ggplot2: forcing space for empty second-level category

Question

I'm trying to keep "empty space" for multi-level grouped boxplots.

set.seed(42)
n <- 100
dat <- data.frame(x=runif(n),
                  cat1=sample(letters[1:4], size=n, replace=TRUE),
                  cat2=sample(LETTERS[1:3], size=n, replace=TRUE))
ggplot(dat, aes(cat1, x)) + geom_boxplot(aes(fill=cat2))

If I force one of the groups to be empty:

dat <- subset(dat, ! (cat1 == 'b' & cat2 == 'B'))
table(dat$cat1, dat$cat2)
##    
##      A  B  C
##   a  9  9  7
##   b  8  0  5
##   c 13 11  6
##   d 11 10  5
ggplot(dat, aes(cat1, x)) + geom_boxplot(aes(fill=cat2))

The second group, "b", is now expanded to fill the space. What I'd like is:

SO 9818835 (forcing an empty level to appear) works fine on the top level, but I can't figure out how to get it to work for a second level of categories. in scale_x_discrete(...), I tried setting:

• breaks=letters[1:4]
• breaks=LETTERS[1:3]
• breaks=list(letters[1:4], LETTERS[1:3]) (a stab)
• breaks=NULL
• breaks=func where func <- function(x, ...) { browser(); 1; } in order to troubleshoot; it only offered letters[1:4] and never prompted for the second level

Using interactions(letters[1:4], LETTERS[1:3]) still does not leave empty space. I tried a workaround by injecting an out-of-bounds x value and forcing it off the screen with scale_y_continuous(limits), but ggplot2 is too smart for me and closes the gap again.

Are there elegant (i.e., "correct" in ggplot2 mechanisms) solutions?

Answer 1

Could coord_cartesian be a solution that you are looking for?

It will zoom in and will not try to "outsmart" the data like scale_y_continuous

library(dplyr)
library(ggplot2)

set.seed(42)
n <- 100
dat <- data.frame(x=runif(n),
                  cat1=sample(letters[1:4], size=n, replace=TRUE),
                  cat2=sample(LETTERS[1:3], size=n, replace=TRUE))

LARGE_VALUE <- 2

dat <- dat %>%
  mutate(x = ifelse(cat1 == 'b' & cat2 == 'B', 
                    LARGE_VALUE,
                    x))

ggplot(dat, aes(cat1, x)) + 
  geom_boxplot(aes(fill=cat2)) + 
  coord_cartesian(ylim = c(0,1))