library(palmerpenguins)
data("penguins")ggrain::geom_rain()
BSTA 526 Functions of the Week
Create rain plots with ggplot
1 ggrain::geom_rain()
geom_rain() is from the ggrain package. This function allows for the creation of so-called “rain” plots with ggplot.
Additional information can be found on the CRAN website, or by typing ?geom_rain in the console.
2 What is it for?
In conjunction with ggplot, geom_rain() will take numeric (y-axis) and categorical (x-axis) data to generate a rain plot, which is a compound plot with density (i.e., half-violin), dot plot, and boxplot representations of the data. These plots get their name from the fact that they look a little like raindrops coming out of a cloud. (Amusingly, the boxplot element is sometimes referred to as the “train in the rain.”)
3 Examples
For the following examples, we’ll primarily be using the palmerpenguins package. The data in penguins represent various measurements of different penguin species recorded near Palmer Station in Antarctica.
(Note: In an effort to minimize code and simplify the examples, plot titles are omitted and variable names are left mostly unchanged.)
3.1 Simple Rain Plots
Through simple implementations of geom_rain(), we can get a sense of the default settings before exploring the various other argument options for plot customization. Plots can be created using a single numeric variable as an input, or by combining a numeric variable with a categorical variable.
3.1.1 One numeric variable
Let’s begin by simply plotting the variable bill_length_mm, which represents penguin bill length (mm):
# Pipe the data
penguins |>
# Create a ggplot
ggplot(aes(
# Select our variable(s)
x = 1,
y = bill_length_mm
)) +
# Add the rain plot layer
geom_rain() +
# Add a theme
theme_classic(base_size = 15) +
# Remove the x-axis labels
theme(axis.title.x = element_blank(),
axis.ticks.x = element_blank(),
axis.text.x = element_blank())
Notice that we needed to provide information about the x-axis (i.e., x = 1), which meant that we also had to use theme() to remove the extraneous information from x-axis.
3.1.2 Rotated plots
By design, geom_rain() only accepts the measurement variable along the y-axis. However, if we want to rotate the plot 90-degrees, we can do so using coord_flip():
penguins |>
ggplot(aes(
x = 1,
y = bill_length_mm
)) +
geom_rain() +
theme_classic(base_size = 15) +
theme(axis.title.y = element_blank(),
axis.ticks.y = element_blank(),
axis.text.y = element_blank()) +
# Swap the x- and y-axes
coord_flip()
3.1.3 Combining numeric and categorical data
If we wish to separate our measurement according to a particular categorical variable, we can do so easily by adjusting aes() in ggplot, much like for any other geom layer. Continuing to use bill_length_mm as our numeric measurement, let’s now add the categorical variable island to our graphic, which represents penguin island of origin:
penguins |>
ggplot(aes(
x = 1,
y = bill_length_mm,
# Fill in color by `island` and increase transparency
fill = island,
alpha = 0.5)) +
geom_rain() +
theme_classic(base_size = 15) +
theme(axis.title.x = element_blank(),
axis.ticks.x = element_blank(),
axis.text.x = element_blank()) +
# Remove the extra legend element for `alpha`
guides(alpha = "none")
(Notice that the dots do not automatically inherit the colors from fill.)
It can be difficult to discern characteristics of the distribution when all groups are aligned in a single column. To clean things up a bit, we can assign our categorical variable to the x-axis:
penguins |>
ggplot(aes(
# Assign `island` to the x-axis
x = island,
y = bill_length_mm,
fill = island,
alpha = 0.5)) +
geom_rain() +
theme_classic(base_size = 15) +
guides(alpha = "none")
3.2 Exploring Additional Arguments
Now that we have a basic understanding of geom_rain(), we can touch on a few other useful arguments for customizing our rain plots.
3.2.1 Adding a covariate
If we’re interested in exploring a third variable as a covariate, we can use the cov argument in geom_rain() to add color shading to the dot plot. We can specify a numeric covariate, such as flipper_length_mm:
penguins |>
ggplot(aes(
x = island,
y = bill_length_mm,
fill = island,
alpha = 0.5)) +
geom_rain(
# Add `flipper_length_mm` as a covariate
cov = "flipper_length_mm"
) +
theme_classic(base_size = 15) +
guides(alpha = "none")
We can also specify a categorical covariate, such as species:
penguins |>
ggplot(aes(
x = island,
y = bill_length_mm,
fill = island,
alpha = 0.5)) +
geom_rain(
# Add `species` as a covariate
cov = "species"
) +
theme_classic(base_size = 15) +
guides(alpha = "none")
3.2.2 Plot orientation
The orientation of rain plots can be mirrored along the y-axis in a number of ways using the argument rain.side. Accepted values tell the plot to orient “right,” “left,” or “flanking,” using shorthand like r, l, f1x1, etc. To demonstrate, let’s select sex as our categorical variable, since it only has two levels. We can adjust the two resulting rain plots to face away/toward each other:
penguins |>
# Select `sex` and `bill_length_mm`
select(sex, bill_length_mm) |>
# Omit missing values
na.omit() |>
ggplot(aes(
x = sex,
y = bill_length_mm,
fill = sex,
alpha = 0.5)) +
geom_rain(
# Tell the plot to adopt a "flanking" orientation
rain.side = "f1x1"
) +
theme_classic(base_size = 15) +
guides(alpha = "none")
3.2.3 Modifying isolated plot elements
On a more granular level, aspects of the density, boxplot, and dot plot elements can be adjusted using violin.args, violin.args.pos, boxplot.args, boxplot.args.pos, point.args, and point.args.pos. To illustrate a few examples, let’s plot body_mass_g and species:
penguins |>
ggplot(aes(
x = species,
y = body_mass_g,
fill = species)) +
geom_rain(
# Label boxplot outliers as asterisks (*) and make them red
boxplot.args = list(
outlier.shape = 8,
outlier.color = "red"),
# Change the transparency of the density plots
violin.args = list(alpha = 0.1),
# Remove jitter and nudge the position of the dot plots
point.args.pos = list(
position = position_nudge(
x = -0.05))
) +
theme_classic(base_size = 15)
Note that we need to use list() to input adjustments to these various arguments. Unfortunately, using list() here can reset certain defaults (e.g., removing jitter of the dots), so some settings may need to be reset!
3.2.4 Paired Data
As a final example, we can draw lines between categories to represent paired/longitudinal data. Unfortunately, penguins will not be a suitable dataset for this example, since we need long-format data with multiple observations for each individual. Instead, we can load the sleep dataset, which shows the effects of two soporific drugs on changes in sleep (extra) at two time points (group) for 10 patients (ID). The data are included as a pre-loaded dataset in R, so there is no need to download an additional package.
To show paired/longitudinal data, use the id.long.var argument:
# Load `sleep` dataset
data("sleep")
sleep |>
ggplot(aes(
x = group,
y = extra,
fill = group,
alpha = 0.5)) +
geom_rain(
rain.side = "f1x1",
# Specify `ID` as the pairing variable
id.long.var = "ID"
) +
theme_classic(base_size = 15) +
# Fix confusing labels on the axes
labs(
x = "Time",
y = "Change in sleep duration (hrs)"
) +
# Remove the legend
theme(legend.position = "none") +
guides(alpha = "none")
As a final note, the lines in this plot can be adjusted using the line.args and line.args.pos arguments.
4 Is it helpful?
4.1 Pros
- Rain plots immediately provide information about sample size, shape, center, and spread of distributions without having to create or query multiple other metrics or plots.
- Argument options offer a fair range of customizability for different plotting scenarios.
4.2 Cons
- Rain plots have a tendency to look cluttered, since they present a lot of (sometimes redundant) information in a very compact space.
- Modifying specific plot elements can be somewhat clunky and unintuitive.
- There is ostensibly no method of re-ordering the dot plot, boxplot, and density graphics (e.g., having the dot plot appear in the middle, between the boxplot and density plot).
4.3 Verdict: Is it helpful?
In short, it depends… These plots can be very informative and aesthetically-pleasing representations of data if we’re working with two or maybe three groups along a common numeric variable. In particular, rain plots seem to be useful for data exploration, since they give us a lot of information very quickly. However, as mentioned above, some of this information (e.g., center; spread) is redundant between plot elements, so it’s not necessarily a concise illustration of the data. In fact, there can sometimes be too much information in these plots.
Ultimately, geom_rain() seems like a good tool have on-hand for data exploration, but I personally don’t plan to use it regularly for finalized reports.