violin plot with constant data? - r

Question

I have some weird behaviour of violin plots, when the data is (in parts) constant.
If I check for constant data and add some small errors artificially (e.g. by adding runif( N, min = -0.001, max = 0.001 ), the script will run. However, that distorts the other violin plot(s) to vertical line(s) (see 1), while it should look something like 2
Question:
Is it possible (when the partial data for a violin plot is constant) to
display a simple horizontal line for the respective constant data
display the other violin plots, as if the constant data wasn't present?
R code:
library(ggplot2)
library(grid)
library(gridExtra)
N <- 20
test_data <- data.frame(
idx <- c( 1:N, 1:N ),
vals <- c( runif(N, 0, 1),
rep( 0.5, N)), # <- R script won't run
#rep( 0.5, N) + runif( N, min = -0.001, max = 0.001 )), # <- delivers graphic (distorted)
type <- c( rep("range", N),
rep("const", N))
)
grid.arrange(
ggplot( test_data, aes( x = idx, y = vals)) +
geom_line( aes(colour = type)),
ggplot( test_data, aes( x = type, y = vals)) +
geom_violin( aes( fill = type),
position = position_dodge(width = 1))
)

Answer 1

I finally managed to get a violin plot with some group(s) having zero variance (standard deviation)
to display a flat line for 0-variance groups
display normal violin plots for other groups
In my example I have 3 groups of data - two without zero variance and the third is constant.
While accumulating the groups, I calculate the standard deviation (variance would be same functionality)
library(ggplot2)
library(gridExtra)
N <- 20
test_data <- data.frame()
# random data from range
for( grp_id in 1:2)
{
group_data <- data.frame(
idx = 1:N,
vals = runif(N, grp_id, grp_id + 1),
type = paste("range", grp_id)
)
group_data$sd_group <- sd( group_data$vals)
test_data = rbind( test_data, group_data)
}
# constant data
group_data = data.frame(
idx = 1:N,
vals = rep( 0.5, N),
type = "const"
)
group_data$sd_group <- sd( group_data$vals)
as suggested I add a little offset to obtain a violin plot for group 'const'
# add a little jittering to get the flat line
if( 0 == group_data$sd_group[1])
{
group_data$vals[1] = group_data$vals[1] + 0.00001
}
test_data = rbind( test_data, group_data)
Only thing now left to do is to scale all violin plots to the same width
grid.arrange(
ggplot( test_data, aes( x = idx)) +
geom_line( aes( y = vals, colour = type)),
ggplot( test_data, aes( x = type, y = vals, fill = type)) +
geom_violin( scale = "width"),
ncol = 1
)