Highcharts has long been a favourite visualisation library of mine, and I’ve written before about Highcharter, my preferred way to use Highcharts in R.

Highcharter has a nice simple function, hcboxplot(), to generate boxplots. I recently generated some for a project at work and was asked: can we see how many observations make up the distribution for each category? This is a common issue with boxplots and there are a few solutions such as: overlay the box on a jitter plot to get some idea of the number of points, or try a violin plot, or a so-called bee-swarm plot. In Highcharts, I figured there should be a method to get the number of observations, which could then be displayed in a tool-tip on mouse-over.

There wasn’t, so I wrote one like this.

First, you’ll need to install highcharter from Github to make it work with the latest dplyr.

Next, we generate a reproducible dataset using the wakefield package. For some reason, we want to look at age by gender, but only for redheads:

library(dplyr)
library(tidyr)
library(highcharter)
library(wakefield)
library(tibble)

set.seed(1001)
sample_data <- r_data_frame(
  n = 1000,
  age(x = 10:90),
  gender,
  hair
) %>%
  filter(hair == "Red")

sample_data %>%
  count(Gender)

## # A tibble: 2 x 2
##   Gender     n
##   <fctr> <int>
## 1   Male    62
## 2 Female    48

Giving us 62 male and 48 female redheads. The tibble package is required because later on, our boxplot function calls the function has_name from that package.

The standard hcboxplot function shows us, on mouse-over, the summary data used in the boxplot, as in the image below.

hcboxplot(x = sample_data$Age, var = sample_data$Gender) %>%
  hc_chart(type = "column")

To replace that with number of observations per group, we need to edit the function. In RStudio, View(hcboxplot) will open a tab with the (read-only) code, which can be copy/pasted and edited. Look for the function named get_box_values, which uses the R boxplot.stats function to generated a data frame:

  get_box_values <- function(x) {
    boxplot.stats(x)$stats %>% t() %>% as.data.frame() %>% 
      setNames(c("low", "q1", "median", "q3", "high"))
  }

Edit it to look like this – the new function just adds a column obs with number of observations:

get_box_values <- function(x) {
    boxplot.stats(x)$stats %>% t() %>% cbind(boxplot.stats(x)$n) %>% as.data.frame() %>% 
      setNames(c("low", "q1", "median", "q3", "high", "obs"))
  }

Save the new function as, for example, my_hcboxplot. Now we can customise the tooltip to use the obs property of the point object:

my_hcboxplot(x = sample_data$Age, var = sample_data$Gender) %>%
  hc_chart(type = "column") %>%
  hc_tooltip(pointFormat = 'n = {point.obs}')

Voilà.