Moon charts, a pie chart alternative for two groups in ggplot2
gggibbous can be installed from CRAN:
install.packages("gggibbous")The development version can be installed from GitHub with the devtools
package:
devtools::install_github("mnbram/gggibbous")A pie chart divides a circle into multiple sections where the arc lengths (and so also the areas) of the sections represent proportions of a whole. A moon chart, similarly, divides a circle into sections where the areas represent proportions of a whole, but in a moon chart the areas are drawn as crescent or gibbous portions of a circle—like the phases of the moon.
The motivation behind using a moon chart instead of a pie chart is primarily one of aesthetic choice. Note also that because the sections of a moon chart are swept from one or the other side of the circle, they are generally only appropriate for depicting one or two groups.
Moon charts are similar to Kosara’s (2019)[1] “circular slice” chart. In studying subjects’ perception of percentages in different chart types, the “circular slice” performed similarly to pie charts. Moon charts differ from the “circular slice” in that the latter slides a second disc of the same size over a base circle, more like a lunar eclipse than the sphases of the moon. Both depend on area as the perceptual cue, however.
gggibbous extends the ggplot2 data visualization package to provide
support for moon charts in R. Unlike the pie charts supported natively
by coord_polar() in R, moon charts in gggibbous do not require any
special coordinate system. They are drawn most similarly to points in
ggplot2: their position is defined by an x and a y coordinate and
their size is defined independently of the coordinate system, so they
always remain circular.
library(gggibbous)
#> Loading required package: ggplot2ggplot(data.frame(x = 1:5, y = 1, size = 2^(0:4)), aes(x, y, size = size)) +
geom_moon() +
geom_point(y = 2) +
lims(x = c(0.5, 5.5), y = c(0.5, 2.5)) +
scale_size(range = c(5, 10))
Two new aesthetics are also important in geom_moon: ratio and
right.
ratio controls the proportion of the moon to be drawn. It must be
between 0 (a “new moon” where nothing is actually drawn) and 1 (a “full
moon”, i.e. a circle).
ggplot(data.frame(x = 1:5, y = 0, ratio = 0:4 * 0.25), aes(x = x, y = y)) +
geom_moon(aes(ratio = ratio), size = 20, fill = "black") +
geom_text(aes(y = y + 1, label = ratio)) +
lims(x = c(0.5, 5.5), y = c(-1, 1.4)) +
theme_void()
right takes a boolean value that controls whether the moon is “waxing”
or “waning”—that is, whether it is “filled” from the right or the left.
One way to make a “complete” moon with two colors is to use right = TRUE for one color and right = FALSE for the other, with
complementary ratios. (See the examples below for some other
approaches.)
tidymoons <- data.frame(
x = rep(1:3, 6),
y = rep(rep(3:1, each = 3), 2),
ratio = c(1:9 / 10, 9:1 / 10),
right = rep(c(TRUE, FALSE), each = 9)
)
ggplot(tidymoons) +
geom_moon(aes(x, y, ratio = ratio, right = right, fill = right)) +
lims(x = c(0.5, 3.5), y = c(0.5, 3.5))
gggibbous includes three key glyphs for different types of legends:
draw_key_moon, draw_key_moon_left, and draw_key_full_moon
draw_key_moon—the default in geom_moon—draws a gibbous moon with
right = TRUE (see above).
draw_key_moon_left draws a crescent moon from the left that is
complementary to the gibbous moon in draw_key_moon, which is useful
for combined legends:
ggplot(tidymoons, aes(x, y, ratio = ratio, right = right, size = 2^x)) +
geom_moon(data = subset(tidymoons, right), fill = "violetred") +
geom_moon(
data = subset(tidymoons, !right), fill = "turquoise3",
key_glyph = draw_key_moon_left
) +
lims(x = c(0.5, 3.5), y = c(0.5, 3.5)) +
scale_size("size", range = c(5, 10), breaks = 2^(1:3))
draw_key_full_moon draws a circle. It is similar to the “point”
key glyph, but the size is calculated slightly differently, so it is
more appropriate if you want the size of legend moons and the size of
the plot moons to match.
ggplot(tidymoons) +
geom_moon(
aes(x, y, ratio = ratio, right = right, fill = right, size = 2^x),
key_glyph = draw_key_full_moon
) +
lims(x = c(0.5, 3.5), y = c(0.5, 3.5)) +
scale_size("size", range = c(5, 10), breaks = 2^(1:3)) +
scale_fill_manual(values = c("firebrick1", "dodgerblue2")) +
theme(legend.box = "horizontal")
One common use of multiple pie charts is to represent proportions at different coordinates on a map. The x and y dimensions are already committed to the map coordinates, so proportional visualizations like bar charts are more difficult. This is a perfect opportunity to try out moon charts!
Pie chart maps are popular in population genetics, so let’s look at an
example from that field. The dmeladh data[2] contains frequencies of
two variants of the Adh gene in Australian and Papua New Guinean fruit
fly populations. Many of these populations are close together, so we
have to deal with overplotting, which we do manually below.
dmeladh_adj <- dmeladh
dmeladh_adj$long <- dmeladh$Longitude + c(
-2, 0, -2, 2, -3, 3, 3, 2, 3, 4, -2.5, -2.5, -1, -2, -2.5, -4, 2.5,
5, 6, 7, 2, -7, -5.5, -3, 0, -7, -2, 3, 5.5, 0.5, -1, -1.5, -3, 2)
dmeladh_adj$lat <- dmeladh$Latitude + c(
-2, 2, 0, 1, 0, 0, 0, 2, 0.5, -1, 1, -1.5, 2, 4, 1.5, 0, 2,
1, -1, -3, -2, 1, -1, -2, -3, -2, -4, -3, -1, 1.5, 2, 2, -2, 0)
moonmap <- ggplot(dmeladh_adj, aes(long, lat)) +
geom_polygon(
data = map_data(
"world", region = "(Australia)|(Indonesia)|(Papua New Guinea)"),
aes(group = group),
fill = "gray80"
) +
geom_segment(aes(xend = Longitude, yend = Latitude), color = "gray20") +
geom_point(aes(Longitude, Latitude), size = 0.75, color = "gray20") +
scale_size(range = c(4, 10)) +
coord_map(xlim = c(110, 160), ylim = c(-45, -5)) +
theme_void() +
theme(
legend.position = c(0.05, 0.05),
legend.direction = "horizontal",
legend.justification = c(0, 0)
)
moonmap +
geom_moon(
aes(ratio = AdhS / 100, size = N),
right = FALSE, fill = "gold", color = "gold",
key_glyph = draw_key_moon_left
) +
geom_moon(
aes(ratio = AdhF / 100, size = N),
fill = "forestgreen", color = "forestgreen"
)
If we want to label the alleles in the legend explicitly, then we need to map them to a group, which requires that we rearrange the data into a “longer” (“tidy”) format.
tidyadh <- reshape(
dmeladh_adj,
varying = c("AdhF", "AdhS"),
v.names = "percent",
timevar = "allele",
times = c("AdhF", "AdhS"),
idvar = c("Locality", "Latitude", "Longitude", "long", "lat", "N"),
direction = "long"
)
tidyadh$right <- rep(c(TRUE, FALSE), each = nrow(dmeladh_adj))
moonmap +
geom_moon(
data = tidyadh, key_glyph = draw_key_full_moon,
aes(ratio = percent / 100, fill = allele, color = allele, right = right,
size = N)
) +
scale_fill_manual(values = c("forestgreen", "gold")) +
scale_color_manual(values = c("forestgreen", "gold"))
Sometimes you just want to plot literal representations of the moon. The
lunardist data, adapted from NASA, contains the distance from the
Earth to the Moon for each day in 2019, as well as the dates (in UTC) of
each occurrence of the four principal phases of the moon. We can plot
those principal phases using moon charts (which in this case are
identical to pie charts).
moonphase <- subset(lunardist, !is.na(phase))
moonphase$percent <- ifelse(
moonphase$phase == "new", 0, ifelse(moonphase$phase == "full", 1, 0.5))
ggplot(lunardist, aes(date, distance)) +
geom_line() +
# Plotting the lower layer as a full circle also works in most cases
geom_moon(data = moonphase, ratio = 1, size = 5, fill = "black") +
geom_moon(
data = moonphase, aes(ratio = percent),
size = 5, fill = "yellow", right = moonphase$phase == "first quarter"
)
“Harvey balls” are essentially pie charts used for qualitative comparisons, often in tabular format. We can use moon charts for the same purpose.
First, let’s make up some data:
rest_names <- c(
"Anscombe's Luncheonette", "Chai Squared", "Tukey's Honest Southern Diner",
"Bagels ANOVA", "Spearmint Row"
)
restaurants <- data.frame(
Restaurant = factor(rest_names, levels = rest_names),
Food = c(5, 3, 4, 4, 1),
Decor = c(2, 5, 3, 1, 5),
Service = c(4, 2, 3, 3, 5),
Price = c(4, 5, 2, 5, 2)
)As a regular table:
| Restaurant | Food | Decor | Service | Price |
|---|---|---|---|---|
| Anscombe’s Luncheonette | 5 | 2 | 4 | 4 |
| Chai Squared | 3 | 5 | 2 | 5 |
| Tukey’s Honest Southern Diner | 4 | 3 | 3 | 2 |
| Bagels ANOVA | 4 | 1 | 3 | 5 |
| Spearmint Row | 1 | 5 | 5 | 2 |
And now as a table with Harvey moons:
# First we reshape the data into "long" format to facilitate plotting
rest_cats <- c("Food", "Decor", "Service", "Price")
tidyrest <- reshape(
restaurants,
varying = rest_cats,
v.names = "Score",
timevar = "Category",
times = factor(rest_cats, levels = rest_cats),
idvar = "Restaurant",
direction = "long"
)
ggplot(tidyrest, aes(0, 0)) +
geom_moon(aes(ratio = (Score - 1) / 4), fill = "black") +
geom_moon(aes(ratio = 1 - (Score - 1) / 4), right = FALSE) +
facet_grid(Restaurant ~ Category, switch = "y") +
theme_minimal() +
theme(
panel.grid = element_blank(),
strip.text.y.left = element_text(angle = 0, hjust = 1),
axis.text = element_blank(),
axis.title = element_blank()
)
-
Kosara, R. 2019. Circular Part-to-Whole Charts Using the Area Visual Cue. EuroVis 2019 - Short Papers. https://doi.org/10.2312/evs.20191163
-
Oakeshott, J.G., et al. 1982. Alcohol dehydrogenase and glycerol-3-phosphate dehydrogenase clines in Drosophila melanogaster on different continents. Evolution, 36(1): 86-96.