How to add y axis title for each facet row in ggplot?

Question

I am doing a scatterplot with a facet_grid() like that:

library(ggplot2)
ggplot(df, aes(x, y)) +
  geom_point() +
  facet_grid(group1 ~ group2)

I want the y axis title y to be in the middle of each row like this (paint solution):

The numbers of facet rows is two in this example because df$group2 has two different values. For my actual use case there may be more than two rows depending on the used facet variable; the y axis title is supposed to be in the middle of each facet row.

Best solution so far is adding spaces which is a mess since using y axis titles of different length shifts the text away from the middle of the rows. It must be with ggplot2, i.e. without the usage of additional packages. I make a package and do not want to rely on/ include too many packages.

Data used here:

df <- data.frame(x= rnorm(100), y= rnorm(100),
                 group1= rep(0:1, 50), group2= rep(2:3, each= 50))

Answer 1

Here is a version with annotation, using ggplot2 only. It should be scalable.

No messing with grobs. The disadvantage is that the x positioning and the plot margins need to be semi-manually defined and this might not be very robust.

library(ggplot2)

df <- data.frame(x= rnorm(100), y= rnorm(100),
                 group1= rep(0:1, 50), group2= rep(2:3, each= 50))

## define a new data frame based on your groups, so this is scalable
annotate_ylab <- function(df, x, y, group1, group2, label = "label") {
  ## make group2 a factor, so you know which column will be to the left
  df[[group2]] <- factor(df[[group2]])
  lab_df <- data.frame( 
    ## x positioning is a bit tricky,
    ## I think a moderately robust method is to
    ## set it relativ to the range of your values
    x = min(df[[x]]) - 0.2 * diff(range(df[[x]])),
    y = mean(df[[y]]),
    g1 = unique(df[[group1]]),
    ## draw only on the left column
    g2 = levels(df[[group2]])[1],
    label = label
  )
  names(lab_df) <- c(x, y, group1, group2, "label")
  lab_df
}

y_df <- annotate_ylab(df, "x", "y", "group1", "group2", "y")

ggplot(df, aes(x, y)) +
  geom_point() +
  geom_text(data = y_df, aes(x, y, label = label), angle = 90) +
  facet_grid(group1 ~ group2) +
  coord_cartesian(xlim = range(df$x), clip = "off") +
  theme(axis.title.y = element_blank(), 
        plot.margin = margin(5, 5, 5, 20))

y_df_mtcars <- annotate_ylab(mtcars, "mpg", "disp", "carb", "vs", "y")

ggplot(mtcars, aes(mpg, disp)) +
  geom_point() +
  geom_text(data = y_df_mtcars, aes(mpg, disp, label = label), angle = 90) +
  facet_grid(carb ~ vs) +
  coord_cartesian(xlim = range(mtcars$mpg), clip = "off") +
  theme(axis.title.y = element_blank(), 
        plot.margin = margin(5, 5, 5, 20))

^{Created on 2021-11-24 by the reprex package (v2.0.1)}

Answer 2

You can copy the axis labels into new grobs in the gtable. Note that although this uses the grid and gtable packages, these are already imported by ggplot2, so this does not add any new dependencies that are not already available and used internally by ggplot.

library(grid)
library(gtable)

g = ggplot(df, aes(x, y)) +
  geom_point() +
  facet_grid(group1 ~ group2)

gt = ggplot_gtable(ggplot_build(g))
which.ylab = grep('ylab-l', gt$layout$name)
gt = gtable_add_grob(gt, gt$grobs[which.ylab], 8, 3)
gt = gtable_add_grob(gt, gt$grobs[which.ylab], 10, 3)
gt = gtable_filter(gt, 'ylab-l', invert = TRUE) # remove the original axis title
grid.draw(gt)

The above works for OP's example with just two facets. If we want to generalise this for an arbitrary number of facets we can do that simply enough by searching the gtable to see which rows contain y-axes.

gt = ggplot_gtable(ggplot_build(g))
which.ylab = grep('ylab-l', gt$layout$name)
which.axes = grep('axis-l', gt$layout$name)
axis.rows  = gt$layout$t[which.axes]
label.col  = gt$layout$l[which.ylab]
gt = gtable::gtable_add_grob(gt, rep(gt$grobs[which.ylab], length(axis.rows)), axis.rows, label.col)
gt = gtable::gtable_filter  (gt, 'ylab-l', invert = TRUE) 
grid::grid.draw(gt)

In the version above, I also use :: to explicitly specify the namespace for the functions from the grid and gtable packages. This will allow the code to work without even loading the additional packages into the search path.

Demonstrating this code with another example with four facet rows:

df <- data.frame(x= rnorm(100), y= rnorm(100),
                 group1= rep(1:4, 25), group2= rep(1:2, each= 50))

Answer 3

Without using another package, I felt that the best method would be to build upon the spaces solution you linked in the original question. So I wrote a function to make the label spacing a little bit more robust.

ylabel <- function(label1,label2){
  L1 <- nchar(label1)
  L2 <- nchar(label2)
  scaler <- ifelse(L1 + L2 > 8, 4, 0)
  space1 = paste0(rep("",27 - (L1/2)),collapse = " ")
  space2 = paste0(rep("",44 - (L1/2 + L2/2) - scaler), collapse = " ")
  space3 = paste0(rep("",22 - (L2/2)), collapse = " ")
  paste0(space1,label1,space2,label2,space3)
}

Application:

test <- ylabel("automobiles", "trucks")
ggplot(df, aes(x, y)) +
  geom_point() +
  facet_grid(group1 ~ group2) +
  ylab(test)

Still playing around with the scaler parameter, it's not perfect:

test2 <- ylabel("super long label", "a")
ggplot(df, aes(x, y)) +
  geom_point() +
  facet_grid(group1 ~ group2) +
  ylab(test2)

Will continue to refine the function/parameters, but am thinking this will get you close to what you're looking for.

Answer 4

You may consider switching to library(cowplot) for more control

The following code could be added to a function, but I left it long for clarity. Create 4 dataframes and feed them to four plots. Then arrange the plots

library(tidyverse)
df <- data.frame(x= rnorm(100), y= rnorm(100),
                 group1= rep(0:1, 50), group2= rep(2:3, each= 50))


library(cowplot)
df1 <- df %>% 
  filter(group2 == 2) %>% 
         filter(group1 == 0)

df2 <- df %>% 
  filter(group2 == 3) %>% 
  filter(group1 == 0)

df3 <- df %>% 
  filter(group2 == 2) %>% 
  filter(group1 == 1)

df4 <- df %>% 
  filter(group2 == 3) %>% 
  filter(group1 == 1)

plot1 <- ggplot(df1, aes(x, y)) +
  geom_point() +
  facet_grid(group1 ~ group2)+
  xlim(c(-3, 3))+
  ylim(c(-3, 2))+
  theme(strip.text.y = element_blank(), 
        axis.title.x = element_blank(), 
        axis.text.x = element_blank(), 
        axis.ticks.x = element_blank()
        )
plot1


plot2 <- ggplot(df2, aes(x, y)) +
  geom_point() +
  facet_grid(group1 ~ group2)+
  xlim(c(-3, 3))+
  ylim(c(-3, 2))+
  theme(axis.title.y = element_blank(), 
        axis.text.y = element_blank(), 
        axis.ticks.y = element_blank(), 
        axis.title.x = element_blank(), 
        axis.text.x = element_blank(), 
        axis.ticks.x = element_blank()
        )
plot2


plot3 <- ggplot(df3, aes(x, y)) +
  geom_point() +
  facet_grid(group1 ~ group2)+
  xlim(c(-3, 3))+
  ylim(c(-3, 2))+
  theme(strip.text.x = element_blank(),
        strip.text.y = element_blank())
plot3


plot4 <- ggplot(df4, aes(x, y)) +
  geom_point() +
  facet_grid(group1 ~ group2)+
  xlim(c(-3, 3))+
  ylim(c(-3, 2))+
  theme(axis.title.y = element_blank(), 
        strip.text.x = element_blank(),
        axis.text.y = element_blank(), 
        axis.ticks.y = element_blank())
plot4

plot_grid(plot1, plot2, plot3, plot4)