Agenda

• Axes and aspect ratios

• Annotations

• Themes (a little bit)

What we won't get to

Each of the following are pretty fundamental to good data viz, but we won't have time to go over them today. Please make sure to read the corresponding chapters:

• Handling high data density (lots of overlapping points)
• Compound figures
  • See {patchwork} and {cowplot}
• Exporting figures

Learning Objectives

• Understand how to make a wide variety of tweaks to ggplot to essentially make it look however you want it to.

• Understand common modifications to plots to make them more clear and reduce cognitive load

Axes

• Cartesian coordinates - what we generally use

Different units

Aspect ratio

Same scales

Use coord_fixed()

Changing aspect ratio

• Explore how your plot will look in its final size

• No hard/fast rules (if on different scales)

• Not even really rules of thumb

• Keep visual perception in mind

• Try your best to be truthful - show the trend/relation, but don't exaggerate/hide it

Handy function

(from an apparently deleted tweet from @tjmahr)

here's my favorite helper #rstats function. preview ggsave() output

ggpreview <- function (..., device = "png") {
fname <- tempfile(fileext = paste0(".", device))
ggplot2::ggsave(filename = fname, device = device, ...)
system2("open", fname)
invisible(NULL)
}

— tj mahr 🍕🍍 (@tjmahr)

Gist

(side note: gists are a good way to share things)

• See the full code/example here

• Let's take 5 minutes to play around:

  • Create a plot (could even be the example in the gist)

  • Try different aspect ratios by changing the width/length

Scale transformations

Raw scale

library(gapminder)
ggplot(gapminder, aes(year, gdpPercap)) +
  geom_line(aes(group = country),
            color = "gray70")

Log10 scale

ggplot(gapminder, aes(year, gdpPercap)) +
  geom_line(aes(group = country),
            color = "gray70") +
  scale_y_log10(labels = scales::dollar)

Scales

d <- tibble(x = c(1, 3.16, 10, 31.6, 100),
            log_x = log10(x))
ggplot(d, aes(x, 1)) +
  geom_point(color = "#0072B2")
ggplot(d, aes(x, 1)) +
  geom_point(color = "#0072B2") +
  scale_x_log10()
ggplot(d, aes(log_x, 1)) +
  geom_point(color = "#0072B2")

Scales

Don't transform twice

ggplot(d, aes(log_x, 1)) +
  geom_point(color = "#0072B2") +
  scale_x_log10() +
  xlim(-0.2, 2.5)

Careful with labeling

• Has the scale or the data been log transformed?
• Specify the base

library(ggtext)
ggplot(d, aes(log_x, 1)) +
  geom_point(color = "#0072B2") +
  labs(x = "log<sub>10</sub>(x)") +
  theme(axis.title.x = element_markdown())

Labels should denote the data, not the scale of the axis

ggplot(d, aes(x, 1)) +
  geom_point(color = "#0072B2") +
  scale_x_log10()

Labeling the above with $lo{g}_{10}\left(x\right)$ would be ambiguous and confusing

Disclaimer

• APA style requires the labels be made in specific ways

• Much of the following discussion still applies

• Our book (Wilke) uses a similar style throughout

Title

What is the point of your figure?

What are you trying to communicate

• Figures should have only one title
• Use integrated title/subtitles for sharing with a broad audience
  • Blog posts
  • Social media
  • Reports to stakeholders
• Keep figures in subtext when there's a designated format you must adhere to
• Make sure your figure has a title
  • Should not start with "This figure displays/shows..."

Caption

Consider stating the data source

Other details relevant to the figure but not important enough for a subtitle

Axis labels

• The title for the axis

• Critical for communication

• Never use variable names (very common and very poor practice)

• State the measure and the unit (if quantitative)

  • e.g., "Brain Mass (grams)", "Support for Measure (millions of people)", "Dollars spent"

  • Categorical variable likely will not need to the measurement unit

Omission

• Consider omitting obvious or redundant labels
  • Use labs(x = NULL) or labs(x = "")
  • If already using scale_x/y_*() just supply the name argument

Omission

• Do not omit axis titles that are not obvious

Don't overdo it

Among the most effective

• If possible, try to remove legends, and just include annotations

Building up a plot

remotes::install_github("clauswilke/dviz.supp")
head(tech_stocks)

## # A tibble: 6 x 6
##   company  ticker date        price index_price price_indexed
##   <chr>    <chr>  <date>      <dbl>       <dbl>         <dbl>
## 1 Alphabet GOOG   2017-06-02 975.6        285.2      342.0757
## 2 Alphabet GOOG   2017-06-01 966.95       285.2      339.0428
## 3 Alphabet GOOG   2017-05-31 964.86       285.2      338.3100
## 4 Alphabet GOOG   2017-05-30 975.88       285.2      342.1739
## 5 Alphabet GOOG   2017-05-26 971.47       285.2      340.6276
## 6 Alphabet GOOG   2017-05-25 969.54       285.2      339.9509

ggplot(tech_stocks, aes(date, price_indexed, color = ticker)) +
  geom_line()

ggplot(tech_stocks, aes(date, price_indexed, color = ticker)) +
  geom_line() +
  scale_color_OkabeIto()

ggplot(tech_stocks, aes(date, price_indexed, color = ticker)) +
  geom_line() +
  scale_color_OkabeIto(name = "Company",
                       breaks = c("GOOG", "AAPL", "FB", "MSFT"),
                       labels = c("Alphabet", "Apple", "Facebook", "Microsoft"))

Bad

ggplot(tech_stocks, aes(date, price_indexed, color = ticker)) +
  geom_line() +
  scale_color_OkabeIto(name = "Company",
                       breaks = c("FB", "GOOG", "MSFT", "AAPL"),
                       labels = c("Facebook", "Alphabet", "Microsoft", "Apple"))

Good

ggplot(tech_stocks, aes(date, price_indexed, color = ticker)) +
  geom_line() +
  scale_color_OkabeIto(name = "Company", 
                       breaks = c("FB", "GOOG", "MSFT", "AAPL"),
                       labels = c("Facebook", "Alphabet", "Microsoft", "Apple")) +
  scale_x_date(name = "year",
               limits = c(ymd("2012-06-01"), ymd("2018-12-31")),
               expand = c(0,0)) +
  geom_text(data = filter(tech_stocks, date == "2017-06-02"),
            aes(y = price_indexed, label = company),
            nudge_x = 280)

ggplot(tech_stocks, aes(date, price_indexed, color = ticker)) +
  geom_line() +
  scale_color_OkabeIto(name = "Company", 
                       breaks = c("FB", "GOOG", "MSFT", "AAPL"),
                       labels = c("Facebook", "Alphabet", "Microsoft", "Apple")) +
  scale_x_date(name = "year",
               limits = c(ymd("2012-06-01"), ymd("2018-12-31")),
               expand = c(0,0)) +
  geom_text(data = filter(tech_stocks, date == "2017-06-02"),
            aes(y = price_indexed, label = company),
            nudge_x = 280,
            hjust = 0)

ggplot(tech_stocks, aes(date, price_indexed, color = ticker)) +
  geom_line() +
  scale_color_OkabeIto(name = "Company", 
                       breaks = c("FB", "GOOG", "MSFT", "AAPL"),
                       labels = c("Facebook", "Alphabet", "Microsoft", "Apple")) +
  scale_x_date(name = "year",
               limits = c(ymd("2012-06-01"), ymd("2018-10-31")),
               expand = c(0,0)) +
  geom_text(data = filter(tech_stocks, date == "2017-06-02"),
            aes(y = price_indexed, label = company),
            color = "gray40",
            nudge_x = 20,
            hjust = 0) +
  guides(color = "none")

ggplot(tech_stocks, aes(date, price_indexed, color = ticker)) +
  geom_line() +
  scale_color_OkabeIto(name = "Company", 
                       breaks = c("FB", "GOOG", "MSFT", "AAPL"),
                       labels = c("Facebook", "Alphabet", "Microsoft", "Apple")) +
  scale_x_date(name = "",
               limits = c(ymd("2012-06-01"), ymd("2018-10-31")),
               expand = c(0,0)) +
  scale_y_continuous(name = "Stock Price, Indexed",
                     labels = scales::dollar) +
  geom_text(data = filter(tech_stocks, date == "2017-06-02"),
            aes(y = price_indexed, label = company),
            color = "gray40",
            nudge_x = 20,
            hjust = 0,
            size = 10) +
  guides(color = "none") +
  labs(title = "Tech growth over time",
       caption = "Data from Wilke (2019): Fundamentals of Data Visualization")

Labeling bars

avs <- tech_stocks %>% 
  group_by(company) %>% 
  summarize(stock_av = mean(price_indexed)) %>% 
  ungroup() %>% 
  mutate(share = stock_av / sum(stock_av))
avs

## # A tibble: 4 x 3
##   company    stock_av     share
## * <chr>         <dbl>     <dbl>
## 1 Alphabet  141.0205  0.2292441
## 2 Apple      77.08241 0.1253058
## 3 Facebook  274.7427  0.4466240
## 4 Microsoft 122.3088  0.1988261

Bar plot

ggplot(avs, aes(fct_reorder(company, share), share)) +
  geom_col(fill = "#0072B2")

Horizontal

ggplot(avs, aes(share, fct_reorder(company, share))) +
  geom_col(fill = "#0072B2",
           alpha = 0.9)

ggplot(avs, aes(fct_reorder(company, share), share)) +
  geom_col(fill = "#0072B2",
           alpha = 0.9) +
  coord_flip() +
  theme(panel.grid.major.y = element_blank(),
        panel.grid.minor.x = element_blank(),
        panel.grid.major.x = element_line(color = "gray80"))

Quick aside

Let's actually make a bar plot theme

bp_theme <- function(...) {
  theme_minimal(...) +
    theme(panel.grid.major.y = element_blank(), 
          panel.grid.minor.x = element_blank(), 
          panel.grid.major.x = element_line(color = "gray80"),
          plot.title.position = "plot")
}

ggplot(avs, aes(fct_reorder(company, share), share)) +
  geom_col(fill = "#0072B2",
           alpha = 0.9) +
  geom_text(aes(company, share, label = round(share, 2)),
            nudge_y = 0.02,
            size = 8) +
  coord_flip() +
  bp_theme(base_size = 25)

ggplot(avs, aes(fct_reorder(company, share), share)) +
  geom_col(fill = "#0072B2",
           alpha = 0.9) +
  geom_text(aes(company, share, label = paste0(round(share*100), "%")),
            nudge_y = 0.02,
            size = 8) + 
  coord_flip() +
  scale_y_continuous("Market Share", labels = scales::percent) +
  labs(x = NULL,
       title = "Tech company market control",
       caption = "Data from Clause Wilke Book: Fundamentals of Data Visualizations") +
  bp_theme(base_size = 25)

ggplot(avs, aes(fct_reorder(company, share), share)) +
  geom_col(fill = "#0072B2",
           alpha = 0.9) +
  geom_text(aes(company, share, label = paste0(round(share*100), "%")),
            nudge_y = 0.02,
            size = 8) + 
  coord_flip() +
  scale_y_continuous("Market Share", 
                     labels = scales::percent,
                     expand = c(0, 0, 0.05, 0)) + 
  labs(x = NULL,
       title = "Tech company market control",
       caption = "Data from Clause Wilke Book: Fundamentals of Data Visualizations") +
  bp_theme(base_size = 25)

Last alternative

ggplot(avs, aes(fct_reorder(company, share), share)) +
  geom_col(fill = "#0072B2",
           alpha = 0.9) +
  geom_text(aes(company, share, label = paste0(round(share*100), "%")), 
            nudge_y = -0.02,
            size = 8,
            color = "white") +
  coord_flip() +
  scale_y_continuous("Market Share", 
                     labels = scales::percent,
                     expand = c(0, 0, 0.05, 0)) + 
  labs(x = NULL,
       title = "Tech company market control",
       caption = "Data from Clause Wilke Book: Fundamentals of Data Visualizations") +
  bp_theme(base_size = 25)

Distributions

ggplot(iris, aes(Sepal.Length, fill = Species)) +
  geom_density(alpha = 0.3,
               color = "white")

ggplot(iris, aes(Sepal.Length, fill = Species)) +
  geom_density(alpha = 0.3,
               color = "white") +
  scale_fill_OkabeIto()

Labeling

One method

label_locs <- tibble(Sepal.Length = c(5.45, 6, 7),
                     density = c(1, 0.8, 0.6),
                     Species = c("setosa", "versicolor", "virginica"))
ggplot(iris, aes(Sepal.Length, fill = Species)) +
  geom_density(alpha = 0.3,
               color = "white") +
  scale_fill_OkabeIto() +
  geom_text(aes(label = Species, y = density, color = Species),
            data = label_locs)

ggplot(iris, aes(Sepal.Length, fill = Species)) +
  geom_density(alpha = 0.3,
               color = "white") +
  scale_fill_OkabeIto() +
  scale_color_OkabeIto() +
  geom_text(aes(label = Species, y = density, color = Species),
            data = label_locs) +
  guides(color = "none",
         fill = "none")

label_locs <- tibble(Sepal.Length = c(5.4, 6, 6.9),
                     density = c(1, 0.75, 0.6),
                     Species = c("setosa", "versicolor", "virginica"))
ggplot(iris, aes(Sepal.Length, fill = Species)) +
  geom_density(alpha = 0.3,
               color = "white") +
  scale_fill_OkabeIto() +
  scale_color_OkabeIto() +
  geom_text(aes(label = Species, y = density),
            color = "gray40",
            data = label_locs) +
  guides(fill = "none")

Other options

• Rather than using a new data frame, you could use multiple calls to annotate.

• One is not necessarily better than the other, but I prefer the data frame method

• Keep in mind you can always use multiple data sources within a single plot

  • Each layer can have its own data source
  • Common in geographic data in particular

Annotate example

ggplot(iris, aes(Sepal.Length, fill = Species)) +
  geom_density(alpha = 0.3) +
  scale_fill_OkabeIto() +
  scale_color_OkabeIto() +
  annotate("text", label = "setosa", x = 5.45, y = 1, color = "gray40") +
  annotate("text", label = "versicolor", x = 6, y = 0.8, color = "gray40") +
  annotate("text", label = "virginica", x = 7, y = 0.6, color = "gray40") +
  guides(fill = "none")

Plot text directly

cars <- rownames_to_column(mtcars)
ggplot(cars, aes(hp, mpg)) +
  geom_text(aes(label = rowname))

Repel text

library(ggrepel)
ggplot(cars, aes(hp, mpg)) +
  geom_text_repel(aes(label = rowname))

Slightly better

ggplot(cars, aes(hp, mpg)) +
  geom_point(color = "gray70") +
  geom_text_repel(aes(label = rowname),
                  min.segment.length = 0)

Common use cases

• Label some sample data that makes some theoretical sense (we've seen this before)

• Label outliers

• Label points from a specific group (e.g., similar to highlighting - can be used in conjunction)

Some new data

remotes::install_github("kjhealy/socviz")
library(socviz)

by_country <- organdata %>% 
  group_by(consent_law, country) %>%
  summarize(donors_mean= mean(donors, na.rm = TRUE),
            donors_sd = sd(donors, na.rm = TRUE),
            gdp_mean = mean(gdp, na.rm = TRUE),
            health_mean = mean(health, na.rm = TRUE),
            roads_mean = mean(roads, na.rm = TRUE),
            cerebvas_mean = mean(cerebvas, na.rm = TRUE))

by_country

## # A tibble: 17 x 8
## # Groups:   consent_law [2]
##   consent_law country     donors_mean donors_sd gdp_mean health_mean roads_mean
##   <chr>       <chr>             <dbl>     <dbl>    <dbl>       <dbl>      <dbl>
## 1 Informed    Australia      10.635   1.142808  22178.54    1957.5    104.8757 
## 2 Informed    Canada         13.96667 0.7511607 23711.08    2271.929  109.2601 
## 3 Informed    Denmark        13.09167 1.468121  23722.31    2054.071  101.6363 
## 4 Informed    Germany        13.04167 0.6111960 22163.23    2348.75   112.7887 
## 5 Informed    Ireland        19.79167 2.478437  20824.38    1479.929  117.7742 
## 6 Informed    Netherlands    13.65833 1.551807  23013.15    1992.786   76.09357
## # … with 11 more rows, and 1 more variable: cerebvas_mean <dbl>

Scatterplot

ggplot(by_country, aes(gdp_mean, health_mean)) +
  geom_point()

Outliers

ggplot(by_country, aes(gdp_mean, health_mean)) +
  geom_point() +
  geom_text_repel(data = filter(by_country,
                                gdp_mean > 25000 |
                                gdp_mean < 20000),
                  aes(label = country))

Combine with highlighting

library(gghighlight)
ggplot(by_country, aes(gdp_mean, health_mean)) +
  geom_point() +
  gghighlight(gdp_mean > 25000 | gdp_mean < 20000) +
  geom_text_repel(aes(label = country))

• Notice you only have to specify the points to highlight and geom_text_repel will then only label those points

Combine with highlighting

Switch to make outliers grayed out and labeled

ggplot(by_country, aes(gdp_mean, health_mean)) +
  geom_point() +
  gghighlight(gdp_mean > 20000 & gdp_mean < 25000 ) +
  geom_text_repel(data = filter(by_country, 
                                gdp_mean > 25000 |
                                gdp_mean < 20000),
                  aes(label = country),
                  color = "#BEBEBEB3")

Note I found the exact gray color by looking at the source code. Specifically, it is the output from ggplot2::alpha("grey", 0.7)

By group

ggplot(by_country, aes(gdp_mean, health_mean)) +
  geom_point() +
  geom_text_repel(data = filter(by_country, 
                                consent_law == "Presumed"),
                  aes(label = country))

By group

ggplot(by_country, aes(gdp_mean, health_mean)) +
  geom_point(color = "#DC5265") +
  gghighlight(consent_law == "Presumed") +
  geom_text_repel(aes(label = country),
                  min.segment.length = 0,
                  box.padding = 0.75) +
  labs(title = "GDP and Health",
         subtitle = "Countries with a presumed organ donation consent are highlighted",
         caption = "Data from the General Social Science Survey, Distributed through the socviz R package",
         x = "Mean GDP",
         y = "Mean Health")

Annotating groups of points

Consider using any of the following from ggforce to annotate specific points

• geom_mark_rect()
• geom_mark_circle()
• geom_mark_ellipse()
• geom_mark_hull()

Examples

library(palmerpenguins)
library(ggforce)
penguins %>% 
  drop_na() %>% # Can't take missing data
ggplot(aes(bill_length_mm, bill_depth_mm)) +
  geom_mark_ellipse(aes(group = species, label = species)) +
  geom_point(aes(color = species)) +
  coord_cartesian(xlim = c(28, 62), ylim = c(13, 23)) +
  guides(color = "none")

Limit to a single group

penguins %>% 
  drop_na() %>% 
ggplot(aes(bill_length_mm, bill_depth_mm)) +
  geom_point(aes(color = species)) +
  geom_mark_ellipse(aes(group = species, label = species),
                    data = filter(drop_na(penguins),
                                  species == "Gentoo")) +
  coord_cartesian(xlim = c(28, 62), ylim = c(13, 23))

Switch to hull

Note - requires the concaveman package be installed

penguins %>% 
  drop_na() %>% 
ggplot(aes(bill_length_mm, bill_depth_mm)) +
  geom_point(aes(color = species)) +
  geom_mark_hull(aes(group = species, label = species),
                    data = filter(drop_na(penguins),
                                  species == "Gentoo")) +
  coord_cartesian(xlim = c(28, 62), ylim = c(13, 23))

Change expand

penguins %>% 
  drop_na() %>% 
ggplot(aes(bill_length_mm, bill_depth_mm)) +
  geom_point(aes(color = species)) +
  geom_mark_hull(aes(group = species, label = species),
                 expand = unit(1, "mm"),
                 data = filter(drop_na(penguins), 
                               species == "Gentoo")) + 
  coord_cartesian(xlim = c(28, 62), ylim = c(13, 23))

More in-depth annotations

First create a description

penguins <- penguins %>% 
  mutate(desc = ifelse(species != "Gentoo", "", "During deep dives, gentoo penguins reduce their heart rate from 80 to 100 beats per minute (bpm) down to 20 bpm. Gentoo penguins use nesting materials ranging from pebbles and molted feathers in Antarctica to vegetation on subantarctic islands. Gentoos are the third largest penguin, following the emperor and king."))

Now add as a description

penguins %>% 
  drop_na() %>% 
ggplot(aes(bill_length_mm, bill_depth_mm)) +
  geom_point(aes(color = species)) +
  geom_mark_ellipse(aes(group = species, 
                   label = species,
                   description = desc),
               data = filter(drop_na(penguins), 
                             species == "Gentoo"),
               label.fill = "#b3cfff") +
  coord_cartesian(xlim = c(28, 62), ylim = c(13, 23))

Similar

We can also just add a textbox through {ggtext}

txtbox <- tibble(
  bill_length_mm = 23,
  bill_depth_mm = 16,
  lab = '"They may all waddle around in their tuxedolike feathers, but the penguins of the Antarctic Peninsula are not equal in their ability to adapt to a warming climate. While the populations of the Adélie and chinstrap penguin species are currently declining, the gentoo species is increasing. But this has not always been the case, according to a recent study published in the journal Scientific Reports." - Scientific American'
)

penguins %>% 
  drop_na() %>% 
ggplot(aes(bill_length_mm, bill_depth_mm)) +
  geom_point(aes(color = species)) +
  ggtext::geom_textbox(aes(label = lab),
                       data = txtbox) +
  coord_cartesian(xlim = c(17, 62), ylim = c(13, 22))

Last bit

The ggforce package is well worth exploring more.

See here for a nice walkthrough that has good data viz and uses some of the ggforce functions (as well as illustrating a few other cool packages)

ggthemes

• Good place to start. All sorts of themes.
• Includes color scales, etc., that align with themes
• You can even conform with other software
  • fit into an economics conference with theme_stata

See the themes here

BBC

The BBC uses ggplot for most of its graphics. They've developed a package with a theme and some functions to help make it match their style more.

See the repo here

Their Journalism Cookbook is really nice too

ggthemeassist

• Another great place to start with making major modifications/creating your own custom theme
• Can't do everything, but can do a lot
• See here

[demo]

theme() for everything else

• You can basically change your plot to look however you want through theme
• Generally a bit more complicated
• I've used ggplot for years and only really now gaining fluency with it

Quick example

From Lab 3

library(fivethirtyeight)
g <- google_trends %>% 
  pivot_longer(starts_with("hurricane"), 
               names_to = "hurricane", 
               values_to = "interest",
               names_pattern = "_(.+)_")
landfall <- tibble(date = lubridate::mdy(c("August 25, 2017", 
                                           "September 10, 2017", 
                                           "September 20, 2017")),
                   hurricane = c("Harvey Landfall", 
                                 "Irma Landfall", 
                                 "Maria Landfall"))

p <- ggplot(g, aes(date, interest)) +
  geom_ribbon(aes(fill = hurricane, ymin = 0, ymax = interest),
              alpha = 0.6) + 
  geom_vline(aes(xintercept = date), landfall,
             color = "gray80", 
             lty = "dashed") +
  geom_text(aes(x = date, y = 80, label = hurricane), landfall,
            color = "gray80",
            nudge_x = 0.5, 
            hjust = 0) +
  labs(x = "", 
       y = "Google Trends",
       title = "Hurricane Google trends over time",
       caption = "Source: https://github.com/fivethirtyeight/data/tree/master/puerto-rico-media") + 
  scale_fill_brewer("Hurricane", palette = "Set2")

p + theme(panel.grid.major = element_line(colour = "gray30"), 
          panel.grid.minor = element_line(colour = "gray30"), 
          axis.text = element_text(colour = "gray80"), 
          axis.text.x = element_text(colour = "gray80"), 
          axis.text.y = element_text(colour = "gray80"),
          axis.title = element_text(colour = "gray80"),
          legend.text = element_text(colour = "gray80"), 
          legend.title = element_text(colour = "gray80"), 
          panel.background = element_rect(fill = "gray10"), 
          plot.background = element_rect(fill = "gray10"), 
          legend.background = element_rect(fill = NA, color = NA), 
          legend.position = c(0.20, -0.1), 
          legend.direction = "horizontal",
          plot.margin = margin(10, 10, b = 20, 10),
          plot.caption = element_text(colour = "gray80", vjust = 1), 
          plot.title = element_text(colour = "gray80"))