Boxplot is too narrow when used with facetting

[Nekochef]

Boxplot is too narrow when used with faceting, and this issue never appears in 2.2.1 release.
I encountered this issue with a dataset similar to below:

va <- c(rnorm(10, 1, 1))
vb <- c(rnorm(10, 3, 1))
vc <- c(rnorm(10, 2, 1))
vd <- c(rnorm(10, 3, 1))

tf <- data.frame(
                 factor1 = rep(c('A', 'B'), each = 20,times=4),
                 factor2 = rep(c('A', 'B', 'A', 'B'), each = 10, times = 4),
                 factor3 = rep(c(LETTERS[1:4]), each = 40),
                 value = rep(c(va, vb, vc, vd), times = 4)
                 )

The plot below is with a discrete x-axis and a continuous y-axis:

ggplot(tf, aes(factor1, value)) +
    geom_boxplot(aes(fill = factor2)) +
    facet_wrap(~factor3, ncol = 2) +
    theme(aspect.ratio = 2 / 3)

And the plot produced:

The width of each boxplot is too narrow. This also happens when I use facet_grid():

ggplot(tf, aes(factor1, value)) +
    geom_boxplot(aes(fill = factor2)) +
    facet_grid(.~factor3) +
    theme(aspect.ratio = 2 / 3)

If I remove facet_wrap(), the plot looks good:

ggplot(tf, aes(factor1, value)) +
    geom_boxplot(aes(fill = factor2)) +
    theme(aspect.ratio = 2 / 3)

If I increase width, the boxes are off the ticks:

ggplot(tf, aes(factor1, value)) +
    geom_boxplot(aes(fill = factor2), width = 2) +
    facet_wrap(~factor3, ncol = 2) +
    theme(aspect.ratio = 2 / 3)

I don't know if it is a bug, or something internal changed from the 2.2.1 release. My original code works fine on release version, but has this issue on the dev version.

[karawoo]

This looks like a bug with the new position_dodge2() function that controls box plot placement in the dev version 😞 . A workaround is to use the preserve = "total" argument:

ggplot(tf, aes(factor1, value)) +
    geom_boxplot(aes(fill = factor2), position = position_dodge2(preserve = "total")) +
    facet_wrap(~factor3, ncol = 2)

[Nekochef]

Thanks! That solved the issue. Hoping for further improvement.

[hadley]

@karawoo any ideas on the source of the bug? I can take a look if you can point me in the right direction.

[karawoo]

Actually I think it's not really a bug, but just a consequence of the fact that position_dodge2()'s preserve argument defaults to "single" whereas position_dodge() uses "total" by default. We could change the default for position_dodge2() to match position_dodge(). However this makes the default value incompatible with variable-width box plots, which need preserve = "single".

[karawoo]

@hadley do you have an opinion on what the default behavior should be here? I can put together a PR.

[hadley]

The facetting shouldn’t affect the width - I think there should be enough information to avoid that? (But maybe I’m forgetting something)

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/