4  Grouped plots and overplotting

4.1 Intended learning outcomes

By the end of this chapter you will be able to:

• Map a categorical variable to colour or fill to display three variables on one plot
• Distinguish setting an aesthetic inside aes() from setting it as a fixed value outside aes()
• Split a plot into panels with facet_wrap() and facet_grid()
• Reduce overplotting with alpha and geom_jitter()

4.2 Functions used

• ggplot2: aes(), geom_point(), geom_jitter(), geom_smooth(), facet_wrap(), facet_grid(), coord_cartesian()
• colour and fill aesthetics

4.3 Adding a third variable with colour

So far our plots have used at most two variables. To show how the number who died on the Middle Passage varied across the period of British involvement, we plot arrival_year against died and map start_port to colour.

ggplot(dat_filter, aes(arrival_year, died, colour = start_port)) +
  geom_point(alpha = 0.4) +
  geom_smooth() +
  scale_colour_viridis_d() +
  labs(x = "Year of arrival",
       y = "Number of deaths per voyage",
       colour = "Port of origin") +
  coord_cartesian(xlim = c(1641, 1815))

Figure 4.1: Number of deaths per voyage by year of arrival, coloured by port of origin.

TipInside or outside aes()?

• Inside aes(): the aesthetic varies with the data. aes(colour = start_port) means each port gets a different colour.
• Outside aes(): the aesthetic is a fixed value. geom_point(colour = "dodgerblue") colours all points the same regardless of port.

The same rule applies to fill, shape, size and linetype.

4.4 Facetting

facet_wrap() splits the plot into one panel per level of a variable. It is an alternative way to display the same three-variable relationship and is often easier to read than overlaid colours.

ggplot(dat_filter, aes(arrival_year, died)) +
  geom_point(aes(colour = start_port), show.legend = FALSE, alpha = 0.4) +
  geom_smooth() +
  scale_colour_viridis_d() +
  facet_wrap(~ start_port, nrow = 3) +
  labs(x = "Year of arrival",
       y = "Number of deaths per voyage",
       title = "Deaths during the Middle Passage, by port of origin") +
  coord_cartesian(xlim = c(1641, 1815))

Figure 4.2: Same data, faceted by port of origin.

The base mapping for the smoother uses no colour mapping, so a single smoother is drawn per facet. Inside geom_point() we map colour = start_port so the points within each facet are coloured by port even though there is only one port per panel; this is purely decorative.

• Try changing nrow = 3 to nrow = 1 and rendering. What happens?

TipActivity

Replace facet_wrap(~ start_port, nrow = 3) with facet_grid(start_port ~ .). Read the help with ?facet_grid to see when one is preferred over the other.

4.5 Overplotting

The earliest scatterplots in sec-basic-plots had so many overlapping points that the density of the data was hidden. Two techniques help.

alpha makes each point semi-transparent, so dense regions appear darker.

geom_jitter() adds a small random offset, so points stacked on the same value are visible separately.

opaque <- ggplot(dat_filter, aes(start_port, mortality)) +
  geom_point() +
  labs(title = "alpha = 1 (default)")

transparent <- ggplot(dat_filter, aes(start_port, mortality)) +
  geom_point(alpha = 0.1) +
  labs(title = "alpha = 0.1")

jittered <- ggplot(dat_filter, aes(start_port, mortality)) +
  geom_jitter(alpha = 0.2, width = 0.3, height = 0) +
  labs(title = "geom_jitter()")

opaque + transparent + jittered

Figure 4.3: Three approaches to overplotting.

Warning

geom_jitter() moves points slightly. That is fine for showing density on a categorical axis, but if you jitter on a continuous axis you are misrepresenting the data. Set the width and height arguments deliberately. Here we use height = 0 so vertical positions are unchanged.