Foundations of {ggplot2}

Eric R. Scott

2024-06-06

Learning Objectives

• Understand the “grammar of graphics” and how it inspired ggplot2’s design
• Know what it means to “map data to aesthetics”
• Learn to plot different data sets on the same visualization
• Understand the relationship between a “geom” and a “stat”
• Learn to customize scales and guides

A Grammar of Graphics

• A book by Wilkinson (2005)
• Inspired the development of many graphics programs including ggplot2 (Wickham 2010)
• Alternative to having a function for every kind of plot
• A framework for layering elements to create any kind of plot

Grammar of Graphics Components

1. Data
2. Aesthetics
3. Scales (and guides)
4. Geometric objects
5. Statistics
6. Facets
7. Coordinate system

We’re going to go through each of these and examine how they appear in this example plot

Data

Observations and variables to be visualized.

What data is being visualized?

• Body mass measured in grams

• Bill length measured in mm

• Island and species categories

Aesthetics

Visual elements (color, shape, position, size, etc.) used to encode data.

What aesthetics are used to encode these data?

• Color

• x position

• y position

Scales (and guides)

Scales translate data units into visual units, guides translate visual units back to data units.

What scales are used?

• Continuous and linear x and y axes (scales)

• Discrete (categorical) color scale

Scale translates data units (e.g. grams) into visual units (pixels). Guides translate visual units (e.g. color hue) back to data units (species category). Walk through this with annotations.

Geometric Objects (“geoms”)

Objects, often having multiple aesthetics, that represent data visually.

What geometric objects are used?

• Data is represented using circles/points
• Trend is represented as a line

Both aesthetics and geoms are related to the visual representation of data. To distinguish aesthetics and geoms, think about how geoms can have many aesthetics. E.g. points have position, shape, size, color, and transparency and different data could be mapped to each of those aesthetics.

Statistics (“stats”)

Any calculations or transformations applied to the data in order to plot it.

What “stats” are used?

• For points, none (stat = “identity”)

• For trend lines, linear regression

If the data are 344 observations, are any transformations needed to draw one of the two “geoms” in the plot?

Facets

Plots can be split into small multiples or “facets” by a variable.

What is the faceting variable?

• Faceted by island

Not every plot has facets. Plots can be faceted by multiple (2) variables in a grid

Coordinate System

How spatial positions are represented on paper (or screen)—e.g. map projections.

What coordinate system is used?

• Cartesian coordinates

Other options include polar coordinates or the many possible coordinate systems used in maps.

Practice

Identify each of the seven components of this plot

1. Data
2. Aesthetics
3. Scales
4. Geometric Objects
5. Statistics
6. Facets
7. Coordinate System

Break ⏰

10:00

ggplot2 and the Grammar of Graphics

Now we’re going to go through each of the seven components and see some examples of how they are implemented in ggplot2

Data

• Data is inherited from ggplot() by all layers, but can be overridden for specific layers

• Worked example: jitter plot of raw data with mean ± standard deviation

library(tidyverse)
library(palmerpenguins)

#summarize dataset
peng_summary <- 
  penguins |> 
  group_by(island) |> 
  summarize(
    mean_mass = mean(body_mass_g, na.rm = TRUE),
    lower_sd = mean_mass - sd(body_mass_g, na.rm = TRUE),
    upper_sd = mean_mass + sd(body_mass_g, na.rm = TRUE)
  )

ggplot(peng_summary, aes(x = island, y = mean_mass)) +
  #mean
  geom_point(shape = "square", color = "blue", size = 2.5) +
  #sd
  geom_errorbar(
    data = peng_summary,
    aes(y = mean_mass, ymin = lower_sd, ymax = upper_sd),
    width = 0.1,
    color = "blue"
  ) +
  #add raw data:
  geom_jitter(
    data = penguins,
    aes(y = body_mass_g),
    alpha = 0.4,
    height = 0
  )

Start by summarizing the penguins dataset to get some summary statistics for the error bars. Notice that ggplot() uses the raw data though.

Then add the raw data points, in this case with geom_jitter()

Then add the geom_pointrange(). Notice that we override the inherited data with data = peng_summary . The x mapping is inherited, but we need to supply new aesthetic mappings for y, ymin, and ymax

Aesthetics

• Aesthetics are inherited when placed in ggplot() but can also be specified per layer
• Aesthetics can be mapped to data or set as constant
• All the aesthetics and their possible values: https://ggplot2.tidyverse.org/articles/ggplot2-specs.html
• Worked example: box plot for each island layered ontop of jitter plot for each island x sex combination.

This might be a good place to look through help files and figure out which aesthetics are valid for mapping

Aesthetic mappings supplied to ggplot() are inherited, aesthetic mappings supplied to a geom only affect that geom.

ggplot(penguins, aes(x = island, y = body_mass_g)) +
  geom_boxplot() +
  geom_jitter(aes(color = sex))

Scales

• scale_ functions affect the range (limits) and breaks of scales and the labels and appearance of corresponding guides.
• Worked example: 1) re-order and re-color a discrete color scale, 2) change the number of breaks on the axes

p + 
  scale_color_manual(
    values = c(
      "Adelie" = "#7570b3",
      "Chinstrap" = "#d95f02",
      "Gentoo" = "#1b9e77"
    ),
    breaks = c("Gentoo", "Chinstrap", "Adelie")
  ) +
  scale_x_continuous(
    n.breaks = 10
  ) +
  scale_y_continuous(
    breaks = seq(from = 30, to = 65, by = 5.5)
  )

values takes a named vector and is used to assign a color (hex code or color name) to each level of the variable. breaks changes the order.

Geoms

• Every geom has a default stat, but it can be overridden
• Not all geoms use the same aesthetics
• Worked example: explore the anatomy of a help file (e.g. ?geom_point())
  • How can you determine the default stat for a geom?
  • How can you find out what aesthetics a geom has?

df <- expand_grid(x = LETTERS[1:5], y = 1:5)
ggplot(df) +
  geom_point(aes(x = x, y = y, 
        color = x, shape = x,
        size = y, alpha = y, stroke = y))

Caution

With great power, comes great responsibility! It’s not always a good idea to map data to aesthetics just because you can. Stay tuned for part 2 of this series for more!

Stats

• Every stat has a default geom and you usually use the geom_*() function
• Some stats calculate multiple values that are available with after_stat()
• Worked examples:
  1. Adding mean ± SD on top of jitter plot with stat_summary()
  2. Making a binned density plot with geom_histogram() and after_stat()

stat_summary()

ggplot(penguins, aes(x = island, y = body_mass_g)) +
  geom_jitter(alpha = 0.4, height = 0) +
  stat_summary(fun.data = mean_sdl, fun.args = list(mult = 1),
               color = "blue", shape = "square")

Binned density plot with geom_histogram() and after_stat()

ggplot(penguins) +
  geom_histogram(aes(x = body_mass_g, y = after_stat(density))) +
  facet_wrap(vars(island))

Facets

• Implemented with facet_wrap() and facet_grid()
• Facets will be explored more in part 2 of this series

Coords

• Adjusting x and y limits with coord_cartesian() is different than adjusting limits in a scale
• coord_polar() for polar data, coord_sf() for maps
• Worked example: zooming in on data

Setting axis limits in scale_x_continuous() removes data that is out of range

p + scale_x_continuous(limits = c(4000, 5000))

`geom_smooth()` using formula = 'y ~ x'

Warning: Removed 222 rows containing non-finite outside the scale range
(`stat_smooth()`).

Warning: Removed 222 rows containing missing values or values outside the scale range
(`geom_point()`).

Importantly, notice that the best-fit line from geom_smooth() is now only considering the points between 4000 and 5000 g body mass!

Setting axis limits in coord_cartesian() simply zooms in

p + coord_cartesian(xlim = c(4000, 5000))

`geom_smooth()` using formula = 'y ~ x'

Notice that points whose centers are < 4000 or > 5000 still show up

Resources

• ggplot2: Elegant Graphics for Data Analysis (3e)
• The R Graph Gallery
• ggplot2 documentation
• CCT Data Science drop-in hours (Tuesdays 9-10am)
• Data Science Incubator

References

Wickham, Hadley. 2010. “A Layered Grammar of Graphics.” Journal of Computational and Graphical Statistics 19 (1): 3–28. https://doi.org/10.1198/jcgs.2009.07098.

Wilkinson, Leland. 2005. The Grammar of Graphics. 2nd ed. Statistics and Computing. New York: Springer-Verlag. https://doi.org/10.1007/0-387-28695-0.