How to create a customized "geom_box" in ggplot2?

Question

I'd like to have something similar to geom_boxplot, but that will only have a box, and that I can set the function for the lower and upper parts of the box, for example, showing plus minus 1 SD of the data from the mean. I am not if stat_boxplot can be used for this purpose or if some other function would fit better.

This can be (almost) done manually on data using stat="identity" and pre-computation, for example:

y <- rnorm(100)
y1 <- mean(y) - sd(y)
y2 <- mean(y) + sd(y)
df1 <- data.frame(y)
df2 <- data.frame(
  x = 1,
  y0 = y1,
  y25 = y1,
  y50 = y1, # this is a problem...
  y75 = y2,
  y100 = y2
)

ggplot(df1, aes(x=1,y=y)) +
geom_boxplot() + 
  geom_boxplot(data = df2,
   mapping = aes(x = 1, y = 1, ymin = y0, lower = y25, middle = y50, upper = y75, ymax = y100),
   stat = "identity", alpha = 0.1, fill = "red")

This example has several problems:

the box is not of the width of the boxplot
the lower part of the box has a thicker line than the upper part (since I needed to say where "middle" should be)
The computation needs to happen manually per data (which could be wrapped in a function, but due to the other problems, I didn't get to that yet).

In short, I'd like something like geom_box but couldn't find it from a google search, and I'd be happy for some directions on how to proceed with writing such a customized geom function (I guess this is a start, but some more help would be welcomed).

@zx8754 well, if it is a geom_box, than they could both be superimposed on each other or not (i.e.: ggplot(df1) + geom_boxplot(...stuff...) + geom_box(...slightly other stuff...) ) — Tal Galili
Apparently width is a computed variable (along with the box and whisker ranges). It appears that the width of the original boxplot is 0.75, so you could add to the aes of your second geom_boxplot call. An option for middle would be to put it at the median, which would then draw it in the same place as the original boxplot. — aosmith
@aosmith - that is possible, but what if I want to have the box without the boxplot? maybe it could be based on geom_rect?! — Tal Galili

Brian Brian · Accepted Answer · 2017-09-29T21:26:24

newbox <- function(values) {
  values <- na.omit(values)
  data.frame(
    ymin = mean(values) - sd(values),
    lower = mean(values) - sd(values),
    middle = mean(values),
    upper = mean(values) + sd(values),
    ymax = mean(values) + sd(values),
    width = 0.75
  )
}

ggplot(iris, aes(Species, Sepal.Length)) + 
  stat_summary(fun.data = newbox, geom = "boxplot", fatten = NA)

Like that?

If you want it with no fill, you could use built-in functions like:

ggplot(iris, aes(Species, Sepal.Length)) + 
  stat_summary(fun.data = mean_sdl, fun.args = list(mult = 1), 
               geom = "crossbar", fatten = NA)

How to create a customized "geom_box" in ggplot2?

1 Answers