Split Violin Plots

Tom Kelly

2021-07-27

##Violin Plots

Therefore violin plots are a powerful tool to assist researchers to visualise data, particularly in the quality checking and exploratory parts of an analysis. Violin plots have many benefits:

As shown below for the iris dataset, violin plots show distribution information that the boxplot is unable to.

###General Set up

library("vioplot")

We set up the data with two categories (Sepal Width) as follows:

data(iris)
summary(iris$Sepal.Width)
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##   2.000   2.800   3.000   3.057   3.300   4.400
table(iris$Sepal.Width > mean(iris$Sepal.Width))
## 
## FALSE  TRUE 
##    83    67
iris_large <- iris[iris$Sepal.Width > mean(iris$Sepal.Width), ]
iris_small <- iris[iris$Sepal.Width <= mean(iris$Sepal.Width), ]

###Boxplots

First we plot Sepal Length on its own:

An indirect comparison can be achieved with par:

Violin Plots

First we plot Sepal Length on its own:

An indirect comparison can be achieved with par:

Split Violin Plots

A more direct comparision can be made with the side argument and add = TRUE on the second plot:

vioplot(Sepal.Length~Species, data=iris_large, col = "palevioletred", plotCentre = "line", side = "right")
vioplot(Sepal.Length~Species, data=iris_small, col = "lightblue", plotCentre = "line", side = "left", add = T)
title(xlab = "Species", ylab = "Sepal Length")
legend("topleft", fill = c("lightblue", "palevioletred"), legend = c("small", "large"), title = "Sepal Width")

Custom axes labels

Custom axes labels are supported for split violin plots. However, you must use these arguments on the first call of vioplot.

vioplot(Sepal.Length~Species, data=iris_large, col = "palevioletred", plotCentre = "line", side = "right", xlab = "Iris species", ylab = "Length", main = "Sepals", names=paste("Iris", levels(iris$Species)))
vioplot(Sepal.Length~Species, data=iris_small, col = "lightblue", plotCentre = "line", side = "left", add = T)
legend("topleft", fill = c("lightblue", "palevioletred"), legend = c("small", "large"), title = "Width")

Note that this is disabled for the second vioplot call to avoid overlaying labels.

vioplot(Sepal.Length~Species, data=iris_large, col = "palevioletred", plotCentre = "line", side = "right")
vioplot(Sepal.Length~Species, data=iris_small, col = "lightblue", plotCentre = "line", side = "left", add = T, xlab = "Iris species", ylab = "Length", main = "Sepals", names=paste("Iris", levels(iris$Species)))
## Warning in vioplot.formula(Sepal.Length ~ Species, data = iris_small, col = "lightblue", : Warning: names can only be changed on first call of vioplot (when add = FALSE)
## Warning in vioplot.formula(Sepal.Length ~ Species, data = iris_small, col = "lightblue", : Warning: x-axis labels can only be changed on first call of vioplot (when add = FALSE)
## Warning in vioplot.formula(Sepal.Length ~ Species, data = iris_small, col = "lightblue", : Warning: y-axis labels can only be changed on first call of vioplot (when add = FALSE)
## Warning in vioplot.default(x, ...): Warning: names can only be changed on first call of vioplot (when add = FALSE)
## Warning in vioplot.default(x, ...): Warning: main title can only be changed on first call of vioplot (when add = FALSE)
legend("topleft", fill = c("lightblue", "palevioletred"), legend = c("small", "large"), title = "Width")

Median

The line median option is more suitable for side by side comparisions but the point option is still available also:

vioplot(Sepal.Length~Species, data=iris_large, col = "palevioletred", plotCentre = "point", side = "right", pchMed = 21, colMed = "palevioletred4", colMed2 = "palevioletred2")
vioplot(Sepal.Length~Species, data=iris_small, col = "lightblue", plotCentre = "point", side = "left", pchMed = 21, colMed = "lightblue4", colMed2 = "lightblue2", add = T)
title(xlab = "Species", ylab = "Sepal Length")
legend("topleft", fill = c("lightblue", "palevioletred"), legend = c("small", "large"), title = "Sepal Width")

It may be necessary to include a points command to fix the median being overwritten by the following plots:

vioplot(Sepal.Length~Species, data=iris_large, col = "palevioletred", plotCentre = "point", side = "right", pchMed = 21, colMed = "palevioletred4", colMed2 = "palevioletred2")
vioplot(Sepal.Length~Species, data=iris_small, col = "lightblue", plotCentre = "point", side = "left", pchMed = 21, colMed = "lightblue4", colMed2 = "lightblue2", add = T)
points(1:length(levels(iris$Species)), as.numeric(sapply(levels(iris$Species), function(species) median(iris_large[grep(species, iris_large$Species),]$Sepal.Length))), pch = 21, col = "palevioletred4", bg = "palevioletred2")
title(xlab = "Species", ylab = "Sepal Length")
legend("topleft", fill = c("lightblue", "palevioletred"), legend = c("small", "large"), title = "Sepal Width")

Similarly points could be added where a line has been used previously:

vioplot(Sepal.Length~Species, data=iris_large, col = "palevioletred", plotCentre = "line", side = "right", pchMed = 21, colMed = "palevioletred4", colMed2 = "palevioletred2")
vioplot(Sepal.Length~Species, data=iris_small, col = "lightblue", plotCentre = "line", side = "left", pchMed = 21, colMed = "lightblue4", colMed2 = "lightblue2", add = T)
points(1:length(levels(iris$Species)), as.numeric(sapply(levels(iris$Species), function(species) median(iris_large[grep(species, iris_large$Species),]$Sepal.Length))), pch = 21, col = "palevioletred4", bg = "palevioletred2")
points(1:length(levels(iris$Species)), as.numeric(sapply(levels(iris$Species), function(species) median(iris_small[grep(species, iris_small$Species),]$Sepal.Length))), pch = 21, col = "lightblue4", bg = "lightblue2")
title(xlab = "Species", ylab = "Sepal Length")
legend("topleft", fill = c("lightblue", "palevioletred"), legend = c("small", "large"), title = "Sepal Width")

Here it is aesthetically pleasing and intuitive to interpret categorical differences in mean and variation in a continuous variable.

Sources

These extensions to vioplot here are based on those provided here:

These have previously been discussed on the following sites: