As part of a page proof request, a boxplot with the ggpubr package was flagged as needing exponential notation for the small pvalues (e.g., 1x10^-4 vs 1e-04). That seemed easy. And it was, after a long battle with expressions, bquote, and other R concepts…

Getting Started

I had used the excellent ggpubr package for a boxplot with points (using add="jitter") and a wilcoxon test of differences using stat_compare_means(). Something equivalent to:

ggpubr::ggboxplot(
    iris,
    x="Species",
    y="Sepal.Length",
     add="jitter"
) + 
    ggpubr::stat_compare_means()

which generates:

So if I can switch from p < 2.2e-16 to p < 2.2^-16 I’d be in good shape.

First Step

To get the p value in different format, it looked like I would need the following:

ggpubr::stat_compare_means(
    ggplot2::aes(
        label = do_something(
            ggplot2::after_stat(p.format)
        ),
    parse = TRUE
)

This actually consists of two different concepts which took me a while to figure out (both of them).

after_stat

The easiest was that I had to change the aesthetics for the compare_means function call. Notice that I have a magic function do_something() that I haven’t figured out yet. But it will eventually format the p value to what I need. The magic of ggplot2::after_stat(p.format) is actually the way to get the calculated p values to do something with them. As should be obvious, p.format would be the formatted p value. Except nothing is ever completely obvious. Most of the time it is the numerical p value as string unless the p value is very small. Then it becomes <2e-16 (for instance). But either way, I can get the computed p value and format it how I’d like.

plotmath

The second piece is the parse=TRUE. There is a whole universe of using math notation (like latex) within graphs in R. This is called plotmath and is definitely worth knowing. Lots of nice math layout is possible using this approach.

But, when using plotmath the goal is to pass expressions (like 10^2) which can be then evaluated and rendered appropriately. The ggplot family of things make like easier by having a parameter parse=TRUE that can be passed along with the strings that represent the expressions. This allows us to use notation like 10^2 which would render 10².

A solution

Given this setup, I just need a do_something() that would create a parseable string representing a plotmath function. I’ve seen several stackexchange posts about this:

gsub("e", " %*% 10^", pvalue)

This cleverly replaces the e with x 10^, meaning you get the correctly formatted p value in plotmath format.

Therefore,

do_something <- function(x) {
   paste0(
        "p ",
        ifelse(
            startsWith(x, "<"), 
            "",
            "== "
        ),
        gsub("e", " %*% 10^",x)
    )
}

Note that the code here is vectorized, since the labels can be numerous if the figure is faceted. Also note that it handles the < val or val approach.

Putting it together,

do_something <- function(x) {
   paste0(
      "p ",
      ifelse(startsWith(x, "<"), "","== "),
      gsub("e", " %*% 10^",x)
    )
}
ggpubr::ggboxplot(
  iris, 
  y= "Sepal.Length",
  x="Species",
  add="jitter"
) +
  ggpubr::stat_compare_means(
    ggplot2::aes(
        label=do_something(
            ggplot2::after_stat(p.format)
        )
    ),
    parse = TRUE
  )

The result is:

(Note: The p value is slightly cutoff, but you can add vjust=1 to the call to adjust it slightly downward).