Visualizing distributions 2

Author

Claus O. Wilke

Introduction

In this worksheet, we will discuss how to display many distributions at once, using boxplots, violin plots, strip charts, sina plots, and ridgeline plots.

First we need to load the required R packages. Please wait a moment until the live R session is fully set up and all packages are loaded.

Next we set up the data.

The dataset we will be working with contains information about the mean temperature for every day of the year 2016 in Lincoln, NE:

Boxplots and violins

We start by drawing the distributions of mean temperatures for each month of the year (columns month and mean_temp in the dataset lincoln_temps), using boxplots. We can do this in ggplot with the geom geom_boxplot(). Try this for yourself.

Hint 
ggplot(lincoln_temps, aes(x = ___, y = ___)) +
  geom_boxplot()
Solution 
ggplot(lincoln_temps, aes(x = month, y = mean_temp)) +
  geom_boxplot()

Next, do the same but now using violins (geom_violin()) instead of boxplots.

Solution 
ggplot(lincoln_temps, aes(x = month, y = mean_temp)) +
  geom_violin()

Customize the violins by trying some of the following:

  • Change the fill or outline color.
  • Swap the x and y mappings.
  • Change the bandwidth (parameter bw) or kernel (parameter kernel). These parameters work just like in geom_density() as discussed in the previous worksheet.
  • Set trim = FALSE. What does this do?

Strip charts and jittering

Both boxplots and violin plots have the disadvantage that they don’t show the individual data points. We can show individual data points by using geom_point(). Such a plot is called a strip chart.

Make a strip chart for the Lincoln temperature data set. Hint: Use size = 0.75 to reduce the size of the individual points.

Hint 
ggplot(lincoln_temps, aes(x = month, y = mean_temp)) +
  geom_point(___)
Solution 
ggplot(lincoln_temps, aes(x = month, y = mean_temp)) +
  geom_point(size = 0.75)

Frequently when we make strip charts we want to apply some jitter to separate points away from each other. We can do so by setting the argument position = position_jitter() in geom_point().

When using position_jitter() we will normally have to specify how much jittering we want in the horizontal and vertical direction, by setting the width and height arguments: position_jitter(width = 0.15, height = 0). Both width and height are specified in units representing the resolution of the data points, and indicate jittering in either direction. So, if data points are 1 unit apart, then width = 0.15 means the jittering covers 0.3 units or 30% of the spacing of the data points.

Try this for yourself, by making a strip chart with jittering.

Hint 
ggplot(lincoln_temps, aes(x = month, y = mean_temp)) +
  geom_point(
    size = 0.75,
    position = position_jitter(___)
  )
Solution 
ggplot(lincoln_temps, aes(x = month, y = mean_temp)) +
  geom_point(
    size = 0.75,
    position = position_jitter(width = 0.15, height = 0)
  )

The function position_jitter() applies random jittering to the data points, which means the plot looks different each time you make it. (Verify this.) We can force a specific, fixed arrangement of jittering by setting the seed parameter. This parameter takes an arbitrary integer value, e.g. seed = 1234. Try this out.

Hint 
ggplot(lincoln_temps, aes(x = month, y = mean_temp)) +
  geom_point(
    size = 0.75,
    position = position_jitter(width = 0.15, height = 0, seed = ___)
  )
Solution 
ggplot(lincoln_temps, aes(x = month, y = mean_temp)) +
  geom_point(
    size = 0.75,
    position = position_jitter(width = 0.15, height = 0, seed = 1234)
  )

Finally, try to figure out what the parameter height does, by setting it to a value other than 0, or by removing it entirely.

Sina plots

We can create a combination of strip charts and violin plots by making sina plots, which jitter points into the shape of a violin. We can do this with geom_sina() from the ggforce package. Try this out.

Solution 
ggplot(lincoln_temps, aes(x = month, y = mean_temp)) +
  geom_sina(size = 0.75)

It often makes sense to draw a sina plot on top of a violin plot. Try this out.

Solution 
ggplot(lincoln_temps, aes(x = month, y = mean_temp)) +
  geom_violin() +
  geom_sina(size = 0.75)

Finally, customize the violins by removing the outline and changing the fill color.

Solution 
ggplot(lincoln_temps, aes(x = month, y = mean_temp)) +
  geom_violin(color = NA, fill = "cornsilk") +  # `NA` means no color
  geom_sina(size = 0.75)

Ridgeline plots

As the last alternative for visualizing multiple distributions at once, we will make ridgeline plots. These are multiple density plots staggered vertically. In ridgeline plots, we normally map the grouping variable (e.g. here, the month) to the y axis and the dependent variable (e.g. here, the mean temperature) to the x axis.

We can create ridgeline plots using geom_density_ridges() from the ggridges package. Try this out. Use the column month_long instead of month for the name of the month to get a slightly nicer plot. Hint: If you get an error about a missing y aesthetic you need to swap your x and y mappings.

Solution 
ggplot(lincoln_temps, aes(x = mean_temp, y = month_long)) +
  geom_density_ridges()

What happens when you use month instead of month_long? Can you explain why?

It is often a good idea to prune the ridgelines once they are close to zero. You can do this with the parameter rel_min_height, which takes a numeric value relative to the maximum height of any ridgeline anywhere in the plot. So, rel_min_height = 0.01 would prune all lines that are less than 1% of the maximum height in the plot.

Hint 
ggplot(lincoln_temps, aes(x = mean_temp, y = month_long)) +
  geom_density_ridges(rel_min_height = ___)
Solution 
ggplot(lincoln_temps, aes(x = mean_temp, y = month_long)) +
  geom_density_ridges(rel_min_height = 0.01)