ggplot2 Tutorial: Components, Layers, and Examples

1. Learn to deconstruct an image
2. Learn to understand how to read ggplot code
3. Understand how ggplot constructs your code into a figure

• R (DSL): brew install homebrew/science/r
• RStudio (IDE): brew install Caskroom/cask/rstudio
• Assumes you have homebrew installed (if you don't, you should!)
• ggplot2 (library) and HSAUR3 (library)

install.packages("ggplot2")
install.packages("HSAUR3")

A data frame with 53 observations on the following 3 variables.

• logdose = the logarithm (base 10) of the dose of drug in milligrams
• bloodp = average systolic blood pressure achieved while the drug was being administered
• recovtime = time in minutes before the patient's systolic blood pressure returned to 100mm of mercury

require(HSAUR3)
head(bp)

  logdose bloodp recovtime
1    2.26     66         7
2    1.81     52        10
3    1.78     72        18
4    1.54     67        NA
5    2.06     69        10
6    1.74     71        13

When your data is in the right data structure, summarizing your data is cake!:

summary(bp)

    logdose          bloodp        recovtime   
 Min.   :1.180   Min.   :51.00   Min.   : 4.0  
 1st Qu.:1.740   1st Qu.:60.00   1st Qu.:10.5  
 Median :1.900   Median :67.00   Median :21.0  
 Mean   :1.992   Mean   :66.34   Mean   :22.4  
 3rd Qu.:2.260   3rd Qu.:69.00   3rd Qu.:26.5  
 Max.   :2.780   Max.   :88.00   Max.   :72.0  
                                 NA's   :10    

In R, graphing data is easy!

require(HSAUR3)
plot(bp)

• Default graphics in R is easy, but less customizable
• Default graphics in R is easy, but figures can be made more domain specific
• Default graphics in R is easy, but arguably less aesthetic

ggplot2 is a visualization library based on the Grammar of Graphics (by Leland Wilkinson, who is now at Tableau).

In short, it is the hypothesis that all graphics can be composed from the same material parts.

• Geoms (used to represent data points)
• Aesthetics (used to customize the “look” of data points)
• Stats (used to transform data)
• Scales (controls how data values relate to visual values)
• Coordinate System (coordinate plane)
• Position Adjustments (arrangement of geoms)
• Labels
• Facets (dividing a plot into subplots)
• Legends
• Zooming
• Themes (color schemes, University of Washington colors, etc.)

• Geoms (used to represent data points)
• Aesthetics (used to customize the “look” of data points)
• Stats (used to transform data)
• Scales (controls how data values relate to visual values)
• Coordinate System (coordinate plane)
• Position Adjustments (arrangement of geoms)
• Labels
• Facets (dividing a plot into subplots)
• Legends
• Zooming
• Themes (color schemes, University of Washington colors, etc.)

require(ggplot2)
ggplot(bp, aes(logdose, bloodp)) +
  geom_point() +
  coord_cartesian()

require(ggplot2)
ggplot(bp, aes(logdose, bloodp)) +
  geom_point() +
  coord_cartesian()

• “require” loads the ggplot2 library. You can also use “library”
• “bp” refers to the dataset (it must be a data frame)
• “aes” refers to aesthetics (what we attach visual encodings to)
• “+” tells ggplot2 that we are about to add a layer
• “geom_point()” is a function that adds a points layer
• “coord_cartesian()” is a function that adds a coordinate layer
• We will refer back to this “stub” code several times later in the presentation

What happens if we just run

ggplot(bp, aes(logdose, bloodp))

without any layers?:

What happens if we just run

ggplot(bp, aes(logdose, bloodp))

without any layers?:

  Error: No layers in plot
  Execution halted

1. ggplot2 operates via layers, and needs at least one to display something
2. layers are additive (and there are defaults)
3. ggplot2 requires datasets to be data frames
4. foundational idea: every figure can be composed using a combination of core components
5. visual encoding assignments happen in aes, but encoding details are provided in layers

Previous code:

  ggplot(bp, aes(logdose, bloodp)) +
  geom_point() +
  coord_cartesian()

If you deconstruct the plot, what is the main difference between the original plot and this one? The coordinate system!

  ggplot(bp, aes(logdose, bloodp)) +
  geom_point() +
  coord_polar()

  logdose bloodp recovtime
1    2.26     66         7
2    1.81     52        10
3    1.78     72        18
4    1.54     67        NA
5    2.06     69        10
6    1.74     71        13

ggplot(bp, 
 aes(logdose, bloodp, color=recovtime)
) +
geom_point() +
scale_color_continuous(
  low="#FFFFFF", 
  high="#990000"
)

What do you see? Can anyone describe how they might create this plot?

  logdose bloodp recovtime
1    2.26     66         7
2    1.81     52        10

ggplot(bp, 
 aes(logdose, bloodp, color=recovtime, size=recovtime)
) +
geom_point() +
scale_color_continuous(
  low="#FFFFFF", 
  high="#990000"
)

I don't think you'd want to map shapes in this way. I've demonstrated this solely as an excuse to use the shape encoding!

ggplot(bp, 
 aes(logdose, bloodp, color=recovtime, shape=factor(bp$recovtime > 50), size=recovtime)
) +
geom_point() +
scale_color_continuous(
  low="#FFFFFF", 
  high="#990000"
)

Can anyone describe how they might create this plot? What do you see?

ggplot(bp, 
 aes(logdose, bloodp, color=recovtime, size=recovtime)
) +
geom_point() +
geom_smooth(method=lm) +
theme(legend.position="none") +
scale_color_continuous(
  low="#FFFFFF", 
  high="#990000"
)

ggplot(bp, 
  aes(logdose, bloodp)
) +
geom_point() +
coord_cartesian() +
ggtitle("log10-transformed hypotensive agent dosage \npositively correlated with\naverage systolic blood pressure (while drug was being administered)") +
theme(plot.title = element_text(
  lineheight=.8, 
  face="bold")
)

• The examples we just walked through were designed to illustrate how to call and use layers in ggplot2
• Double and triple encoding the same groups of data can facilitate improved perceptual performance. However, there are diminishing returns and figure can end up looking “busy”.

How might we make a histogram?

ggplot(bp, 
  aes(logdose)
) +
geom_histogram(
  binwidth= 0.5
)

How might we make a histogram?

ggplot(bp, 
  aes(logdose, 
      fill=cut(
          bp$logdose, 
          breaks = c(1,1.5,2,2.5,3)
        )
    )
) +
geom_histogram(
  binwidth= 0.5
)

ggplot(bp, 
  aes(logdose, 
      fill=cut(
        bp$logdose, 
        breaks = c(1,1.5,2,2.5,3)
       )
    )
) +
geom_histogram(
  binwidth= 0.5
) +
coord_polar()

• A Layered Grammar of Graphics
• RStudio Cheat Sheets
• ColorBrewer
• Advanced R