'I've been waiting for a guide to come and take me by the hand': Ridgeline plots with {ggridges}

I really like ridgeline plots but only recently I have learned how to do them myself. Of course, the most famous ridgeline plot ever is the one you find on the cover of Joy Division’s album “Unknown Pleasures”. I wonder how many ridgeline plots done with the {ggridges} package try to replicate the look of this famous (and great!) album. And - of course - I will try, too.

But first, we gonna need some data. I am using a dataset I have compiled myself in my job at the Leibniz Institute for the German Language. I will try to explain with a few words what’s in the data. To download the data and read it into the res object, I am using code I found on SE. Note that this ZIP file will change whenever I update the data (usually on Fridays). So, your results may vary. Basically, more data should be plotted each week.

temp <- tempfile()
download.file("https://www.owid.de/plus/cowidplus2020/data/unigram-results.csv.zip",
              temp)
con <- unz(temp, "unigram-results.csv")
res <- read.table(con, sep = ",", header = T)
unlink(temp)
scroll_box(kable_styling(kable(res)),
           width = "100%", height = "400px")

The scroll box above shows the data. The dataset contains measures for a linguistic corpus of German online news RSS feeds. In the analyses presented on OWID^plus (only in German), we are interested in the diversity of the vocabulary in these RSS feeds. We show that during the start of the coronavirus pandemic, the vocabulary gets less diverse. This effect is strongest from mid-March to mid-April. The columns in the dataset encode the following information (I will only introduce the two variables we gonna need below):

• date: The date, duh!
• top100share: This is a very simple measure of lexical diversity. It simply the total frequency of the 100 most frequent types (different words) in the corpus on that day divided by the total frequency (column tok). So, on January 1^st, 2020, approx. 43% of all tokens (running words) belonged to the 100 most frequent word forms.

redundancy and msttr are two other measures of lexical diversity. top100share and msttr (mean segmental type-token ratio) are highly correlated (redundancy not so much), so it might not matter which one we use. I’ll go with the top100share.

cor(res[,c("redundancy", "msttr", "top100share")])

##             redundancy      msttr top100share
## redundancy   1.0000000  0.0498225   0.2105027
## msttr        0.0498225  1.0000000  -0.8798542
## top100share  0.2105027 -0.8798542   1.0000000

Next, I am loading the necessary packages. The column week shall contain the week number of 2020. I am re-ordering the week factor because I want to display the weeks from top to bottom in the final plot. I’m finding this a little more intuitive than displaying a continuous variable from bottom to top - I guess it’s up to personal preference.

library(ggridges)
library(ggplot2)
suppressPackageStartupMessages(library(lubridate))
res$week <- week(res$date)
res$week <- factor(res$week, levels = max(res$week):1)

Next up, I’m plotting the ridgeline with the default settings.

ggplot(res, aes(x = top100share, y = week, fill = week)) +
  geom_density_ridges(scale = 4) +
  scale_y_discrete(expand = c(0, 0)) +
  scale_x_continuous(expand = c(0, 0)) +
  coord_cartesian(clip = "off") +
  theme_ridges(grid = F) +
  guides(fill = F)

## Picking joint bandwidth of 0.00317

For a scientific audience who wants to understand what’s going on, this plot might be fine. The annotations help us identifying the weeks and a little axis annotation certainly doesn’t hurt. Also you can clearly see how the density distribution of vocabulary diversity shifts to the right during March and April and gradually turns back to the original value(s).

But that’s not what we’re here for, isn’t it? Let’s tweak some parameters to get that Joy Division vibe. That involves filling the density curves black, using white lines and basically deleting all of the annotations. The larger margins resemble the original album cover.

ggplot(res, aes(x = top100share, y = week)) +
  geom_density_ridges(scale = 4, col = "white", fill = "black") +
  scale_y_discrete(expand = c(0, 0)) +
  scale_x_continuous(expand = c(0, 0)) +
  coord_cartesian(clip = "off") +
  theme_ridges(grid = F) +
  labs(x = "", y = "") +
  theme(axis.text.x = element_blank(),
        axis.ticks = element_blank(),
        axis.text.y = element_blank(),
        plot.background = element_rect(fill = "black"),
        plot.margin = unit(c(5,5,5,5), "cm"))

## Picking joint bandwidth of 0.00317

Nice.