I’m a big fan of ggplot2. Recently, I ran into a situation which called for a useful feature that I had not used previously: aes_string.

Imagine that you have data consisting of observations for several variables – let’s say A, B, C – where each observation is from one of two groups – call them X and Y:

df1 <- data.frame(A = rnorm(50), B = rnorm(50), 
                  C = rnorm(50), group = rep(LETTERS[24:25], 25))
head(df1)
#           A          B           C group
# 1 0.2748922 -0.4805635 -1.80242191     X
# 2 0.0060852 -1.2972077  0.64262069     Y
# 3 0.1994655 -0.4628783  0.07670911     X
# 4 0.5416900  0.3853958  0.50193895     Y
# 5 0.3118773  0.9488503 -0.55855749     X
# 6 2.0924626  0.3027878 -0.03000122     Y

If you were interested in the distribution of variable A by group, you might generate a boxplot like so:

png("A.png", width = 800, height = 600) print(ggplot(df1) + geom_boxplot(aes(group, A, fill = group)) + theme_bw()) dev.off()

Boxplot of A by group

Here, the arguments to aes() are expressions (group, A) which ggplot interprets as column names from the data frame.

What if you wanted to generate plots for each of variable A, B and C using a loop? You might start like this:

for(i in names(df1)[1:3])
# oh wait, these are characters not expressions
# [1] "A" "B" "C"

You see the problem. How do we pass the column names which are characters, not expressions, to aes()?

The answer: use aes_string() instead.

Description:
     Aesthetic mappings describe how variables in the data are mapped
     to visual properties (aesthetics) of geoms.  Compared to aes this
     function operates on strings rather than expressions.

And so:

for(i in names(df1)[1:3]) {
  png(paste(i, "png", sep = "."), width = 800, height = 600)
  df2 <- df1[, c(i, "group")]
  print(ggplot(df2) + geom_boxplot(aes_string(x = "group", y = i, fill = "group")) + theme_bw())
  dev.off()
}

It’s a little ugly as it stands (better to write a function using one of the apply family). However, the key point is: you can pass data frame column names as expressions to aes() or as characters to aes_string().

With thanks to Hadley’s contribution to this mailing list thread.