Question

Ggplot2 allows one to add error bars to a plot. To calculate the error bar limits for you, it wraps functions from Hmisc. For example, to bootstrap one can use the mean_cl_boot option:

m <- ggplot(mtcars, aes(x=cyl, y=am)) + stat_summary(fun.y=mean,geom="point")
m2 <- m + stat_summary(fun.data = "mean_cl_boot", geom = "errorbar", conf.int=.95)
m2

But what if you need to write a custom function to calculate the error bar limits? How should the function be written to be invoked from a stat_summary call?

Answer 1

Here’s an example using the add4ci function from PropCIs providing a confidence interval. You just need to have the function return a list of numbers named "y", "ymin", and "ymax":

library(PropCIs)
add4ciForGgplot <- function(x,conf.int) {
  numCorrect <- sum(x)
  numTrials <- length(x)
  CI <- add4ci(numCorrect,numTrials,conf.int)
  triplet <- data.frame(numCorrect/numTrials, CI$conf.int[1], CI$conf.int[2])
  names(triplet) <- c("y","ymin","ymax") #this is what ggplot is expecting
  return (triplet)
}

m <- ggplot(mtcars, aes(x=cyl, y=am)) + stat_summary(fun.y=mean,geom="point")
mCustom <- m + stat_summary(fun.data = "add4ciForGgplot", geom = "errorbar", conf.int=.95)
mCustom