Plot design

The default ggplot2 design is not particularly elegant, in my opinion. In particular, the gray background grid is a prime example of chartjunk, which stands for graphical elements that distract the viewer without providing any useful information. For example, see this ggplot2 visualization of the mpg data set:

ggplot(mpg, aes(x = cty, y = hwy, colour = factor(cyl))) + 
   geom_point(size=2.5) + theme_gray()

The default design of cowplot, by contrast, is clean and simple. It looks similar to ggplot2’s theme_classic(), but there are important but subtle difference we’ll discuss later.

In particular, the cowplot default theme works nicely in conjunction with the save_plot() function the package provides, such that the output pdfs are nicely formatted and scaled and don’t require any additional parameter fiddling:

require(cowplot)
plot.mpg <- ggplot(mpg, aes(x = cty, y = hwy, colour = factor(cyl))) + 
  geom_point(size=2.5)
# use save_plot() instead of ggsave() when using cowplot
save_plot("mpg.pdf", plot.mpg,
          base_aspect_ratio = 1.3 # make room for figure legend
          )

By default, cowplot disables grid lines on the plot. In many cases, this is the cleanest and most elegant way to display the data. However, sometimes gridlines may be useful, and thus copwlot provides a simple way of adding gridlines, via the function background_grid():

plot.mpg + background_grid(major = "xy", minor = "none")

While the same result could be obtained using the function theme(), the function background_grid() makes the most commonly used option easily accessible. See the reference documentation for details.

Arranging graphs into a grid

A major problem with ggplot2 is that it is not particularly easy to add labels and other annotations to a plot. ggplot2 strictly separates the plot panel (the part inside the axes) from the rest of the plot, and while it’s generally easy to modify one or the other we cannot easily change both. To solve this issue in a generic way, cowplot implements a generic drawing layer on top of ggplot2. In this drawing layer, you can add arbitrary graphical elements on top of a graph. This concept will be discussed in detail in the next section. For now, let’s discuss how this capability allows us to make nice compound plots.

The need for compound plots arises frequently when writing scientific publications. For example, let’s assume we have created the following two plots, and we would like to combine them into a two-part figure with parts A and B. In our case, the plots are the following:

plot.mpg <- ggplot(mpg, aes(x = cty, y = hwy, colour = factor(cyl))) + 
  geom_point(size=2.5)
plot.mpg

plot.diamonds <- ggplot(diamonds, aes(clarity, fill = cut)) + geom_bar() +
  theme(axis.text.x = element_text(angle=70, vjust=0.5))
plot.diamonds

cowplot allows us to combine them into one graph via the function plot_grid():

plot_grid(plot.mpg, plot.diamonds, labels=c("A", "B"))

By default, the function will attempt to achieve a reasonable layout of the plots provided. However, you can precisely manipulate the layout by specifying the number of rows or columns or both:

plot_grid(plot.mpg, NULL, NULL, plot.diamonds, labels=c("A", "B", "C", "D"), ncol = 2)

plot_grid(plot.mpg, plot.diamonds, plot.mpg, labels=c("A", "B", "C"), nrow = 1)

The function plot_grid() works nicely in combination with the function save_plot(), which can be told about the grid layout. For example, if we want to save a 2-by-2 figure, we might use this code:

plot2by2 <- plot_grid(plot.mpg, NULL, NULL, plot.diamonds,
                      labels=c("A", "B", "C", "D"), ncol = 2)
save_plot("plot2by2.pdf", plot2by2,
          ncol = 2, # we're saving a grid plot of 2 columns
          nrow = 2, # and 2 rows
          # each individual subplot should have an aspect ratio of 1.3
          base_aspect_ratio = 1.3
          )

The advantage of saving figures in this way is that you can first develop the code for individual figures, and once each individual figure looks the way you want it to you can easily combine the figures into a grid. save_plot() will make sure to scale the overall figure size such that the individual figures look the way they do when saved individually (as long as they all have the same base_aspect_ratio).

Generic plot annotations

Finally, let’s discuss how we can use cowplot to create more unusual plot designs. For example, let’s take the mpg image from the previous section, label it with an A in the top-left corner, and mark it as a draft:

ggdraw(plot.mpg) + 
  draw_plot_label("A", size = 14) + 
  draw_text("DRAFT!", angle = 45, size = 80, alpha = .2)

The function ggdraw() sets up the drawing layer, and functions that are meant to operate on this drawing layer all start with draw_. The resulting object is again a standard ggplot2 object, and you can do with it whatever you might do with a regular ggplot2 plot, such as save it with ggsave(). [However, as mentioned before, I recommend using save_plot() instead.]

In fact, because ggdraw() produces a standard ggplot2 object, we can draw on it with standard geoms if we want to. For example:

t <- (0:1000)/1000
spiral <- data.frame(x = .45+.55*t*cos(t*15), y = .55-.55*t*sin(t*15), t)
ggdraw(plot.mpg) + 
  geom_path(data = spiral, aes(x = x, y = y, colour = t), size = 6, alpha = .4)

I don’t know if this is useful in any way, but it shows the power of the approach.

Importantly, though, in all cases discussed so far, the main plot was below all other elements. Sometimes, you might want the plot on top. In this case, you can initialize an empty drawing canvas by calling ggdraw() without any parameters. You then place the plot by calling draw_plot(). Notice the difference in the two plots produced by the following code:

boxes <- data.frame(
  x = sample((0:36)/40, 40, replace = TRUE),
  y = sample((0:32)/40, 40, replace = TRUE)
)
# plot on top of annotations
ggdraw() + 
  geom_rect(data = boxes, aes(xmin = x, xmax = x + .15, ymin = y, ymax = y + .15),
            colour = "gray60", fill = "gray80") +
  draw_plot(plot.mpg) +
  draw_text("Plot is on top of the grey boxes", x = 1, y = 1,
            vjust = 1, hjust = 1, size = 10, fontface = 'bold')
# plot below annotations
ggdraw(plot.mpg) + 
  geom_rect(data = boxes, aes(xmin = x, xmax = x + .15, ymin = y, ymax = y + .15),
            colour = "gray60", fill = "gray80") + 
  draw_text("Plot is underneath the grey boxes", x = 1, y = 1,
            vjust = 1, hjust = 1, size = 10, fontface = 'bold')

Note that placing a plot on top of annotations only makes sense if the plot background is transparent. This is one of the main differences between theme_cowplot() and theme_classic(). If you tried the same example with theme_classic(), the gray boxes underneath the plot would not show.

Finally, the draw_plot() function also allows us to place graphs at arbitrary locations and at arbitrary sizes onto the canvas. This is useful for combining subplots into a layout that is not a simple grid, e.g. with one sub-plot spanning the entire width of the figure and two other figures using up half of the figure width:

plot.iris <- ggplot(iris, aes(Sepal.Length, Sepal.Width)) + 
  geom_point() + facet_grid(. ~ Species) + stat_smooth(method = "lm") +
  background_grid(major = 'y', minor = "none") + # add thin horizontal lines 
  panel_border() # and a border around each panel
# plot.mpt and plot.diamonds were defined earlier
ggdraw() +
  draw_plot(plot.iris, 0, .5, 1, .5) +
  draw_plot(plot.mpg, 0, 0, .5, .5) +
  draw_plot(plot.diamonds, .5, 0, .5, .5) +
  draw_plot_label(c("A", "B", "C"), c(0, 0, 0.5), c(1, 0.5, 0.5), size = 15)

The functions background_grid() and panel_border() are convenience functions defined by cowplot to save some typing when manipulating the background grid and panel border.

Of course, we can also go crazy:

ggdraw() +
  #geom_rect(data = boxes, aes(xmin = x, xmax = x + .15, ymin = y, ymax = y + .15),
  #          colour = "gray60", fill = "red", alpha=.03) +
  geom_path(data = spiral, aes(x = x, y = y, colour = t), size = 6, alpha = .4) +
  draw_plot(plot.diamonds, -.05, -.1, .55, .55) +
  draw_plot(plot.diamonds, .65, .4, .5, .5) +
  draw_plot(plot.mpg, .3, .3, .4, .4) +
  draw_plot(plot.iris, 0, .7, .7, .35 ) +
  draw_plot(plot.iris, .45, .0, .6, .3 )