{highcharts} doesn’t like empty factors (and the .by argument is quite nice, actually)

Heard of {plotly} for interactive charts? I massively prefer {highcharts} as it’s more accessible, mobile friendly and simply more beautiful. Both of these packages are htmlwidgets and there are so many packages for making interactive charts, maps and more.

But. We’re here to discuss how {highcharter} doesn’t like empty factor levels - with my own YouTubesque thumbnail.

Let’s demonstrate how {ggplot2} handles ordering of factor levels that are empty:

library(tidyverse)

gss_party_by_marital <- gss_cat %>% 
  summarise(n_in_subcategory = n(), .by = c(partyid, marital)) %>% 
  mutate(n_in_category = sum(n_in_subcategory), .by = partyid) %>% 
  mutate(partyid = fct_reorder(partyid, n_in_category))

gss_party_by_marital %>% 
  ggplot(aes(x = n_in_subcategory,
             y = partyid,
             fill = marital)) +
  geom_col() +
  scale_fill_viridis_d(option = "A", direction = 1) +
  guides(fill = guide_legend(reverse = TRUE))

That’s a nicely ordered stacked bar chart! We’ve got the bars going from big to small, even though the final two categories don’t contain all of the marital states.

gss_party_by_marital %>% 
  filter(partyid %in% c("No answer", "Don't know")) %>% 
  knitr::kable()

partyid	marital	n_in_subcategory	n_in_category
No answer	Married	60	154
No answer	Divorced	29	154
No answer	Never married	35	154
No answer	Widowed	16	154
No answer	Separated	9	154
No answer	No answer	5	154
Don’t know	Married	1	1

Let’s throw this into {highcharter} and we see just a complete mess.

library(highcharter)

gss_party_by_marital %>% 
  hchart(
    type = "bar",
    hcaes(
      y = n_in_subcategory,
      x = partyid,
      group = marital
    )
  ) %>% 
  hc_plotOptions(series = list(stacking = "normal"))

Using complete() and talking about that .by argument

You might have missed it in my code above, but I used groups without group_by() or ungroup()! Back in late 2022 there was a suggestion that the grouping functions gain a .by argument which was added as an experimental feature.

I initially I really didn’t like it. But, look how neat this is. I’ve highlighted the summarise() and mutate() lines that make use of the .by argument, it saves having to call group_by() and ungroup()… and just to be clear complete() is filling in empty factor levels with an explicit value of 0.

gss_completed_data <- gss_cat %>% 
  summarise(n_in_subcategory = n(), .by = c(partyid, marital)) %>% 
  complete(partyid,
           marital,
           fill = list(n_in_subcategory = 0)) %>% 
  mutate(n_in_category = sum(n_in_subcategory), .by = partyid) %>% 
  mutate(partyid = fct_reorder(partyid, n_in_category)) %>% 
  arrange(desc(n_in_category))

gss_completed_data %>% 
  filter(partyid %in% c("No answer", "Don't know")) %>% 
  knitr::kable()

partyid	marital	n_in_subcategory	n_in_category
No answer	No answer	5	154
No answer	Never married	35	154
No answer	Separated	9	154
No answer	Divorced	29	154
No answer	Widowed	16	154
No answer	Married	60	154
Don’t know	No answer	0	1
Don’t know	Never married	0	1
Don’t know	Separated	0	1
Don’t know	Divorced	0	1
Don’t know	Widowed	0	1
Don’t know	Married	1	1

Now those empty factor levels are filled we can throw it back into {highcharter}. Make sure to interact with the chart, and click on the legend items to see just how beautiful {highcharter} is.

gss_completed_data %>% 
  hchart(
    type = "bar",
    hcaes(
      y = n_in_subcategory,
      x = partyid,
      group = marital
    )
  ) %>% 
  hc_plotOptions(series = list(stacking = "normal"))

A necessary aside on licensing

{highcharter} is a wrapper for the incredible https://highcharts.com/ JavaScript framework. It is not free to use {highcharter} in commercial projects, but there are free license options for personal projects and educational usage.