Enhancing geom_text() and geom_label()

Warning

Versions of package ‘ggpp’ released before 0.5.0 included a preliminary attempt at the design of these geometries based on a different logic and interface. At the time of writing, version 0.5.2 is available through CRAN. To reproduce the examples at the end of this page please use ‘ggpp’ (>= 0.5.0) and preferably ‘ggpp’ (>= 0.5.2).

Selective mapping of aesthetics

I have defined in package ‘ggpp’ enhanced versions of geom_text() and geom_label() under the names of geom_text_s() and geom_label_s() . The s is for segment, at least it was when I thought of these names. The versions described are included in version 0.5.0 and later.

The idea of better supporting the use of data labels in plots created with R package ‘ggplot2’ is not new. I have been for a long time a user of the repulsive geometries geom_text_repel() and geom_label_repel() from package ‘ggrepel’. I have even contributed some code to ‘ggrepel’. However, some time back when designing some new position functions, the idea of developing non-repulsive geometries suitable for data labels started growing on me.

Considering data visualization principles, and the philosophy of the grammar of graphics the proliferation of multiple versions of the same aesthetics is to be avoided. Colour and mappings of colours to values using a single scale per plot ensure that there is a single meaning for each value of the aesthetic. Aesthetics linewidth and linetype can be applied to the border of label boxes and provide an additional approach for highlighting or distinguishing individual data labels, when colour is not available.

In geom_text() transparency set with alpha affects the text while in geom_label() it affects only fill. In geom_label() the colour aesthetics controls at the same time the colour of the text and the line bordering the label; label.size controls the size of the border line, but it is a parameter rather than an aesthetic. This means that it is not possible to use the border width, its abscence or its colour to highlight selected data labels. Lack of support for the linetype aesthetic of the borber, makes it dificult to highlight individual data labels in black and white printing.

I aimed from the start to keep the use of any new geometries simple, and as consistent with the grammar of graphics as possible. The interfaces of geom_text_s() and geom_label_s() are still not completely stable. However, these geometries are very likely to remain backwards compatible in future versions of ‘ggpp’. By default geom_text() and geom_label() behave (almost?) as their counterparts from ‘ggplot2’. They do not make use of any new aesthetic, they have additional formal parameters instead.

The enhancements are: 1) if used with one of the position functions from ‘ggpp’, which keep the original position, segments are drawn connecting the text or label to the observation. The justification, vjust and hjust and thus the anchoring point of the segment to the text or label are computed automatically based on the direction of displacement. 2) A new formal parameter colour.target makes it possible to select to which elements to apply the mapped colour and to which elements to apply the default colour. 3) Similarly a new formal parameter alpha.target allows similar control of what elements obey the mapped alpha values and which ones obey the default alpha. Currently supported elements are "text", "segment", "box.line" and "box.fill" in any combination, and "box" and "all" as aliases to save typing. 4) Add support for linewidth and linetype aesthetics in geom_label_s() applying them to the box border line, independently of segment.linewidth which a parameter.

The use of segments or arrows to make clear to which data observation data labels are linked is frequent except in very sparse clouds of observations. Using by default automatic justification "position" based on the direction of displacement by nudging, dodging, etc., to select the anchor point ensures that short segments will not cause difficulties even with long text labels. It also means that small displacements away from observations are enough to avoid overlaps between label and labelled observation. Of course, the vjust and hjust aesthetics except for the support of the additional value "position", work as accepting the same values as geom_text() and geom_label() and to the same effect.

In the case of these two geometries, the design stage was time-consuming and included attempts that in use resulted awkward. Implementation of the new features was straightforward. The examples below show some of the new possibilities for the graphical design of text label annotations in ggplots that geom_text_s() and geom_label_s() make possible.

Examples

Tip

Above each plot you will find the code used to create the figures, folded behind a triangle and the word Code. Clicking on the triangle or on “Code” makes the code visible. The code can be easily copied by clicking on the icon that appears when the cursor in on the upper right corner of the box.

The code used for the individual plots requires the core in the code chunk immediately under this callout to be run first to attach the packages used, prepare the data and save a base plot.

Code

library(ggpp)
my.cars <- mtcars[c(TRUE, FALSE, FALSE, FALSE), ]
my.cars$name <- rownames(my.cars)
p <- 
  ggplot(my.cars, aes(wt, mpg, label = name)) +
  scale_colour_discrete(l = 40) # luminance, make colours darker

A simple example, using nudging to displace the labels.

Code

# Use nudging
p +
  geom_text_s(nudge_x = 0.12) +
  geom_point() +
  expand_limits(x = 6.2)

We add arrow points to the segments.

Code

p +
  geom_text_s(nudge_x = 0.15,
              arrow = arrow(length = grid::unit(1.5, "mm")),
              point.padding = 0.4) +
  geom_point() +
  expand_limits(x = 6.2)

Using colour with the default target, which, consistently with geom_text() it is the text label.

Code

p +
  geom_text_s(aes(colour = factor(cyl)),
              angle = 90,
              nudge_y = 1) +
  geom_point() +
  expand_limits(y = 27)

We can select a different graphic element as the target for the colour aesthetic, here we apply colour to the connecting segments.

Code

p +
  geom_text_s(aes(colour = factor(cyl)),
              colour.target = "segment",
              angle = 90,
              nudge_y = 1,
              show.legend = FALSE) +
  geom_point(aes(colour = factor(cyl))) +
  expand_limits(y = 27)

And here we apply colour to all the elements created by geom_text_s().

Code

p +
  geom_text_s(aes(colour = factor(cyl)),
              colour.target = "all",
              angle = 90,
              nudge_y = 1) +
  geom_point() +
  expand_limits(y = 27)

In the case of geom_label_s() additional targets for colour are available. We start with the default, that is consistent with geom_label() applying the colour aesthetic to the border line and text.

Code

p +
  geom_label_s(aes(colour = factor(cyl)), nudge_x = 0.3) +
  geom_point() +
  expand_limits(x = 7)

We set the default coulour to "grey50" to make segments and box borders less prominent, achieving a calm design.

Code

p +
  geom_label_s(aes(colour = factor(cyl)),
               colour.target = "text",
               default.colour = "grey50",
               linewidth = 0.4,
               nudge_x = 0.3) +
  geom_point() +
  expand_limits(x = 7) +
  theme_bw()

We can use the opposite approach, keeping the text black for maximum legibility at a small size and apply colour to the remaining elements.

Code

p +
  geom_label_s(aes(colour = factor(cyl)), 
               size = 2.8,
               colour.target = c("box", "segment"),
               linewidth = 0.4,
               nudge_x = 0.3) +
  geom_point(aes(colour = factor(cyl)), size = 3) +
  expand_limits(x = 7)

Aesthetic linetype can be used to unobtrusively highlight some labels in the previous plot.

Code

p +
  geom_label_s(aes(colour = factor(cyl), linetype = grepl("Merc", name)), 
               size = 2.8,
               colour.target = c("box", "segment"),
               linewidth = 0.4,
               nudge_x = 0.3,
               show.legend = FALSE) +
  geom_point(aes(colour = factor(cyl)), size = 3) +
  expand_limits(x = 7)

A visually simple and space-saving design, by combining fill and colour but.

Code

p +
  geom_label_s(aes(colour = factor(cyl), fill = factor(cyl)),
               nudge_x = 0.2,
               size = 3.5,
               colour.target = "all",
               alpha.target = c("box", "segment"),
               alpha = 0.15,
               linewidth = 0,
               segment.linewidth = 1,
               label.padding = grid::unit(0.12, "lines")) +
  geom_point(aes(colour = factor(cyl)), size = 2.5) +
  expand_limits(x = 7) +
  theme_bw()

If we disable the plotting of segments, geom_tetx_s() behaves like geom_text() and geom_label_s() as geom_label() except for the flexibility in the mapping of colour and alpha aesthetics in geom_label_s() and the default justification "position".

Code

p +
  geom_text_s(aes(colour = factor(cyl)),
              add.segments = FALSE,
              angle = 90,
              nudge_y = 0.4,
              show.legend = FALSE) +
  geom_point(aes(colour = factor(cyl)), size = 2.5) +
  expand_limits(y = 27)

Tip

For additional information, please, see the help page for geometries geom_text_s() and geom_label_s() and the documentation of package ‘ggpp’, including a vignette.

Reuse