Beating lollipops into dumbbells

Shortly after I added lollipop charts to ggalt I had a few requests for a dumbbell geom. It wasn’t difficult to do modify the underlying lollipop Geoms to make a geom_dumbbell(). Here it is in action:

library(ggalt) # devtools::install_github("hrbrmstr/ggalt")

# from:
URL <- ""
fil <- basename(URL)
if (!file.exists(fil)) download.file(URL, fil)

df <- read.csv(fil, stringsAsFactors=FALSE)
df <- arrange(df, desc(Men))
df <- mutate(df, School=factor(School, levels=rev(School)))

gg <- ggplot(df, aes(x=Women, xend=Men, y=School))
gg <- gg + geom_dumbbell(colour="#686868",
gg <- gg + scale_x_continuous(breaks=seq(60, 160, by=20),
                              labels=sprintf("$%sK", comma(seq(60, 160, by=20))))
gg <- gg + labs(x="Annual Salary", y=NULL,
                title="Gender Earnings Disparity",
                caption="Data from plotly")
gg <- gg + theme_bw()
gg <- gg + theme(axis.ticks=element_blank())
gg <- gg + theme(panel.grid.minor=element_blank())
gg <- gg + theme(panel.border=element_blank())
gg <- gg + theme(axis.title.x=element_text(hjust=1, face="italic", margin=margin(t=-24)))
gg <- gg + theme(plot.caption=element_text(size=8, margin=margin(t=24)))


The API isn't locked in, so definitely file an issue if you want different or additional functionality. One issue I personally still have is how to identify the left/right points (blue is male and pink is female in this one).

Working Out With Dumbbells

I thought folks might like to see behind the ggcurtain. It really only took the addition of two functions to ggalt: geom_dumbbell() (which you call directly) and GeomDumbbell() which acts behind the scenes.

There are a few additional, custom parameters to geom_dumbbell() and the mapped stat and position are hardcoded in the layer call. We also pass in these new parameters into the params list.

geom_dumbbell <- function(mapping = NULL, data = NULL, ...,
                          point.colour.l = NULL, point.size.l = NULL,
                          point.colour.r = NULL, point.size.r = NULL,
                          na.rm = FALSE, show.legend = NA, inherit.aes = TRUE) {

    data = data,
    mapping = mapping,
    stat = "identity",
    geom = GeomDumbbell,
    position = "identity",
    show.legend = show.legend,
    inherit.aes = inherit.aes,
    params = list(
      na.rm = na.rm,
      point.colour.l = point.colour.l,
      point.size.l = point.size.l,
      point.colour.r = point.colour.r,
      point.size.r = point.size.r,

The exposed function eventually calls it's paired Geom. There we get to tell it what are required aes parameters and which ones aren't required, plus set some defaults.

We automagically add yend to the data in setup_data() (which gets called by the ggplot2 API).

Then, in draw_group() we create additional data.frames and return a list of three Geom layers (two points and one segment). Finally, we provide a default legend symbol.

GeomDumbbell <- ggproto("GeomDumbbell", Geom,
  required_aes = c("x", "xend", "y"),
  non_missing_aes = c("size", "shape",
                      "point.colour.l", "point.size.l",
                      "point.colour.r", "point.size.r"),
  default_aes = aes(
    shape = 19, colour = "black", size = 0.5, fill = NA,
    alpha = NA, stroke = 0.5

  setup_data = function(data, params) {
    transform(data, yend = y)

  draw_group = function(data, panel_scales, coord,
                        point.colour.l = NULL, point.size.l = NULL,
                        point.colour.r = NULL, point.size.r = NULL) {

    points.l <- data
    points.l$colour <- point.colour.l %||% data$colour
    points.l$size <- point.size.l %||% (data$size * 2.5)

    points.r <- data
    points.r$x <- points.r$xend
    points.r$colour <- point.colour.r %||% data$colour
    points.r$size <- point.size.r %||% (data$size * 2.5)

      ggplot2::GeomSegment$draw_panel(data, panel_scales, coord),
      ggplot2::GeomPoint$draw_panel(points.l, panel_scales, coord),
      ggplot2::GeomPoint$draw_panel(points.r, panel_scales, coord)


  draw_key = draw_key_point

In essence, this new geom saves calls to three additional geom_s, but does add more parameters, so it's not really clear if it saves much typing.

If you end up making anything interesting with geom_dumbbell() I encourage you to drop a note in the comments with a link.

Cover image from Data-Driven Security
Amazon Author Page

10 Comments Beating lollipops into dumbbells

  1. Pingback: Beating lollipops into dumbbells – Mubashir Qasim

  2. Richard

    The problem I see with this kind of lollipop charts is, that I cannot clearly grasp the endpoint. Is it the middle of the lollipop or is it the outer border? It may have some aesthetical appeal but a horizontal box plot without whiskers (or only whiskers with a vertical bar at the end |———————–| ) would be more readable.

      1. Winfried

        I need to think about it. It triggers more questions than answers. Clearly the unit you want to use depends on both the units used on the x-axis and the minimum and maximum difference (but what with differences close to zero).
        How would a dumbbell chart look with a logarithmic scale on the x-axis? Would it be interpretable?

  3. Pingback: Dumbbells get a little gg-smarter? – Mubashir Qasim

  4. Bruno

    I was wondering if is possible to change the shape of starting and ending in a dumbell graph.
    This is because I want to highlight the variation across time so i think that an arrow could be more useful, for example:
    1) a value has risen from 1 to 10 I would see something like ()=======>
    2) a value has fallen from 10 to 1 I would see something like <======()
    If is not possible probably is better to keep the ending point and skipping the beginning like

    1) ========()
    2) ()========

    Do you think something like that is possible?

    Thanks in advance ;)


    1. hrbrmstr

      Thx for both using the pkg and taking the time to comment! Aye, I agree that this is a great idea and shld have time after I’m back from holiday to get it into the next rev of {ggalt}’s geom_dumbbell().

  5. rafabelokurows

    While the actual uses for this plot should be actually well thought-out to not confuse the user, it looks great and I’m glad I found this package/function out so many years after this post. Thanks! :)

    Obs: I see the parameters are a bit different now, I guess the package have evolved a little bit, so whoever uses this nowadays, beware of the difference in how you call the function.


Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.