Nested legend based on colour and shape - r

Question

I want to make an xy plot of nested groups (Group and Subgroup) where points are colored by Group and have shape by Subgroup. A minimal example is below:
DATA<-data.frame(
Group=c(rep("group1",10),rep("group2",10),rep("group3",10) ),
Subgroup = c(rep(c("1.1","1.2"),5), rep(c("2.1","2.2"),5), rep(c("3.1","3.2"),5)),
x=c(rnorm(10, mean=5),rnorm(10, mean=10),rnorm(10, mean=15)),
y=c(rnorm(10, mean=3),rnorm(10, mean=4),rnorm(10, mean=5))
)
ggplot(DATA, aes(x=x, y=y,colour=Group, shape=Subgroup) ) +
geom_point(size=3)
However, because in reality I have many more subgroups than can be easily be identified based on the available shapes I want to repeat the same shapes within each Group. Below is the same code but with an additional column (Shape) specifying the shape:
DATA<-data.frame(
Group=c(rep("group1",10),rep("group2",10),rep("group3",10) ),
Subgroup = c(rep(c("1.1","1.2"),5), rep(c("2.1","2.2"),5), rep(c("3.1","3.2"),5)),
Shape = as.character(c(rep(c(1,2),15) ) ),
x=c(rnorm(10, mean=5),rnorm(10, mean=10),rnorm(10, mean=15)),
y=c(rnorm(10, mean=3),rnorm(10, mean=4),rnorm(10, mean=5))
)
ggplot(DATA, aes(x=x, y=y,colour=Group, shape=Shape) ) +
geom_point(size=3)
Now the shapes and colours are as I want them. However, the legend no longer lists the subgroups. What I want is a legend that lists all subgroups under each respective Group. Something like:
Group1
1.1
1.2
Group2
2.1
2.2
Group3
3.1
3.2
(Ideally, this would be a single nested legend. If nested legends are not possible, perhaps they can be three separate legends with the Groups as titles)
Is this something that can be achieved, and how?
Thanks

Answer 1

One option to achieve your desired result would be via the ggnewscale package which allows for multiple scales and legends for the same aesthetic.
To this end we have to
split the data by GROUP and plot each GROUP via a separate geom_point layer.
Additionally each GROUP gets a separate shape scale and legend which via achieve via ggnewscale::new_scale.
Instead of making use of the color aesthetic we set the color for each group as an argument for which I make use of a named vector of colors
Instead of copying and pasting the code for each group I make use of purrr::imap to loop over the splitted dataset and add the layers dynamically.
One more note: In general the order of legends is by default set via a "magic algorithm". To get the groups in the right order we have to explicitly set the order via guide_legend.
library(ggplot2)
library(ggnewscale)
library(dplyr)
library(purrr)
library(tibble)
DATA_split <- split(DATA, DATA$Group)
# Vector of colors and shapes
colors <- setNames(scales::hue_pal()(length(DATA_split)), names(DATA_split))
shapes <- setNames(scales::shape_pal()(length(unique(DATA$Shape))), unique(DATA$Shape))
ggplot(mapping = aes(x = x, y = y)) +
purrr::imap(DATA_split, function(x, y) {
# Get Labels
labels <- x[c("Shape", "Subgroup")] %>%
distinct(Shape, Subgroup) %>%
deframe()
# Get order
order <- as.numeric(gsub("^.*?(\\d+)$", "\\1", y))
list(
geom_point(data = x, aes(shape = Shape), color = colors[[y]], size = 3),
scale_shape_manual(values = shapes, labels = labels, name = y, guide = guide_legend(order = order)),
new_scale("shape")
)
})
DATA
set.seed(123)
DATA <- data.frame(
Group = c(rep("group1", 10), rep("group2", 10), rep("group3", 10)),
Subgroup = c(rep(c("1.1", "1.2"), 5), rep(c("2.1", "2.2"), 5), rep(c("3.1", "3.2"), 5)),
Shape = as.character(c(rep(c(1, 2), 15))),
x = c(rnorm(10, mean = 5), rnorm(10, mean = 10), rnorm(10, mean = 15)),
y = c(rnorm(10, mean = 3), rnorm(10, mean = 4), rnorm(10, mean = 5))
)