Drawing panel charts in R

In this blog post, I describe how to draw panel charts as an alternative to stacked bar charts in R, using the {ggplot2} package.

Author

Affiliation

Lukas Sönning

University of Bamberg

I recently wrote a paper on the use of bar charts in corpus-linguistic research articles (see here). One of the questions of this systematic review was whether the bar chart is always the best choice. In Section 6.1 of the paper I discuss panel charts (see Camões (2016, 217–18), Schwabish (2021, 88–89)) as an alternative to stacked bar charts. In this blog post, I describe how to draw such a chart using the R package {ggplot2}.

library(tidyverse)

Data

For illustration, we use data from Larsson et al. (2020), a nice corpus study that looks at adverb placement in learner writing. These data are also the basis of Figure 6 and Figure 17 of the bar chart paper referenced above. They include frequency counts for cross-classifications of two categorical variables:

• adverb: n = 13 levels
• position: n = 4 levels

The counts reflect how often a particular adverb was observed in a particular position in the clause.

d <- data.frame(
    adverb = rep(c(
      "maybe", "perhaps", "of course", "possibly", "apparently", 
      "obviously", "certainly", "actually", "clearly", "simply", 
      "probably", "definitely", "really"), 4),
    position = rep(c("M3", "M2", "M1", "I"), each = 13),
    count = c(
      13, 83, 85, 10, 25, 50, 61, 124, 138, 152, 169, 36, 58,
      12, 70, 57, 66, 12, 40, 47, 181, 54, 154, 163, 43, 100,
      8, 24, 36, 21, 31, 48, 47, 305, 142, 171, 119, 16, 121,
      61, 112, 105, 26, 16, 31, 17, 61, 28, 33, 28, 4, 5)
)

Let’s look at the contents of the data frame:

str(d)

'data.frame':   52 obs. of  3 variables:
 $ adverb  : chr  "maybe" "perhaps" "of course" "possibly" ...
 $ position: chr  "M3" "M3" "M3" "M3" ...
 $ count   : num  13 83 85 10 25 50 61 124 138 152 ...

The next step is to add proportions to the data frame. These will reflect the relative frequency of the four positions within each adverb. For each adverb, the proportions therefore sum to 1. The {dplyr} package is useful for this task.

d <- d |> 
  group_by(adverb) |> 
    mutate(n_total = sum(count)) |> 
    ungroup() |> 
    mutate(prop = count / n_total)

In principle, we are now good to go. However, I would also like to be able to order the adverbs in the plot according to the proportional share of position I (initial), so we also add a new column giving this value.

d <- d |> 
    group_by(adverb) |> 
    mutate(prop_initial = prop[position == "I"]) |> 
    mutate(position = fct_rev(position))

Here is what the first few rows of the data frame look like:

head(d)

# A tibble: 6 × 6
# Groups:   adverb [6]
  adverb     position count n_total   prop prop_initial
  <chr>      <fct>    <dbl>   <dbl>  <dbl>        <dbl>
1 maybe      M3          13      94 0.138         0.649
2 perhaps    M3          83     289 0.287         0.388
3 of course  M3          85     283 0.300         0.371
4 possibly   M3          10     123 0.0813        0.211
5 apparently M3          25      84 0.298         0.190
6 obviously  M3          50     169 0.296         0.183

Draw a stacked bar chart

We start by drawing a stacked bar chart, the rudimentary version first.

d |> 
    ggplot(aes(
      x = reorder(adverb,          # adverb on the x-axis
                  -prop_initial),  # order adverbs by prop_initial
      y = prop,                    # the proportional share as bar segments
      fill = position)) +          # position as a fill variable
    geom_col() +                   # geom for bar segments
    scale_fill_grey()              # use grey fill colors

And the following enhanced plotting call creates a graph that mimics Figure 3 in Larsson et al. (2020), and which appears as Figure 6 in the bar chart paper.

d |> 
    ggplot(aes(
      x = reorder(adverb, 
                  -prop_initial), 
      y = prop, 
      fill = position)) +
    geom_col(
      col = 1,                      # add black contours around the bar segments
      linewidth = .4,               # make these contours thin         
      width = .6) +                 # reduce width of the columns of bars
    scale_fill_grey(
      start = .4, end = 1) +        # customize shades of grey
    ylab("Percentage of tokens") +  # new y-axis title
  xlab(NULL) +                    # remove x-axis title
    theme_minimal() +               # use different theme
    scale_x_discrete(
      expand = c(.05, .05)) +       # control padding at left/right margin of plot
    scale_y_continuous(
      labels = c(0, 25, 50, 75, 100), # define locations of tick marks on y-axis
      expand = c(.01, .01),           # control padding at top/bottom margin of plot
      minor_breaks = NULL) +          # omit minor grid lines
  theme(
    axis.text.x = element_text(            # modify tick mark labels on x-axis
      angle = 45,                          #   rotate by 45 degrees
      hjust=1.2,                           #   control horizontal adjustment
      vjust=1.3,                           #   control vertical adjustment
      face = "italic"),                    #   print adverbs in italics
    panel.grid.major.x = element_blank(),  # no vertical grid lines
    legend.key.size = unit(.3, 'cm'))      # reduce size of tiles in key

Panel chart

Next, we draw a panel chart, basic version first. Since we want to order the levels of position logically (initial at the far left), we must first reorder the factor levels of this variable:

d <- d |> 
    mutate(position = fct_rev(position))

To switch to a panel chart, the key change is to make position into a faceting variable, which means that the different positions will appear in different panels. This requires adding four lines of code, which call facet_grid(). The arguments of this function do the following:

• the first argument uses formula notation to state how the panels representing different levles of position should be arrange in the grid:
  • . ~ position: side by side, in a single row
  • position ~ .: on top of one another, as a single column
• scales = "free_x" says that the panels need not have the same limits
• space = "free_x" says that the panels need not have the same physical width

d |>  
    ggplot(aes(
      y = reorder(adverb, -prop_initial), 
      x = prop, 
      fill = position)) +
    geom_col() +
    scale_fill_grey() +
    facet_grid(
      . ~ position,      # draw separate facets for position, in rows
      scales = "free_x", # allow upper limits of x-axis to vary across facets
      space = "free_x")  # allow width of facets to vary

And the following enhanced code draws the version that appears as Figure 17 in the bar chart paper.

d |>  
  ggplot(aes(
    y = reorder(adverb, -prop_initial), 
    x = prop, 
    fill = position)) +
    geom_col(
      col = 1,                      # add black contours around the bar segments
      linewidth = .4,               # make these contours thin         
      width = .6) +                 # reduce width of the columns of bars
    scale_fill_grey(
      start = .4, end = 1) +        # customize shades of grey
    ylab("Percentage of tokens") +  # new y-axis title
  xlab(NULL) +                    # remove x-axis title
    theme_minimal() +               # use different theme
  facet_grid(
    . ~ position, 
    scales = "free_x", 
    space = "free_x") +
  geom_vline(xintercept = 0) +    # add a black reference line at 0
  scale_x_continuous(
    breaks = c(0, .2, .4, .6),    # define locations of tick marks on y-axis
    labels = c(0, 20, 40, 60),    # define tick mark labels
      expand = c(.01, .01),         # control padding at top/bottom margin of plot
      minor_breaks = NULL) +        # omit minor grid lines
  theme(
    legend.position = "none",              # omit legend
    panel.grid.major.y = element_blank(),  # no horizontal grid lines
    axis.text.y = element_text(
      face = "italic"),                    # print adverbs in italics
    panel.spacing = unit(1.2, "lines"))    # adjust spacing between facets

References

Camões, Jorge. 2016. Data at Work: Best Practices for Creating Effective Charts and Information Graphics in Microsoft Excel. New Riders.

Larsson, Tove, Marcus Callies, Hilde Hasselgård, Natalia Judith Laso, Sanne van Vuuren, Isabel Verdaguer, and Magali Paquot. 2020. “Adverb Placement in EFL Academic Writing: Going Beyond Syntactic Transfer.” International Journal of Corpus Linguistics 25 (2): 156–85. https://doi.org/10.1075/ijcl.19131.lar.

Schwabish, Johnathan. 2021. Better Data Visualizations: A Guide for Scholars, Researchers, and Wonks. New York: Columbia University Press.