how to replace legend 'bullet' of geom_text guide (legend)

Question

I would like to replace the 'bullets' in legend (guide) of geom_text. Now it's a tilted a, but I would like a big fat circle or a square or any other shape that will emphasize the color (more).

library(ggplot2)

majdf <- data.frame(lvl = rep(c("A", "B"), each = 50), val = c(rnorm(50, 1), rnorm(50, 3)))
majtxt <- data.frame(species = c("sp1", "sp2", "sp3"), geq = c(0.01, 2, 2.2))

ggplot(majdf, aes(x = val)) +
  geom_density() +
  geom_vline(data = majtxt, aes(xintercept = geq)) +
  geom_text(data = majtxt, aes(x = geq, y = 0.2, label = geq, color = species), angle = 90) +
  facet_wrap(~ lvl)

Answer 1

This is just a hack.

Create the plot with geom_point:

p1 <- ggplot(majdf, aes(x = val)) +
  geom_density() +
  geom_vline(data = majtxt, aes(xintercept = geq)) +
  geom_point(data = majtxt, aes(x = geq, y = 0.2, label = geq, color = species), angle = 90) +
  facet_wrap(~ lvl)

The same with geom_text:

p2 <- ggplot(majdf, aes(x = val)) +
  geom_density() +
  geom_vline(data = majtxt, aes(xintercept = geq)) +
  geom_text(data = majtxt, aes(x = geq, y = 0.2, label = geq, color = species), angle = 90) +
  facet_wrap(~ lvl)

Turn into a grob and find which element is the guide:

g1 <- ggplotGrob(p1)
g1

TableGrob (8 x 10) "layout": 13 grobs
    z         cells       name                                     grob
1   0 ( 1- 8, 1-10) background          rect[plot.background.rect.2339]
2   1 ( 4- 4, 4- 4)    panel-1                gTree[panel-1.gTree.2263]
3   2 ( 4- 4, 7- 7)    panel-2                gTree[panel-2.gTree.2278]
4   3 ( 3- 3, 4- 4)  strip_t-1    absoluteGrob[strip.absoluteGrob.2306]
5   4 ( 3- 3, 7- 7)  strip_t-2    absoluteGrob[strip.absoluteGrob.2312]
6   5 ( 4- 4, 3- 3)   axis_l-1 absoluteGrob[axis-l-1.absoluteGrob.2299]
7   6 ( 4- 4, 6- 6)   axis_l-2         zeroGrob[axis-l-2.zeroGrob.2300]
8   7 ( 5- 5, 4- 4)   axis_b-1 absoluteGrob[axis-b-1.absoluteGrob.2285]
9   8 ( 5- 5, 7- 7)   axis_b-2 absoluteGrob[axis-b-2.absoluteGrob.2292]
10  9 ( 7- 7, 4- 7)       xlab             text[axis.title.x.text.2314]
11 10 ( 4- 4, 2- 2)       ylab             text[axis.title.y.text.2316]
12 11 ( 4- 4, 9- 9)  guide-box                        gtable[guide-box]
13 12 ( 2- 2, 4- 7)      title               text[plot.title.text.2337]

Copy the guide:

g2 <- ggplotGrob(p2)
g2[[1]][[12]] <- g1[[1]][[12]]
plot(g2)

Answer 2

ggplot(majdf, aes(x = val)) +
  geom_point(data = majtxt, aes(x = geq, colour = species), 
             y = 0.2, size = 0) +
  geom_density() +
  geom_vline(data = majtxt, aes(xintercept = geq)) +
  geom_text(data = majtxt, aes(x = geq, y = 0.2, label = geq, color = species), 
            angle = 90, show_guide = FALSE) +
  facet_wrap(~ lvl) +
  scale_colour_discrete(guide=guide_legend(override.aes=list(size=4)))

How this works: Add a point geom with the appropriate colour mapping. This will add a point into the legend. But, to keep it from showing up on the the plot, set the size of the point to 0. In the text geom, tell it not to add that part (the rotated a) to the legend (show_guide = FALSE). Finally, the legend will have just the point that you want and not the sideways a; unfortunately, it is drawn at the same size as in the plot, namely 0. So using the override.aes argument to guide_legend (which is passed to guide in scale_colour_discrete), set the size of the point to something "big".

This approach does not require pulling apart pieces to two different plots and stitching them back together.

An alternative way of specifying the guide parameters is using the guides function instead of passing it as an argument to scale_colour_manual:

ggplot(majdf, aes(x = val)) +
  geom_point(data = majtxt, aes(x = geq, colour = species), 
             y = 0.2, size = 0) +
  geom_density() +
  geom_vline(data = majtxt, aes(xintercept = geq)) +
  geom_text(data = majtxt, aes(x = geq, y = 0.2, label = geq, color = species), 
            angle = 90, show_guide = FALSE) +
  facet_wrap(~ lvl) +
  guides(colour = guide_legend(override.aes=list(size=4)))

The resulting graphic is the same.