Dumbbell Chart with a Gap Column

This post explains how to build a custom dumbbell chart with a gap column on the right hand side with R and ggplot2. Step by step code snippets with explanations are provided. Thanks a lot to Fred Duong for this contribution!

Lollipop section Data to Viz


This tutorial by Fred Duong will show you how to create dumbbell charts with gap columns, similar to the charts you might have seen from Pew and YouGov.

The original dumbbell chart built by the Pew Research Center is shown below:

We can get pretty close to the chart with ggplot2() (and with a lot more tweaking that we won’t go into here, we could probably exactly recreate the chart!)

Here is the final figure that we’ll create.

But first, let’s create the basic figure!


Let’s use the tidyverse that includes ggplot2 and `patchwork`` to combine plots.

library(patchwork) #to combine plots

Data Creation

First, we create a tibble that contains the data from the figure.

    "Spirituality, faith and religion",
    "Freedom and independence",
    "Hobbies and recreation",
    "Physical and mental health",
    "Nature and the outdoors"),


Basic Figure

From this tibble, it’s easy to create the base figure. Though you can keep the data in wide format, I find it simpler to wrangle the data into long format first thanks to the pivot_longer function.

df_long=df %>% 

Now let’s build the plot:

df_long %>% 
  ggplot(aes(x=value,y=labels)) +
  geom_line(aes(group=labels), color="#E7E7E7", linewidth=3.5) + 
  # note that linewidth is a little larger than the point size 
  # so that the line matches the height of the point. why is it 
  # like that? i don't really know
  geom_point(aes(color=name), size=3) +
  theme_minimal() +
  theme(legend.position = "none",
        axis.text.y = element_text(color="black"),
        axis.text.x = element_text(color="#989898"),
        axis.title = element_blank(),
        panel.grid = element_blank()
        ) +
  scale_color_manual(values=c("#436685", "#BF2F24"))+
  scale_x_continuous(labels = scales::percent_format(scale = 1))

That’s it! Not a ton of code! But, you’ll notice that this is missing a few key formatting features from the original Pew figure:

  1. The y-axis is not sorted. By default, ggplot() alphabetizes character vectors that are used for the axes. Then, it displays the y-axis values reverse alphabetized. We can manually sort this however we want, but sorting by descending gaps is common. To do this, we will compute the gaps, and then turn the y-axis labels into a factor so that ggplot() respects the order.

  2. There are no text labels. Some dumbbell figures have data callouts to the left and right, and some have them underneath or above the points. When it’s to the left and right, it’s a little complicated to implement. There are two issues: 1) Because Democrats and Republicans might switch orders in terms of which is the higher value, you can’t rely just the categories themselves to decide which is left and right (Democrats won’t always be the lower value, for example); and 2) by default, positioning the data callouts will place them right on top of the points. But we will want to nudge the data callouts either to the left or to the right of the points. To solve this, we will create a variable that assesses the maximum value per label (or minimum) to assign the left and right callouts. Then, we can either use two geom_text()s, with each one using filter() on the data to only show the maximum or not the maximum and using the color aesthetic set to political party, or–more simply–use one geom_text() that uses an if_else() statement to assign that conditions on if the value == the maximum value. I’ll show the latter case.

  3. There is no color legend for the points. By default, ggplot will attempt to put a typical legend for your color aesthetic that is placed to the right. You’ll see above I turned this off in theme(). It can be much more readable to place the legend above the actual points themselves, as you see from the Pew figure.

  4. There is no gap column. The best way I can figure out how to do this is to create a second ggplot object, and use patchwork to stitch it together with the main plot.

The complete figure will solve all four issues, but this requires a lot more tweaking. The complete figure will consist of the main dumbbell figure and a gap figure.

Clean Dumbbell plot (left part)

A few additional data wrangling steps are necessary to complete the final figure:

df=raw %>% # raw is just the generated data
  # compute the gap
  mutate(gap=Rep-Dem) %>% 
  # find the maximum value by label
  group_by(labels) %>% 
  mutate(max=max(Dem, Rep)) %>% 
  ungroup() %>% 
  # sort the labels by gap value
  # note that its absolute value of gap
  mutate(labels=forcats::fct_reorder(labels, abs(gap)))  

# make into long format for easier plotting  
df_long=df %>% 

df_long %>% head()

The left hand side part of the final plot can then be built as follow:

# set a custom nudge value

df_long %>% 
  # the following 3 lines of code are the same
  ggplot(aes(x=value,y=labels)) +
  geom_line(aes(group=labels), color="#E7E7E7", linewidth=3.5) +
  geom_point(aes(color=name), size=3) +
  # but we want geom_text for the data callouts and the legend
  # data callout
  geom_text(aes(label=value, color=name),
              df_long$value==df_long$max, # if it's the larger value...
              nudge_value,   # move it to the right of the point
              -nudge_value), # otherwise, move it to the left of the point
              df_long$value==df_long$max, #if it's the larger value
              0, # left justify
              1),# otherwise, right justify      
  # legend
  geom_text(aes(label=name, color=name), 
            data=. %>% filter(gap==max(gap)),
            nudge_y =.5, 
  theme_minimal() +
  theme(legend.position = "none",
        axis.text.y = element_text(color="black"),
        axis.text.x = element_text(color="#989898"),
        axis.title = element_blank(),
        panel.grid = element_blank()
        ) +
  labs(x="%",y=NULL) +
  scale_color_manual(values=c("#436685", "#BF2F24")) +
  #extend the y-axis otherwise the legend is cut off
  coord_cartesian(ylim=c(1, 7.5)) +
  #display percentages with % appended
  scale_x_continuous(labels = scales::percent_format(scale = 1)) 


Gap section (right part)

The Gap figure on the right hand side must be created too. Pew formats the gap values in a particular way:

  1. Colors by party

  2. Adds “+” to the beginning of the gap value if positive

  3. Adds “R” or “D” to the end of the gap value

Let’s do that!

df %>%  # note i am using df and not df_long
    label=fct_reorder(labels, abs(gap)), #order label by descending gaps
    # we need a column for the party with the max value
    # format gap values
      paste0("+", abs(gap), gap_party_max) %>% 
      fct_inorder() #turns into factor to bake in the order


And then the plot:

  df_gap %>% 
  ggplot(aes(x=gap,y=labels)) +
  geom_text(aes(x=0, label=gap_label, color=gap_party_max),
            size=3.25) +
  geom_text(aes(x=0, y=7), # 7 because that's the # of y-axis values
            nudge_y =.5, # match the nudge value of the main plot legend    
            size=3.25) +
  theme_void() +
  coord_cartesian(xlim = c(-.05, 0.05), 
                  ylim=c(1,7.5) # needs to match main plot
    plot.margin = margin(l=0, r=0, b=0, t=0), #otherwise it adds too much space
    panel.background = element_rect(fill="#EFEFE3", color="#EFEFE3"),
    legend.position = "none"
  scale_color_manual(values=c("#436685", "#BF2F24"))


Combine figures

Use the package patchwork to put together your final figure. You’ll need to play around with plot_layout() a bit to get your figures to fit together how you want.

  # syntax from `patchwork`
  p_main + p_gap + plot_layout(design=
    area(l=0,  r=45, t=0, b=1), # defines the main figure area
    area(l=46, r=52, t=0, b=1)  # defines the gap figure area


The End

And there it is! To summarize: making the basic figure is quite easy, but adding all the little extras to make it more readable takes a lot more tweaking.

Some final notes:

Thank you for reading! I hope you had as much fun following along as I did making this! – Fred Duong

Related chart types

Spider / Radar
Circular Barplot


This document is a work by Yan Holtz. Any feedback is highly encouraged. You can fill an issue on Github, drop me a message on Twitter, or send an email pasting yan.holtz.data with gmail.com.

Github Twitter