8  Visualization II

8.1 Learning objectives

By the end of this chapter, you should be able to:

  • add color or fill only when it represents useful information;
  • distinguish a fixed appearance from a data mapping;
  • compare several time series;
  • choose between grouped, stacked, and percentage-stacked bars;
  • compare distributions with box plots;
  • separate crowded groups with facets; and
  • save a finished chart.

8.2 When does a chart need more complexity?

Chapter 7 used a line, bars, a histogram, and a scatterplot to answer four basic questions. An advanced chart should begin from one of those forms and add something only because the reporting question requires it.

New reporting need Possible addition
Compare several groups over time Map a group to color
Compare groups directly within categories Grouped bars
Show how groups contribute to a total Stacked bars
Compare composition rather than totals Percentage-stacked bars
Compare several distributions Box plots
Reduce overlap among many groups Facets or small multiples
Continue in the same project

Open djr.Rproj and create 08-data-visualization-ii.Rmd. Reuse data/gpfg_5_years.csv and save finished charts in outputs/.

library(tidyverse)

gpfg <- read_csv("data/gpfg_5_years.csv")

gpfg_latest <- gpfg |>
  filter(year == 2025)

8.3 Map a column or set an appearance

This distinction matters throughout the chapter:

geom_point(color = "#0072B2")

color is outside aes(), so every point receives the same fixed color. It does not create a legend.

geom_point(aes(color = region))

color is inside aes(), so values of region are mapped to different colors. Because color now represents data, ggplot2 creates a legend.

The same rule applies to fill, size, and other visual properties.

8.5 Compare groups with grouped bars

Question and analysis table

How did the total market value of NBIM’s reported equity holdings differ across three industries and three regions in 2025?

Inside filter(), conditions separated by a comma must both be true. The following code therefore keeps a row only when its region and industry are both in the selected vectors:

region_industry <- gpfg_latest |>
  filter(
    region %in% c("Asia", "Europe", "North America"),
    industry %in% c("Technology", "Financials", "Industrials")
  ) |>
  group_by(industry, region) |>
  summarise(market_value_nok = sum(market_value_nok)) |>
  mutate(market_value_nok_billions = market_value_nok / 1000000000)

region_industry

The visual roles are industry on x, value on y, and region mapped to fill. Start with the default bars:

region_industry |>
  ggplot(
    aes(
      x = industry,
      y = market_value_nok_billions,
      fill = region
    )
  ) +
  geom_col()

The default is stacked. This shows the combined height and each region’s contribution, but only the bottom segment begins at the same point, so the other segments are harder to compare precisely.

For direct comparison, change one argument:

region_industry |>
  ggplot(
    aes(
      x = industry,
      y = market_value_nok_billions,
      fill = region
    )
  ) +
  geom_col(position = "dodge")

position = "dodge" places the regions beside one another. The data and aesthetic mappings have not changed; only the bar arrangement has changed.

Add labels and styling after choosing the arrangement that fits the question:

region_industry |>
  ggplot(
    aes(
      x = industry,
      y = market_value_nok_billions,
      fill = region
    )
  ) +
  geom_col(position = "dodge") +
  scale_fill_manual(
    values = c(
      "Asia" = "#009E73",
      "Europe" = "#E69F00",
      "North America" = "#0072B2"
    )
  ) +
  labs(
    title = "Regional values differed across selected industries",
    x = NULL,
    y = "Market value in NOK billions",
    fill = "Region",
    caption = "Three selected industries and regions; year-end 2025\nSource: Norges Bank Investment Management"
  ) +
  theme_minimal() +
  theme(legend.position = "top")

8.6 Compare composition with percentage bars

A different question requires a different bar arrangement:

Within each selected industry, what share of NBIM’s reported total across the three selected regions came from each region?

Starting from the same table and mappings, change the position to "fill":

region_industry |>
  ggplot(
    aes(
      x = industry,
      y = market_value_nok_billions,
      fill = region
    )
  ) +
  geom_col(position = "fill") +
  scale_fill_manual(
    values = c(
      "Asia" = "#009E73",
      "Europe" = "#E69F00",
      "North America" = "#0072B2"
    )
  ) +
  labs(
    title = "Regional composition differed across selected industries",
    x = NULL,
    y = "Share of selected-region market value",
    fill = "Region",
    caption = "Source: Norges Bank Investment Management"
  ) +
  theme_minimal() +
  theme(legend.position = "top")

Every bar now has the same height. This makes composition easier to compare but hides differences in total amounts. A percentage chart must not be used to support a claim about which industry has the largest total.

8.7 Compare distributions with box plots

Chapter 7’s histogram described all 2025 holding values together. Now ask:

Among companies in NBIM’s 2025 data, how did the distribution of reported holding values differ across industries?

The source table already contains the two required columns. geom_boxplot() summarizes a numeric distribution for each group: the line inside a box is the median, the box contains the middle half of the values, and points beyond the whiskers identify values to inspect.

Visual role Column
X-axis market_value_usd
Y-axis industry

Begin without additional scales:

industry_box <- gpfg_latest |>
  ggplot(
    aes(
      x = market_value_usd,
      y = industry
    )
  ) +
  geom_boxplot()

industry_box

The large values compress most boxes near zero. Because all displayed values are positive, add a logarithmic x-axis:

industry_box +
  scale_x_log10() +
  labs(
    title = "Holding values varied widely within every industry",
    x = "Market value in USD (log scale)",
    y = NULL,
    caption = "Source: Norges Bank Investment Management"
  ) +
  theme_minimal()

Points beyond the whiskers deserve inspection, but they are not automatically errors. scale_x_log10() changes the display scale rather than the underlying data. Always disclose a log scale in the axis label.

8.8 Add a third variable with facets

Return to the basic relationship question from Chapter 7:

Among companies in NBIM’s 2025 data, were higher ownership percentages associated with larger holding values, and did that pattern differ by region?

First select three regions so the comparison remains readable:

selected_holdings <- gpfg_latest |>
  filter(region %in% c("Asia", "Europe", "North America"))

Begin with the familiar scatterplot, adding transparency because many points overlap:

relationship_plot <- selected_holdings |>
  ggplot(
    aes(
      x = market_value_usd,
      y = ownership_pct
    )
  ) +
  geom_point(alpha = 0.25)

relationship_plot

Now add one new layer, facet_wrap(), to create a panel for each region. In facet_wrap(~ region), the ~ can be read as “make panels by region”:

relationship_plot +
  scale_x_log10() +
  facet_wrap(~ region) +
  labs(
    title = "The relationship can be inspected separately by region",
    x = "Market value in USD (log scale)",
    y = "Ownership (%)",
    caption = "Source: Norges Bank Investment Management"
  ) +
  theme_minimal()

The panels use the same axes, making their patterns easier to compare than many overlapping colors in one panel.

8.9 Save a finished chart

Before saving, verify that the chart matches the analysis table and that its title, period, measure, unit, and source are clear.

ggsave("outputs/country-trends.png", country_trend_plot)

The filename and chart object are enough here. Width, height, and resolution are optional controls for a later publication requirement.

Interactive charts are optional

Interaction can add tooltips, filters, and zooming, but it also introduces new packages and HTML-specific behavior. First make a correct, readable static chart. Interactive graphics can be considered later when interaction helps a specific reporting task.

8.10 Practice

Choose one of the following:

  1. Compare three industries across five years with colored lines.
  2. Compare selected regions with grouped and stacked bars, then explain which arrangement better answers your question.
  3. Add facets to a relationship plot and explain what became easier to see.

For every version, state the question, list the columns, identify the visual roles, show the basic plot first, and explain every added layer.

8.11 Takeaways

Function or option What it adds
color = column inside aes() Maps a column to line or point color
fill = column inside aes() Maps a column to bar fill
Fixed color outside aes() Changes appearance without representing data
position = "dodge" Places groups beside one another
position = "fill" Converts stacked bars to equal-height composition bars
geom_boxplot() Compares numeric distributions across groups
scale_x_log10() Spreads positive values across a logarithmic axis
facet_wrap() Creates a separate panel for each group
scale_color_manual() Assigns deliberate colors to named groups
ggsave() Saves a completed chart
Want more control?

The official ggplot2 reference documents additional scales, legends, themes, annotations, and saving options. Add them one at a time and only when they serve the reporting question.