variable which has less than six categories, you can use scale_shape().

ggplot(mtcars) +
  geom_point(aes(disp, mpg, shape = factor(cyl))) +
  scale_shape(name = 'Cylinders')

Change legend shape values

To change the default shapes in the legend, use the values argument and supply a numeric vector of shapes. The number of shapes specified must be equal to the number of levels in the categorical variable mapped. In the below example, cyl has 3 levels (4, 6, 8) and hence we have specified 3 different shapes.

ggplot(mtcars) +
  geom_point(aes(disp, mpg, shape = factor(cyl))) +
  scale_shape_manual(values = c(4, 12, 24))

Change legend labels

The labels in the legend can be modified using the labels argument. Let us change the labels to Four, Six and Eight in the next example. Ensure that the labels are intuitive and easy to interpret for the end user of the plot.

ggplot(mtcars) +
  geom_point(aes(disp, mpg, shape = factor(cyl))) +
  scale_shape_manual(values = c(4, 12, 24), labels = c('Four', 'Six', 'Eight'))

Change legend limits

Let us assume that we want to modify the data to be displayed i.e. instead of examining the relationship between mileage and displacement for all cars, we desire to look at only cars with at least 6 cylinders. One way to approach this would be to filter the data using filter from dplyr and then visualize it. Instead, we will use the limits argument and filter the data for visualization.

ggplot(mtcars) +
  geom_point(aes(disp, mpg, shape = factor(cyl))) +
  scale_shape_manual(values = c(4, 24), limits = c(6, 8))

As you can see above, ggplot2 returns a warning message indicating data related to 4 cylinders has been dropped. If you observe the legend, it now represents only 6 and 8 cylinders.

Reorder or hide legend items with breaks

When there are large number of levels in the mapped variable, you may not want the labels in the legend to represent all of them. In such cases, we can use the breaks argument and specify the labels to be used. In the below case, we use the breaks argument to ensure that the labels in legend represent two levels (4, 8) of the mapped variable.

ggplot(mtcars) +
  geom_point(aes(disp, mpg, shape = factor(cyl))) +
  scale_shape_manual(values = c(4, 12, 24), breaks = c(4, 8))

Putting it all together…

ggplot(mtcars) +
  geom_point(aes(disp, mpg, shape = factor(cyl))) +
  scale_shape_manual(name = "Cylinders", labels = c('Six', 'Eight'),  
     values = c(4, 24), limits = c(6, 8), breaks = c(6, 8))

14.2 Change legend point sizes

We will learn to modify the following using scale_size_continuous when size aesthetic is mapped to variables:

title
breaks
limits
range
labels
values

Plot

Let us start with a scatter plot examining the relationship between displacement and miles per gallon from the mtcars data set. We will map the size of the points to the hp variable. Remember, size must always be mapped to a continuous variable.

ggplot(mtcars) +
  geom_point(aes(disp, mpg, size = hp))

As you can see, the legend acts as a guide for the size aesthetic. Now, let us learn to modify the different aspects of the legend.

Change the legend title

The title of the legend (hp) is not very intuitive. If the user does not know the underlying data, they will not be able to make any sense out of it. Let us change it to Horsepower using the name argument.

ggplot(mtcars) +
  geom_point(aes(disp, mpg, size = hp)) +
  scale_size_continuous(name = "Horsepower")

Change the legend size range

The range of the size of points can be modified using the range argument. We need to specify a lower and upper range using a numeric vector. In the below example, we use range and supply the lower and upper limits as 3 and 6. The size of the points will now lie between 3 and 6 only.

ggplot(mtcars) +
  geom_point(aes(disp, mpg, size = hp)) +
  scale_size_continuous(range = c(3, 6))

Change legend limits

Let us assume that we want to modify the data to be displayed i.e. instead of examining the relationship between mileage and displacement for all cars, we desire to look at only cars whose horsepower is between 100 and 350. One way to approach this would be to filter the data using filter from dplyr and then visualize it. Instead, we will use the limits argument and filter the data for visualization.

ggplot(mtcars) +
  geom_point(aes(disp, mpg, size = hp)) +
  scale_size_continuous(limits = c(100, 350))

Reorder or hide legend items with breaks

When the range of the variable mapped to size is large, you may not want the labels in the legend to represent all of them. In such cases, we can use the breaks argument and specify the labels to be used. In the below case, we use the breaks argument to ensure that the labels in legend represent certain midpoints (125, 200, 275) of the mapped variable.

ggplot(mtcars) +
  geom_point(aes(disp, mpg, size = hp)) +
  scale_size_continuous(breaks = c(125, 200, 275))

Change legend labels

The labels in the legend can be modified using the labels argument. Let us change the labels to “1 Hundred”, “2 Hundred” and “3 Hundred” in the next example. Ensure that the labels are intuitive and easy to interpret for the end user of the plot.

ggplot(mtcars) +
  geom_point(aes(disp, mpg, size = hp)) +
  scale_size_continuous(breaks = c(100, 200, 300),
    labels = c("1 Hundred", "2 Hundred", "3 Hundred"))

Putting it all together…

ggplot(mtcars) +
  geom_point(aes(disp, mpg, size = hp)) +
  scale_size_continuous(name = "Horsepower", range = c(3, 6), 
    limits = c(0, 400), breaks = c(100, 200, 300),
    labels = c("1 Hundred", "2 Hundred", "3 Hundred"))

14.3 Change legend transparency

We will learn to modify the following using scale_alpha_continuous() when alpha or transparency is mapped to variables:

title
breaks
limits
range
labels
values

Plot

Let us start with a scatter plot examining the relationship between displacement and miles per gallon from the mtcars data set. We will map the transparency of the points to the hp variable. Remember, alpha must always be mapped to a continuous variable.

ggplot(mtcars) +
  geom_point(aes(disp, mpg, alpha = hp), color = 'blue')

As you can see, the legend acts as a guide for the alpha aesthetic. Now, let us learn to modify the different aspects of the legend.

Change the legend title

ggplot(mtcars) +
  geom_point(aes(disp, mpg, alpha = hp), color = 'blue') +
  scale_alpha_continuous("Horsepower")

Reorder or hide legend items with breaks

ggplot(mtcars) +
  geom_point(aes(disp, mpg, alpha = hp), color = 'blue') +
  scale_alpha_continuous(breaks = c(125, 200, 275))

Change legend limits

Let us assume that we want to modify the data to be displayed i.e. instead of examining the relationship between mileage and displacement for all cars, we desire to look at only cars whose horsepower is between 100 and 350. One way to approach this would be to filter the data using filter from dplyr and then visualize it. Instead, we will use the limits argument and filter the data for visualization.

ggplot(mtcars) +
  geom_point(aes(disp, mpg, alpha = hp), color = 'blue') +
  scale_alpha_continuous(limits = c(100, 350))

Change the legend transparency range

The range of the transparency of points can be modified using the range argument. We need to specify a lower and upper range using a numeric vector. In the below example, we use range and supply the lower and upper limits as 0.4 and 0.8. The transparency of the points will now lie between 0.4 and 0.8 only.

ggplot(mtcars) +
  geom_point(aes(disp, mpg, alpha = hp), color = 'blue') +
  scale_alpha_continuous(range = c(0.4, 0.8))

Change legend labels

ggplot(mtcars) +
  geom_point(aes(disp, mpg, alpha = hp), color = 'blue') +
  scale_alpha_continuous(breaks = c(100, 200, 300),
    labels = c("1 Hundred", "2 Hundred", 
      "3 Hundred"))

Putting it all together…

ggplot(mtcars) +
  geom_point(aes(disp, mpg, alpha = hp), color = 'blue') +
  scale_alpha_continuous("Horsepower", breaks = c(100, 200, 300),
    limits = c(100, 350), range = c(0.4, 0.8),
    labels = c("1 Hundred", "2 Hundred", "3 Hundred"))

14.4 Exercises

Map shape to cylinders with scale_shape_manual(values = ...).
Control point spread with scale_size_continuous(range = ...).
Explain why size/alpha must map to continuous variables.

Worked solutions in solutions/legend-shape-size-alpha.md (attempt first).

Where to go next

ggplot2 book scales and guides chapters