1 Review

Base R Visualization. For an introduction to data visualization in R, see Visualization in Base R.

Introduction to ggplot2. To view the first part of this series on visualization in ggplot2, see Visualization with ggplot2: Part I.


1.1 Installing & Loading

Installing Packages. Packages, or libraries, extend R’s functionality.

  • Install packages with function install.packages() and the package name in quotes ""
  • If you already have ggplot2 installed, you do not need to reinstall it, just load it
  • The following command will install package ggplot2


install.packages("ggplot2")


Additional Packages. Because we’ll focus on graphical themes in this session:

  • We’ll also install package ggthemes and GGally to enhance the ggplot2 package
  • Like ggplot2, skip this if ggthemes is already installed
  • Package ggthemes adds a number of new premade themes to ggplot2
  • Explore dozens of new themes at All Your Figure Are Belong to Us
  • Remember that R is case-sensitive


install.packages("ggthemes")
install.packages("GGally")


Loading Packages. Once installed, packages must be loaded into your session:

  • You can load packages individually using function library()
  • This time, quotes ("") are optional for package names
  • Remember that R is case-sensitive


library(ggplot2)
library(ggthemes)
library(GGally)


1.2 The Grammar of Graphics

Visualization Shorthand. Like human language and parts of speech, e.g. verbs, nouns, adjectives, etc.:

  • Data visualization is also comprised of “parts” called layers
  • See Leland Wilkinson’s work The Grammar of Graphics (1999)
  • “Grammar of Graphics” is the gg in ggplot2
  • ggplot2 was designed by Hadley Wickham using this framework
  • See Wickham’s A Layered Grammar of Graphics (2010)


Layers. “Layers” are the equivalent of parts of speech in human language; there are 7 in total:

  • Data. Your dataset and variables of interest (ggplot())
  • Aesthetics. How you map variables, e.g. x- and y-axis, color, size, etc. (aes())
  • Geometries. The shape of your visualization, e.g. scatter plots, bar plots, etc. (geom_*())
  • Facets. Grouping variables and displaying each over many visualizations (facet_*())
  • Statistics. Statistical elements, such as bins and linear regression (stat_*())
  • Coordinates. Zooming in or changing height and width (coord_*())
  • Themes. Stylizing and polishing plots (theme_*())


Essential Layers. Three layers are essential to any ggplot2 visualization:

  • Data. Because you need some data to visualize them
  • Aesthetics. Because you want to show specific variables from your data
  • Geometries. Because your variables must assume a form


Observe. Here we use the practice dataset diamonds, which we call with function data():

  • Note that we connect ggplot2 functions with the addition operator, or +
  • The diamonds dataset is built into ggplot2
  • Optional: Load diamonds with function data()


ggplot(diamonds) +                # Data
  aes(x = carat, y = price) +     # Aesthetics, here x- and y-axes
  geom_point()                    # Geometries, here a scatterplot


“Proper” Syntax. Though the above is easier, function aes() should be nested in ggplot()

ggplot(diamonds,             # Separate arguments with commas
       aes(x = carat, 
           y = price)) +     # Double end parantheses ends both ggplot() and aes()
  geom_point()


1.3 Data vs. Non-Data Ink

Data Ink. The aesthetics layer is only used to show data ink, in function aes()

  • “Data Ink” refers to any element of a page that depicts your data
  • “Data Ink” is dynamic, so when your data changes, your ink changes
  • Arguments include x =, y =, color =, fill =, size =, etc.
  • Also called aesthetic mappings


ggplot(diamonds,
       aes(x = carat, 
           y = price,
           color = clarity)) +     # Argument "color =" shows variable "clarity"
  geom_point()


Non-Data Ink. The geometries layer also shows non-data ink in geom_*() functions

  • “Non-data Ink” does not depict data, but makes graphics easier to interpret
  • “Non-data Ink” is not dynamic, so it stays the same if your data change
  • Except for x = and y =, all data ink arguments are the same
  • Also called attributes


ggplot(diamonds,
       aes(x = carat, 
           y = price)) +
  geom_point(alpha = 0.075,        # Argument "alpha =" makes data transparent (7.5%)
             color = "tomato")     # Argument "color =" makes your data pop


Conclusion. Combine both data and non-data ink for optimal effect.


ggplot(diamonds,
       aes(x = carat, 
           y = price,
           color = clarity)) +
  geom_point(alpha = 0.35)        # Argument "alpha =" makes data transparent (7.5%)


2 Geometries

There are 60+ geometry functions in package ggplot2. They all begin with geom_.


2.1 Proper Form

What’s Good? Over time, which geometries to use becomes intuitive.

  • Selecting the right geometry begins with one question: “What’s the big idea?”
  • If visualizing for a colleague or collaborator, always consult for context
    • “I want to show revenue over time” indicates a line chart
    • “I want to show the difference in income by ethnicity” indicates a bar chart
    • “I want to that most of our clients are young folks” indicates a histogram
    • “I want to show a pie chart with all 23 flavors in Dr. Pepper” indicates inexperience
  • If you need help, consider visiting “Input” in Data Viz Project


Scatter Plots measure two continuous (numeric) variables, like length and width:

  • Use function geom_point() for a typical scatter plot
  • If many overlapping points, add random noise with geom_jitter()


data(iris)
ggplot(iris, 
       aes(x = Petal.Length,       # Length in centimeters
           y = Petal.Width,        # Width in centimeters
           color = Species)) +
  geom_point()

ggplot(iris, 
       aes(x = Petal.Length,       # Length in centimeters
           y = Petal.Width,        # Width in centimeters
           color = Species)) +
  geom_jitter()


Bar Plots measure one continuous (numeric) and one categorical variable:

  • Note that geom_bar() needs argument stat = to tell R how to plot your geometry
  • Calling options(scipen = 999) disables scientific notation (orders of magnitude)


options(scipen = 999)             # Disable scientific notation


ggplot(diamonds, 
       aes(x = cut,               # Categorical (Quality of Cut)
           y = price)) +          # Continuous (US Dollars)
  geom_bar(stat = "identity",     # Argument "stat =" tells R how to handle stats
           color = "skyblue")


Histograms visualize distributions of a single continuous variable, like mpg:

  • Function geom_histogram() automatically chooses bin widths
  • Histograms show total amount of observations in each range of values
  • Bin widths control how big or small each value range is
  • Note that fill = is needed for areas and color = is for lines and points


data(diamonds)
ggplot(diamonds,
       aes(x = carat)) +
  geom_histogram(fill = "dodgerblue3")     # Argument "fill =" needed for areas


Line Graphs generally visualize changes in data over time.

  • There are several types of line graphs in ggplot2
    • geom_line() seems obvious, but it makes the line touch the x-axis between values
    • geom_step() is an improvement, making horizontal lines between values
    • geom_path() is often ideal, directly connecting each data point with a line
  • Practice dataset economics already has dates and number of unemployed, in thousands


data(economics)
ggplot(economics, 
       aes(x = date,
           y = unemploy)) +
  geom_path()


Conclusion. Choose the right geometries for the variables you want to show.

Then make it pretty. But never too pretty that it obfuscates the clarity of your ideas.


“His clothes were all new and had been cut by a good Moscow tailor. But there was something wrong even with his clothes. They were rather too fashionable, as clothes always are from conscientious but not very talented tailors.”

Fyodor Dostoevsky, “The Idiot”, Part II, Chapter 2



2.2 Factor This In

Factors in R are the same as categorical variables, a.k.a. “nominal”, “discrete”, etc.

  • This is a source of frustration, since you sometimes must tell R explicitly
  • Use function as.factor() on a variable to tell R that it is categorical
  • Question. Why is this an issue?
  • Another Question. How many cylinders are in cars, typically?


data(mtcars)
ggplot(mtcars, aes(x = cyl,        # Number of cylinders
                   y = mpg)) +     # Miles per gallon 
  geom_point()


What the heck? Where are all the 5- and 7-cylinder cars?

  • Right, the 32 cars in dataset mtcars only have 4-, 6-, and 8-cylinder engines
  • R hasn’t been told that variable cyl (“cylinders”) is categorical
  • Let’s try again with function as.factor()


data(mtcars)
ggplot(mtcars, aes(x = as.factor(cyl),     # Cylinders "coerced" (changed) to factor
                   y = mpg)) +
  geom_point()


Better. But we can add some random noise by using the geometry geom_jitter():

  • Argument width = and height = controls degree of random noise


data(mtcars)
ggplot(mtcars, aes(x = as.factor(cyl),     # Still a factor
                   y = mpg)) +
  geom_jitter(width = 0.05)                # Jittered slightly, but "mpg" still precise


Right.


3 Preset Themes

Because we’re cutting it close on time, let’s cover premade themes quickly.

  • The themes layer largely concerns style and non-data ink
  • All themes functions begin with theme_
  • Function theme() allows for total customization
  • Function theme_set() allows for global themes and will apply to all plots
    • More on this at a later time, we just had to cover geometries
  • Simply add your preferred theme with the + operator
  • There are dozens of premade themes, and dozens more with package ggthemes
    • Type theme_ in the console and scroll through the autocomplete dropdown


Try theme_light():

library(ggthemes)
ggplot(diamonds,
       aes(x = cut,
           y = price)) +
  geom_jitter(width = 0.25,
              alpha = 0.05,
              color = "tomato") +
  theme_light()                       # Just add the theme layer onto the graph


Or theme_fivethirtyeight():

library(ggthemes)
ggplot(diamonds,
       aes(x = cut,
           y = price)) +
  geom_jitter(width = 0.25,
              alpha = 0.1,
              color = "tomato") +
  theme_fivethirtyeight()


Or theme_minimal():

library(ggthemes)
ggplot(diamonds,
       aes(x = cut,
           y = price)) +
  geom_jitter(width = 0.25,
              alpha = 0.05,
              color = "tomato") +
  theme_classic()


4 Practice: Labeling & Themes

Instructions. Take any of the above plots and use the + operator to add function labs().

  • In labs(), you can set graph labels for the following arguments:
    • title =
    • subtitle =
    • caption =
    • x =
    • y =
    • Make sure to put your text in quotations!
  • Select any dataset, between mtcars, diamonds, economics, or iris
  • Use function names() with the dataset name inside () to get the variable names
    • Alternatively, use function help() with the dataset for definitions
  • Select data ink for function aes(), a geometry, and non-data ink inside geom_*()
    • Alternatively, reuse an existing plot for convenience
  • Browse and select a theme you like using any theme_*() function
  • Make sure your data and non-data ink convey your finding clearly; polish appropriately


Thanks, everyone.