Draw a line across multiple ggplot figures in a gtable_matrix

Question

I am trying to draw a line across two ggplot histograms in a gtable_matrix, so that the mean of the values in the one histogram is overlaid across both plots.

However, I cannot get at the device coordinates of the plotting area. In base graphics, I would use grconvertX(), but where can I find the device coordinates of the plotting area of ggplot so I can convert numbers on my 'user' scale (0-10) to device coordinates?

In the example below, I have meticulously found the numbers to plug in to get the line at the correct location, but as soon as the plot is rescaled, or the axis labels change, or any other plot element changes, it breaks down. Probably won't work as intended on your machine either.

library(ggplot2)
library(grid)
library(gtable)

n_1 = 10
n_2 = 10
mean_1 = 5.5
sd_1 = 1
mean_2 = 7
sd_2 = 1
data = data.frame(y = c(
  rnorm(n_1, mean_1, sd_1),
  rnorm(n_2, mean_2, sd_2)
),
group = c(rep("1", n_1), rep("2", n_2)))
data$y[data$y > 10] <- 10
data$y[data$y < 0] <- 0

plots <- lapply(c("1", "2"), function(x) {
  ggplotGrob(
    ggplot(data[data$group == x,], aes(y)) +
      geom_histogram(
        breaks = seq(0, 10, length.out = 12),
        fill = ifelse(x == "1", "blue", "red"),
        colour = "black",
        alpha = .2
      ) +
      theme_classic() +
      theme(axis.title.x = element_blank()) +
      ylab(x) +
      scale_x_continuous(expand = c(0, 0), limits = c(0, 10)) +
      scale_y_continuous(expand = c(0, 0), limits = c(0, 4))
  )

})

gt <- gtable_matrix(
  "histograms",
  matrix(plots, nrow = 2, byrow = TRUE),
  widths = unit(1, "null"),
  heights = unit(c(1, 1), "null")
)

left <- textGrob("Frequency", rot = 90, just = c(.5, .5))
gt <-
  gtable_add_cols(gt, widths = grobWidth(left) + unit(0.5, "line"), 0)
gt <- gtable_add_grob(
  gt,
  left,
  t = 1,
  b = nrow(gt),
  l = 1,
  r = 1,
  z = Inf
)

gt <- gtable_add_cols(gt, widths = unit(0.5, "line"))
grid.newpage()
grid.draw(gt)

pushViewport(viewport())

grid.lines(y = c(.05, .98),
           x = (.11 + (5 / 10 * .861)),
           gp = gpar(col = "red"))
popViewport()

Is there a reason why you're building so much of this from scratch instead of using facets? A `facet_wrap` might not get *exactly* what you want, but it will probably be very very close and much more flexible — camille, Jun 04 '18 at 17:03
Camille, thank you for your request for clarification. I am using gtable because this example comes from a larger bit of code. Could you please explain how faceting addresses my question of how to draw a line in the right place across both panels? — Caspar van Lissa, Jun 04 '18 at 18:03
https://stackoverflow.com/questions/42151880/ggplot-drawing-multiple-lines-across-facets — user9902495, Jun 06 '18 at 09:30
The comment by user9902495 above exactly answers the question. — Caspar van Lissa, Jun 06 '18 at 14:25

score 0 · Answer 1 · answered Jun 04 '18 at 19:19

0

Here's a stripped-down version with facets. You can decide whether this accomplishes enough of what you're looking for to drop the gtable stuff.

Use a geom_vline with the intercept set to the mean of your y-values; this will put it in the same place on each facet. I took out the strip text (strip.text = element_blank()) to mimic what you'd done with removing the titles of the two plots. Other than that, it's just a standard facet_wrap by groups.

library(tidyverse)

n_1 = 10
n_2 = 10
mean_1 = 5.5
sd_1 = 1
mean_2 = 7
sd_2 = 1
data = data.frame(y = c(
  rnorm(n_1, mean_1, sd_1),
  rnorm(n_2, mean_2, sd_2)
),
group = c(rep("1", n_1), rep("2", n_2)))
data$y[data$y > 10] <- 10
data$y[data$y < 0] <- 0

ggplot(data, aes(x = y, fill = group)) +
  geom_histogram(breaks = seq(0, 10, length.out = 12)) +
  geom_vline(aes(xintercept = mean(y))) +
  facet_wrap(~ group, ncol = 1) +
  theme_minimal() +
  theme(strip.text = element_blank())

answered Jun 04 '18 at 19:19

camille

16,432
18
38
60

1

Thanks, but this is not the kind of solution I'm looking for. I want to get the location of my x-axis in the units of my graphical device, akin to grconvertX. I'm not looking for a workaround to get a similar visual effect. – Caspar van Lissa Jun 04 '18 at 20:38
Okay, then I'm afraid I don't really understand what you're trying to do. How does the mean of the values relate to the user's viewport? – camille Jun 05 '18 at 16:16
In base R you can convert a value on the user scale (my example: 0-10) to the location of that value on the graphics device. This is useful if you want to plot something over the original graphic at a specific location, e.g., the mean. I want to do something similar, but for ggplot. – Caspar van Lissa Jun 06 '18 at 06:25

Draw a line across multiple ggplot figures in a gtable_matrix

1 Answers1

Linked