Drawing grouped dot plots in R

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

Author

Affiliation

Lukas Sönning

University of Bamberg

The dot plot was introduced by Cleveland (1984) to address some of the shortcomings of bar charts. In Sönning (2016), I discuss and illustrate the potential of this graph type for language data analysis and presentation. In the course of writing that paper, I also created templates for Microsoft Excel, which make it easy to draw such graphs. They can be found here.

In a recent study, I looked at the use of bar charts in corpus-linguistic research articles. In this systematic review, I came across settings in which a grouped bar chart could have been fruitfully replaced by a grouped dot plot. In this blog post, I describe how to draw such a chart using the R package {ggplot2}.

library(tidyverse)

Data

For illustration, I will use data from Maguire, Wisniewski, and Storms (2010), a corpus study on semantic patterns in compounds consisting of two nominal constituents. These data are also the basis of Figure 5 and Figure 16 of the bar chart paper referenced above. The data include percentages for cross-classifications of two categorical variables:

• Category: semantic category of the nominal constituent, n = 25 levels
• Function: function of the nominal constituent in the compound, n = 2 levels (“Modifier” vs. “Head”)

The percentages, which are approximations taken from Figure 2 of Maguire, Wisniewski, and Storms (2010), express the rate at which nouns in these semantic categories appear in compounds, using as a baseline the total number of occurrences of nouns in the category in the corpus. These percentages are reported separately for the two functional roles, modifier and head.

d <- data.frame(
  Category = rep(c(
    "Act", "Animal", "Artifact", "Attribute", "Body", "Cognition",
    "Communication", "Event", "Feeling", "Food", "Group", "Location",
    "Motive", "Object", "Person", "Phenomenon", "Plant", "Possession",
    "Process", "Quantity", "Relation", "Shape", "State", "Substance", "Time"),
    each = 2),
  Function = factor(rep(c("Modifier", "Head"), 25)),
  Percent = c(
    5.9, 3.7, 7.6, 2.2, 6.9, 4.3, 2, 2.1, 6.2, 1.5, 2, 3.1, 
    3.3, 3.2, 5.3, 2.9, 
    1.3, 0.7, 6.4, 3.4, 4.8, 4.7, 9.7, 2.2, 4, 2.2, 8.1, 3.1,
    3.2, 3.4, 7.1, 2.6, 10.5, 4.2, 8.9, 4.7, 4.2, 5.4, 6, 1.8,
    5.9, 5.6, 4.1, 2.3, 5, 2.5, 13.2, 3.7, 5.3, 1.1)*2.5
)

To mimic the original chart as closely as possible, we want the Modifier category to appear to the left of the Head category in each cluster of bars. We must therefore reorder the factor levels of the variable Function; the {dplyr} package is useful for this task.

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

Let’s print the first few rows of the table. Rows 1 and 2 represent nouns belonging to the semantic category Act. Out of all Act nouns in the corpus, 14.75% occur in the modifier slot of a N + N compound; 9.25% appear as the head noun of a N + N compound. This means that, overall, 24% of Act nouns (14.75% + 9.25%) occurred in N + N compounds.

head(d)

  Category Function Percent
1      Act Modifier   14.75
2      Act     Head    9.25
3   Animal Modifier   19.00
4   Animal     Head    5.50
5 Artifact Modifier   17.25
6 Artifact     Head   10.75

Draw a grouped bar chart

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

d |> 
  ggplot(aes(
    x = Category,         # Category on the x-axis
    y = Percent,          # the percentage as bar segments
    fill = Function)) +   # Function as a fill variable
  geom_col(               # geom for bar segments
    position = "dodge") + # dodging for *grouped* bar chart
  scale_fill_grey()       # use grey fill colors

And the following enhanced plotting call creates a graph that mimics Figure 2 in Maguire, Wisniewski, and Storms (2010), and which appears as Figure 5 in the bar chart paper. Note that the semantic categories on the x-axis appear in alphabetical order, which rarely makes sense in data visualization.

d |> 
  ggplot(aes(
    x = Category, 
    y = Percent, 
    fill = Function)) +
  geom_col(
    position = "dodge", 
    col = 1,                      # add black contours around the bars
    width = .66) +                # make these contours thin
  scale_fill_grey(
    start = .8, end = .6) +       # customize shades of grey
  ylab("Percentage of tokens") +  # add y-axis title
  xlab(NULL) +                    # remove x-axis title
  theme_minimal() +               # use different theme
  scale_x_discrete(
    expand = c(.02, .02)) +       # control padding at left/right margin
  scale_y_continuous(
    expand = c(.01, .01),         # control padding at top/bottom margin
    minor_breaks = NULL) +        # omit minor grid lines
  theme(
    axis.text.x = element_text(   # modify tick mark labels on x-axis
      angle = 90,                 #   rotate by 90 degrees
      hjust = 1,                  #   control horizontal adjustment
      vjust = .5,                 #   control vertical adjustment
      size = 10),                 #   control font size
    panel.grid.major.x = element_blank(),  # no vertical grid lines
    legend.position = "top",               # remove key
    legend.key.size = unit(.3, 'cm'))      # reduce size of tiles in key

Grouped dot pots

Next, we draw a grouped dot plot. To be able to order the semantic categories in the plot in a sensible way, we need to add new columns to the data frame. There are (at least) three ways in which we could rearrange them:

• based on the proportional share of the Modifier category
• based on the proportional share of the Head category, or
• based on the difference between these percentages

We add three new columns that record this information:

d <- d |> 
    group_by(Category) |> 
    mutate(
        percent_head = Percent[Function == "Head"],
        percent_modifier = Percent[Function == "Modifier"],
        difference = percent_head - percent_modifier)

We start by drawing a basic version of the grouped dot plot:

d |> 
    ggplot(aes(
      y = reorder(
        Category,           # Category on the y-axis
        percent_modifier),  # order categories by percent_modifier
      x = Percent,          # the percentage on the x-axis
      shape = Function)) +  # different plotting symbols for Function
    geom_point()            # geom for points

Next, we reproduce the version that appears as Figure 16 in the bar chart paper. This version includes an appended panel that directly shows the difference between the two groups that appear in the dot plot; see Sönning (2016, 113); also Amit, Heiberger, and Lane (2008) and Heiberger and Holland (2015, 566). This means that we draw two plots, which we then combine into a single display with the function plot_grid() in the {cowplot} package (Wilke 2024).

We start by drawing the main panel, which we store as a new plot object p1:

p1 <- d |> 
  ggplot(aes(
    y = reorder(
      Category,                  # Category on the y-axis
      percent_modifier),         # order categories by percent_modifier
    x = Percent,                 # the percentage on the x-axis
    shape = Function)) +         # different plotting symbols for Function
  geom_point(                    # geom for points
    size = rep(c(1.2, 1), 25)) + # control size of plotting symbols
  scale_shape_manual(
    values = c(1, 3)) +          # control choice of plotting symbols
  xlab("Percentage of tokens") + # new x-axis title
  ylab(NULL) +                   # remove y-axis title
  geom_vline(xintercept = 0) +   # add reference line at 0
  theme_minimal() +              # use different theme
  scale_y_discrete(
    expand = c(.02, .02)) +      # control padding at top/bottom margin
  scale_x_continuous(
    expand = c(.02, .02),        # control padding at left/right margin
    limits = c(0, NA),           # control limits of x-axis
    minor_breaks = NULL) +       # omit minor grid lines
  annotate("text",               # add direct label "Head"
           x = 25, y = 20,       #   position
           label = "Head",       #   label
           size = 3.1,           #   font size
           color = "grey30") +   #   color
  annotate("text",               # add direct label "Modifier"
           x = 15, y = 25,       #   position
           label = "Modifier",   #   label
           size = 3.1,           #   font size
           color = "grey30") +   #   color
  theme(
    legend.position = "none")    # omit legend

The we draw the appended panel, which we assign to object p2. We have to be careful to use the same parameter settings as for p1 (relevant lines are marked by (!) in the code).

p2 <- d |> 
  ggplot(aes(
    y = reorder(Category,          # (!) Category on the y-axis
                percent_modifier), # (!) order categories by percent_modifier
    x = difference)) +             # percentage point difference on the x-axis
  geom_point(size = 1.1) +         # control size of plotting symbols
  scale_shape_manual(
    values = 19) +                 # control choice of plotting symbol
  xlab("Difference") +             # new x-axis title
  ylab(NULL) +                     # remove y-axis title
  geom_vline(xintercept = 0) +     # add reference line at 0
  theme_minimal() +                # use different theme
  scale_y_discrete(
    expand = c(.02, .02)) +        # (!) control padding at top/bottom margin
  scale_x_continuous(
    expand = c(.05, .05),          # control padding at left/right margin
    breaks = c(-20, -10, 0),       # control location of tick marks
    labels = c("\u221220", "\u221210", "0"), # tick mark labels: proper minus signs
    minor_breaks = NULL) +                   # omit minor grid lines
  theme(axis.text.y = element_blank())       # omit tick mark labels on y-axis

Then we combine and print the two panels:

cowplot::plot_grid(
  p1, NULL, p2, 
  nrow = 1, 
  rel_widths = c(2,.05,1.2))

References

Amit, Ohad, Richard M. Heiberger, and Peter W. Lane. 2008. “Graphical Approaches to the Analysis of Safety Data from Clinical Trials.” Pharmaceutical Statistics 7 (1): 20–35. https://doi.org/10.1002/pst.254.

Cleveland, William S. 1984. “Graphical Methods for Data Presentation: Full Scale Breaks, Dot Charts, and Multibased Logging.” The American Statistician 38 (4): 270–80. https://doi.org/10.1080/00031305.1984.10483224.

Heiberger, Richard M., and Burt Holland. 2015. Statistical Analysis and Data Display: An Intermediate Course with Examples in r. New York: Springer.

Maguire, Phil, Edward J. Wisniewski, and Gert Storms. 2010. “A Corpus Study of Semantic Patterns in Compounding.” Corpus Linguistics and Linguistic Theory 6 (1). https://doi.org/10.1515/cllt.2010.003.

Sönning, Lukas. 2016. “The Dot Plot: A Graphical Tool for Data Analysis and Presentation.” In A Blend of MaLT: Selected Contributions from the Methods and Linguistic Theories Symposium, edited by Hanna Christ, Daniel Klenovšak, Lukas Sönning, and Valentin Werner, 101–29. Bamberg: University of Bamberg Press. https://doi.org/10.20378/irbo-51101.

Wilke, Claus O. 2024. Cowplot: Streamlined Plot Theme and Plot Annotations for ’Ggplot2’. https://doi.org/10.32614/CRAN.package.cowplot.