Inconsistent behavior in violinplot when near-zero variance vs zero variance

[stefaneng]

Description

Converting this internal discussion into an issue for better visability: https://github.com/uclahs-cds/project-cohort-PrecisionHealth-WXPH-000086-150KPH/pull/12

Lattice issue here: deepayan/lattice#27

Violinplots appear different with near-zero variance vs zero variance.
The reason for this is that in lattice the limits are pulled from range(density(data)$x) which returns a large interval when there is no variance

> range(density(simple.data$values)$x)
[1] -95.69051 295.69051
> range(density(simple.data$values2)$x)
[1]  99.99981 100.00048

Originally posted in deepayan/lattice#27 (comment)

Example

library(BoutrosLab.plotting.general)

set.seed(13)
simple.data <- data.frame(
    values = rep(100, 5),
    values2 = rep(100, 5) + rnorm(5, mean = 0.0001, sd = 0.0001),
    variable = rep('a', 5)
    )

print(sd(simple.data$values))
#> [1] 0

print(sd(simple.data$values2))
#> [1] 8.052756e-05

create.violinplot(
    formula = values ~ variable,
    data = simple.data,
    ylim = c(-20, 200),
    main = 'Zero variance'
    )

create.violinplot(
    formula = values2 ~ variable,
    data = simple.data,
    ylim = c(-20, 200),
    main = 'Near zero variance'
    )

The most dangerous case is when we have multiple categories and we plot without knowing that we have zero variance in one of the categories.

library(BoutrosLab.plotting.general)

set.seed(13)
simple.data <- data.frame(
    values = c(rep(5, 5), rnorm(25, sd = 3)),
    variable = c(rep('a', 5), rep('b', 25))
    )

create.violinplot(
    formula = values ~ variable,
    data = simple.data,
    ylim = c(-10, 15)
    )

Just looking at the plot it is not clear that a has zero variance which could lead to incorrect interpretation if published. It would be more clear that it has zero variance if a single line is draw representing a point mass with probability one.

^{Created on 2023-02-15 with reprex v2.0.2}