通过构面分隔y轴标签或删除图例但保留空间

Question

Ok, I'm stumped on a home-brew ggplot.

What I would like to do is have a three row, one column faceted plot with a different y-axis label for each facet. The units of the y-axis are all the same. This would be the most convenient, but googling tells me it may not be possible.

Alternatively, I found this solution using grid.arrange, which seems like it will work. However, I want to keep a legend only for one plot and remove it from the other two, but maintain the spacing as if it were still there so that everything lines up nice. Someone had the same problem a few years ago, but the suggested solution is depreciated and I can't sort out how to make it work in modern ggplot.

Any help is appreciated! Using facets would be easiest!

Edited to add copy of plot after using user20560's gridArrange solution below. Very nearly there, just would like to get back the box around the top and bottom facet panels!

Answer 1

I have assumed (possibly wrongly) that you are wanting to add separate y-axis titles rather than axis labels. [If it is the labels you want different you can use the scales argument in facet_grid ]

There will be a ggplot way to do this but here are a couple of ways you could tweak the grobs yourself.

So using mtcars dataset as example

library(ggplot2)
library(grid)
library(gridExtra)

One way

p <- ggplot(mtcars, aes(mpg, wt, col=factor(vs))) +   geom_point() + 
                                 facet_grid(gear ~ .)

# change the y axis labels manually
g <- ggplotGrob(p)
yax <- which(g$layout$name=="ylab")

# define y-axis labels
g[["grobs"]][[yax]]$label <- c("aa","bb", "cc")

# position of labels (ive just manually specified)
g[["grobs"]][[yax]]$y <- grid::unit(seq(0.15, 0.85, length=3),"npc")

grid::grid.draw(g)

Or using grid.arrange

# Create a plot for each level of grouping variable and y-axis label
p1 <- ggplot(mtcars[mtcars$gear==3, ], aes(mpg, wt, col=factor(vs))) + 
                                 geom_point() + labs(y="aa") + theme_bw()
p2 <- ggplot(mtcars[mtcars$gear==4, ], aes(mpg, wt, col=factor(vs))) +  
                                 geom_point() + labs(y="bb") + theme_bw()
p3 <- ggplot(mtcars[mtcars$gear==5, ], aes(mpg, wt, col=factor(vs))) +  
                                 geom_point() + labs(y="cc") + theme_bw()

# remove legends from two of the plots
g1 <- ggplotGrob(p1)
g1[["grobs"]][[which(g1$layout$name=="guide-box")]][["grobs"]] <- NULL

g3 <- ggplotGrob(p3)
g3[["grobs"]][[which(g3$layout$name=="guide-box")]][["grobs"]] <- NULL

gridExtra::grid.arrange(g1,p2,g3)

If it is the axis titles you want to add I should ask why you want a different titles - can the facet strip text not do?

Answer 2

Following the comments by Axeman and aosmith (thank you), here's a way to do this using the facet labels using ggplot2 version 2.2.0

library(ggplot2) # From sessionInfo(): ggplot2_2.2.0

ggplot(mtcars, aes(mpg, wt, col=factor(vs))) + geom_point() + 
  facet_grid(gear ~ ., switch = 'y') +
  theme( axis.title.y = element_blank(), # remove the default y-axis title, "wt"
         strip.background = element_rect(fill = 'transparent'), # replace the strip backgrounds with transparent
         strip.placement = 'outside', # put the facet strips on the outside
         strip.text.y = element_text(angle=180)) # rotate the y-axis text (optional)
# (see ?ggplot2::theme for a list of theme elements (args to theme()))

Answer 3

I know this is an old post, but after finding it, I could not get @user20560's response to work.

I've edited @user20560's grid.extra approach as follows:

library(ggplot2)
library(gridExtra)
library(grid)

# Create a plot for each level of grouping variable and y-axis label
p1 <- ggplot(mtcars[mtcars$gear==3, ], aes(mpg, wt, col=factor(vs))) + 
  geom_point() + labs(y="aa") + theme_bw()
p2 <- ggplot(mtcars[mtcars$gear==4, ], aes(mpg, wt, col=factor(vs))) +  
  geom_point() + labs(y="bb") + theme_bw()
p3 <- ggplot(mtcars[mtcars$gear==5, ], aes(mpg, wt, col=factor(vs))) +  
  geom_point() + labs(y="cc") + theme_bw()

# get the legend as a grob
legend <- ggplotGrob(p1)
legend <- legend$grobs[[which(legend$layout$name=="guide-box")]]
lheight <- sum(legend$height)
lwidth <- sum(legend$width)

# remove the legend from all the plots
p1 <- p1 + theme(legend.position = 'none')
p2 <- p2 + theme(legend.position = 'none')
p3 <- p3 + theme(legend.position = 'none')

# force the layout to the right side
layoutMat <- matrix(c(1,2,3,4,4,4),ncol = 2)

grid.arrange(p1,p2,p3,legend, layout_matrix = layoutMat, ncol = 2, 
             widths = grid::unit.c(unit(1,'npc') - lwidth, lwidth))

This example is somewhat specific to this particular layout. There is a more general approach on the ggplot2 wiki .

Answer 4

I too had trouble getting the first approach in the answer of user20560 (above) to work. This is probably because the internals of ggplot2 have evolved, and there is no guarantee that these internals should stay the same. In any case, here is a version that currently works:

library(ggplot2) # From sessionInfo(): ggplot2_2.1.0
library(grid)

p <- ggplot(mtcars, aes(mpg, wt, col=factor(vs))) + geom_point() + facet_grid(gear ~ .)
g <- ggplotGrob(p)
yax <- which(g$layout$name == "ylab")
g[["grobs"]][[yax]]$children[[1]]$label <- c('fo','bar','foobar')
g[["grobs"]][[yax]]$children[[1]]$y <- grid::unit(seq(0.15, 0.85, length=3), "npc")
grid.draw(g)

Note that this is the approach that keeps the facets and does not repeat the x-axes.