Plot divergent stacked bar chart with ggplot2

Question

Is there a way to use ggplot2 to create divergent stacked bar charts like the one on the right-hand side of the image below?

Data for reproducible example

library(ggplot2)
library(scales)
library(reshape)

dat <- read.table(text = "    ONE TWO THREE
                  1   23  234 324
                  2   34  534 12
                  3   56  324 124
                  4   34  234 124
                  5   123 534 654",sep = "",header = TRUE)

# reshape data
datm <- melt(cbind(dat, ind = rownames(dat)), id.vars = c('ind'))

# plot
ggplot(datm,aes(x = variable, y = value,fill = ind)) + 
  geom_bar(position = "fill",stat = "identity") +
  coord_flip()

Answer 1

Sure, positive values stack positively, negative values stack negatively. Don't use position fill. Just define what you want as negative values, and actually make them negative. Your example only has positive scores. E.g.

ggplot(datm, aes(x = variable, y = ifelse(ind %in% 1:2, -value, value), fill = ind)) + 
    geom_col() +
    coord_flip()

If you want to also scale to 1, you need some preprocessing:

library(dplyr)
datm %>% 
  group_by(variable) %>% 
  mutate(value = value / sum(value)) %>% 
  ggplot(aes(x = variable, y = ifelse(ind %in% 1:2, -value, value), fill = ind)) + 
  geom_col() +
  coord_flip()

Answer 2

An extreme approach might be to calculate the boxes yourself. Here's one method

dd <- datm %>% group_by(variable) %>% 
  arrange(desc(ind)) %>% 
  mutate(pct = value/sum(value), right = cumsum(pct), left=lag(right, default=0))

then you can plot with

ggplot(dd) + 
  geom_rect(aes(xmin=right, xmax=left, ymin=as.numeric(variable)-.4, ymax=as.numeric(variable)+.4, fill=ind)) + 
  scale_y_continuous(labels=levels(dd$variable), breaks=1:nlevels(dd$variable))

to get the left plot. and to get the right, you just shift the boxes a bit. This will line up all the right edges of the ind 3 boxes.

ggplot(dd %>% group_by(variable) %>% mutate(left=left-right[ind==3], right=right-right[ind==3])) + 
  geom_rect(aes(xmin=right, xmax=left, ymin=as.numeric(variable)-.4, ymax=as.numeric(variable)+.4, fill=ind)) + 
  scale_y_continuous(labels=levels(dd$variable), breaks=1:nlevels(dd$variable))

So maybe overkill here, but you have a lot of control this way.