5  Reshaping data and combining plots

5.1 Intended learning outcomes

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

• Reshape data from wide to long with pivot_longer()
• Plot grouped distributions using one mapping per categorical column rather than one column per category
• Order categorical axes deliberately with factor() or fct_relevel()
• Combine multiple plots into a single figure with the patchwork package

5.2 Functions used

• tidyr: pivot_longer(), drop_na()
• forcats: fct_relevel()
• ggplot2: geom_boxplot()
• patchwork: +, /, plot_layout(), plot_annotation()

5.3 From wide to long

The dataset has three columns recording the demographic make-up of each voyage. men, women and children are each a proportion of the people embarked. Most voyages in the database do not record this breakdown, so these three columns are missing for the great majority of rows. Before reshaping, we drop the voyages with no demographic data using drop_na(), which removes any row that is missing a value in one of the named columns.

dat_demographic <- dat_filter |>
  drop_na(men, women, children)

head(dat_demographic[, c("start_port", "men", "women", "children")])

start_port	men	women	children
Bristol	0.38397	0.23207	0.38397
Bristol	0.39200	0.25600	0.35200
Bristol	0.48780	0.17480	0.33740
Bristol	0.26943	0.22280	0.50777
Bristol	0.14881	0.13095	0.72024
Bristol	0.47737	0.18519	0.33745

This is the wide form. Each demographic category has its own column. To plot them together we need the long form, where one column holds the category and another holds the value.

pivot_longer() does the conversion. We tell it which columns to gather (cols), the name of the new column that will hold the original column names (names_to) and the name of the new column that will hold the values (values_to).

dat_long <- dat_demographic |>
  pivot_longer(cols = c(men, women, children),
               names_to = "demographic",
               values_to = "proportion")

head(dat_long[, c("start_port", "demographic", "proportion")])

start_port	demographic	proportion
Bristol	men	0.38397
Bristol	women	0.23207
Bristol	children	0.38397
Bristol	men	0.39200
Bristol	women	0.25600
Bristol	children	0.35200

Each voyage now appears three times, once per demographic category. The column is called demographic because it holds the categories man, woman and child. The values are proportions of the people embarked.

Note

The demographic figures describe only the voyages that recorded them, a minority of the total. The original column names men, women and children are inherited from the source database. In the reshaped data, the column demographic holds those labels as values.

5.4 Ordering categorical axes

By default, ggplot will plot demographic in alphabetical order, which puts children first. We probably want men, women, children or the other way around. fct_relevel() from forcats is the easiest way.

dat_long <- dat_long |>
  mutate(demographic = fct_relevel(demographic, "men", "women", "children"))

5.5 Grouped boxplots

We can now show the distribution of the demographic proportions, broken down by port of origin, all on one plot.

ggplot(dat_long, aes(x = demographic, y = proportion, fill = start_port)) +
  geom_boxplot(alpha = 0.6) +
  scale_fill_viridis_d(option = "E") +
  labs(x = "Demographic category",
       y = "Proportion of those embarked",
       fill = "Port of origin")

Figure 5.1: Demographic composition of voyages from each port.

Men make up the majority of those embarked on every type of voyage, but the proportion of women and children varies markedly by port.

TipActivity

Change the value of alpha and observe what happens. Try alpha = 0.2 and alpha = 1.

5.6 Piping straight into ggplot

You can combine filter() and ggplot() in a single pipe rather than creating an intermediate object. The example below takes the long-format data, filters to voyages that disembarked in the Americas, and plots the result.

dat_long |>
  filter(outcome == "Slaves disembarked in Americas") |>
  ggplot(aes(x = demographic, y = proportion, fill = demographic)) +
  geom_boxplot(show.legend = FALSE) +
  scale_fill_viridis_d(option = "E") +
  labs(x = NULL,
       y = "Proportion of those embarked",
       title = "Demographic composition of voyages disembarking in the Americas")

Figure 5.2: Demographic composition of voyages that disembarked in the Americas.

5.7 Combining plots with patchwork

patchwork lets you arrange ggplot objects into multi-panel figures. The syntax is intuitive: + puts plots side by side, / stacks them vertically, and plot_layout() controls the arrangement.

First, save two plots as objects.

men_plot <- dat_long |>
  filter(demographic == "men") |>
  ggplot(aes(start_port, proportion, fill = start_port)) +
  geom_boxplot(show.legend = FALSE, alpha = 0.6) +
  scale_fill_viridis_d(option = "E") +
  labs(x = NULL, y = "Proportion men")

children_plot <- dat_long |>
  filter(demographic == "children") |>
  ggplot(aes(start_port, proportion, fill = start_port)) +
  geom_boxplot(show.legend = FALSE, alpha = 0.6) +
  scale_fill_viridis_d(option = "E") +
  labs(x = NULL, y = "Proportion children")

Then combine them.

men_plot + children_plot

Figure 5.3: Two boxplots side by side with patchwork.

Stacked vertically.

men_plot / children_plot

Figure 5.4: The same two boxplots stacked vertically.

plot_annotation() adds a single overall title across multiple panels.

(men_plot + children_plot) +
  plot_annotation(title = "Demographic composition by port of origin")

Figure 5.5: Combined figure with a shared title.

See the patchwork documentation for more elaborate layouts, shared legends and tag_levels for automatic panel labels (A, B, C).

5.8 Activities

TipActivity 1

Build a third plot showing the proportion of women by port, and combine all three plots (men_plot, women_plot, children_plot) in a single row.

NoteSolution

women_plot <- dat_long |>
  filter(demographic == "women") |>
  ggplot(aes(start_port, proportion, fill = start_port)) +
  geom_boxplot(show.legend = FALSE, alpha = 0.6) +
  scale_fill_viridis_d(option = "E") +
  labs(x = NULL, y = "Proportion women")

men_plot + women_plot + children_plot

TipActivity 2

Reshape the data back into wide format using pivot_wider(). (Hint: it takes names_from and values_from arguments.)