Matt Bannert - 7 months ago 48

R Question

Yes, I know it's been around, I've also found Hadley's answer on google groups that there is no notches yet for

`ggplot2`

I mean I do not need the notch optic, representing the confidence bounds by some shaded area that is suitably placed in another layer over the boxplot, would look nice, too.

Also added a screenshot because I heard a graphics question is never complete without the graphic

Answer

**Update**
In addition to the options detailed below, version 0.9.0 of **ggplot2** includes this feature in `geom_boxplot`

. Examining `?geom_boxplot`

reveals a `notch`

and `notchwidth`

argument:

```
+ geom_boxplot(notch = TRUE, notchwidth = 0.5)
```

Not elegant graphics but here is an example:

```
# confidence interval calculated by `boxplot.stats`
f <- function(x) {
ans <- boxplot.stats(x)
data.frame(ymin = ans$conf[1], ymax = ans$conf[2])
}
# overlay plot (upper panel below)
p <- ggplot(iris, aes(Species, Sepal.Length)) + geom_boxplot() +
stat_summary(fun.data = f, geom = "linerange", colour = "skyblue", size = 5)
p
# base graphics (lower panel below)
boxplot(Sepal.Length ~ Species, data = iris, notch = TRUE)
```

you can change the apparence of CI bar by tweaking the arguments of `stat_summary`

.

crossbar version:

```
f <- function(x) {
ans <- boxplot.stats(x)
data.frame(ymin = ans$conf[1], ymax = ans$conf[2], y = ans$stats[3])
}
p <- ggplot(iris, aes(Species, Sepal.Length)) +
geom_boxplot(width = 0.8) +
stat_summary(fun.data = f, geom = "crossbar",
colour = NA, fill = "skyblue", width = 0.8, alpha = 0.5)
p
```