Make a rectangular legend, with rows and columns labeled, in grid

Question

I've got a ggplot where I'm mapping factors to both fill and alpha, like this:

set.seed(47)
the_data <- data.frame(value = rpois(6, lambda=20),
                       cat1 = rep(c("A", "B"), each = 3),
                       cat2 = rep(c("X", "Y", "Z"), 2))

ggplot(the_data, aes(y = value, x = cat2, alpha = cat1, fill = cat2)) +
    geom_bar(stat = "identity", position = "dodge") +
    scale_alpha_discrete(range = c(0.5, 1)) +
    theme_bw()

The people I'm producing it for don't find the legend for alpha very clear. I think a good alternative would be something like this (which I hacked together in base graphics):

I know I can't generate a legend like that with high-level ggplot commands, but can I do it in grid and put it on top of my plot?

Answer 1

Here is one possible starting point. I create two different plots which have the appropriate legends - a 'bright' and a 'pale'. Extract the legends from the plot objects. Then use grid viewports, one for the plot, and one for each legend, to put the pieces together.

library(grid)
library(gtable)

# create plot with legend with alpha = 1
g1 <- ggplot(the_data, aes(y = value, x = cat2, alpha = cat1, fill = cat2)) +
  geom_bar(stat = "identity", position = "dodge") +
  scale_alpha_discrete(range = c(0.5, 1)) +
  theme_bw() +
  guides(fill = guide_legend(title = "A",
                             title.hjust = 0.4),
         alpha = FALSE) +
  theme_bw() +
  theme(legend.text = element_blank())

g1

# grab legend
legend_g1 <- gtable_filter(ggplot_gtable(ggplot_build(g1)), "guide-box") 


# create plot with 'pale' legend
g2 <- ggplot(the_data, aes(y = value, x = cat2, alpha = cat1, fill = cat2)) +
  geom_bar(stat = "identity", position = "dodge") +
  scale_alpha_discrete(range = c(0.5, 1)) +
  guides(fill = guide_legend(override.aes = list(alpha = 0.5),
                             title = "B",
                             title.hjust = 0.3),
         alpha = FALSE) +
  theme_bw()
g2

# grab legend
legend_g2 <- gtable_filter(ggplot_gtable(ggplot_build(g2)), "guide-box") 



# arrange plot and legends

# legends to the right

# define plotting regions (viewports)
vp_plot <- viewport(x = 0.4, y = 0.5,
                    width = 0.8, height = 1)

vp_legend_g1 <- viewport(x = 0.85, y = 0.5,
                           width = 0.4, height = 0.4)

vp_legend_g2 <- viewport(x = 0.90, y = 0.5,
                           width = 0.4, height = 0.4)


# clear current device
grid.newpage()

# add objects to the viewports
# plot without legend
print(g1 + theme(legend.position = "none"), vp = vp_plot)
upViewport(0)

pushViewport(vp_legend_g1)
grid.draw(legend_g1)
upViewport(0)

pushViewport(vp_legend_g2)
grid.draw(legend_g2) 

# legends on top
vp_plot <- viewport(x = 0.5, y = 0.4,
                    width = 1, height = 0.85)

vp_legend_g1 <- viewport(x = 0.5, y = 0.9,
                         width = 0.4, height = 0.4)

vp_legend_g2 <- viewport(x = 0.55, y = 0.9,
                         width = 0.4, height = 0.4)

grid.newpage()

print(g1 + theme(legend.position = "none"), vp = vp_plot)
upViewport(0)

pushViewport(vp_legend_g1)
grid.draw(legend_g1)
upViewport(0)

pushViewport(vp_legend_g2)
grid.draw(legend_g2)  

Answer 2

@Henrik

This might be a little easier,

g1 <- ggplotGrob(p1)
g2 <- ggplotGrob(p2)

leg1 <- gtable_filter(g1, "guide-box") 
leg2 <- gtable_filter(g2, "guide-box") 
leg <- gtable:::cbind_gtable(leg1[["grobs"]][[1]],  leg2[["grobs"]][[1]], "first")

g1$grobs[g1$layout$name == "guide-box"][[1]] <- leg
g1$widths[max(subset(g1$layout, name == "guide-box")[["r"]])] <- list(leg1$width + leg2$width)

grid.newpage()
grid.draw(g1)