Learning geom

This does not work…

ggplot(data=mtcars2, aes(x=hp, y=mpg)) 

… because we did not specify a geometry. Because we have defined two continuous variables in aes(), we want to use a scatterplot, so we add geom_point with a +:

ggplot(data=mtcars2, aes(x=hp, y=mpg)) +
  geom_point() 

ggplot2 is object-oriented. This is useful when we build up plots:

ggp <- ggplot(data=mtcars2, aes(x=hp, y=mpg)) +
          geom_point() 
ggp

Now we can add a smoothed trendline using geom_smooth:

ggp + geom_smooth() 

## `geom_smooth()` using method = 'loess' and formula 'y ~ x'

Notice the default is to provide a smoothed loess curve for < 1000 observations surrounded by a shaded 95% confidence interval. ?geom_smooth describes many options, including a method argument to define the type of trendline calculation and se for whether one wants to turn the shaded area off:

ggp + geom_smooth(method = 'lm', se = FALSE) 

Component placement

Note that both data and aes() can be given in either the main ggplot() call or in the geom_. Furthermore, data can be ‘piped’ in to ggplot() via %>%. Note that data= must be used inside a geom_ but can be omitted inside ggplot().¹

These produce identical results:

# 1
  ggplot(mtcars2, aes(x=hp, y=mpg)) + 
    geom_point()
# 2
  ggplot() + 
    geom_point(data=mtcars2, aes(x=hp, y=mpg))
# 3
  ggplot(mtcars2) + 
    geom_point(aes(x=hp, y=mpg))
# 4
  mtcars2 %>%
    ggplot() + 
      geom_point(aes(x=hp, y=mpg))

Adding variables to aes()

Adding additional information via the points is easy with ggplot: we simply add any additional variables in the dataset to aes() to specify the color and/or shape of each point.

First, we can represent the number of cylinders in each car’s engine via point shape; notice how ggplot automatically creates and adds a legend:

ggplot(mtcars2, 
       aes(x=hp, y=mpg, 
           shape=cyl)) +
    geom_point()

Perhaps even more visually appealing is separating the engines by color, as well:

ggplot(mtcars2, 
       aes(x=hp, y=mpg, 
           color=cyl,
           shape=cyl)) +
    geom_point()

But maybe we’d prefer to convey another variable with one of those aesthetics:

ggplot(mtcars2, 
       aes(x=hp, y=mpg, 
           color=cyl,
           shape=transmission)) +
    geom_point()

This scatterplot now conveys four bits of information–engine power & cylinder number, fuel economy, and transmission type–with the same number of points (and thus essentially the same amount of ink) as the first scatterplot.

Note the critical difference between these sets of plots:

ggplot(mtcars2) +
 geom_point(aes(x=hp, y=mpg,
                shape = cyl,
                color = cyl)) 

ggplot(mtcars2) +
 geom_point(aes(x=hp, y=mpg, 
                shape = cyl), 
                color = 'blue') 

To change a component of the plot according to a variable in the dataset, the command must be included in aes(). Otherwise, changing a component will apply across the entire geom. Pay careful attention to properly closing parentheses in ggplot, especially for aes() within a geom.

Using facets

Adding additional information to a single graph pane can be messy. Fortunately, ggplot gives us several options to arrange our data in multiple panes, called facets, according to one or more (categorical) variables.

The simplest is facet_wrap, which just creates a unique plot window for data in each level of a categorical variable:

ggplot(mtcars2, 
       aes(x=hp, y=mpg, 
           color=cyl)) +
    geom_point() +
    facet_wrap(~transmission)

facet_grid plots levels of one variable against another:

ggplot(mtcars2, 
       aes(x=hp, y=mpg, 
           color=cyl)) +
    geom_point() +
    facet_grid(origin~transmission)

Note there is only one manual transmission model made in the USA in this snapshot of a few automobiles in 1974.

With panes and colors taking care of categorical variables, we can add a continuous variable, wt, to show the mass of cars via the relative size of the plotted points; i.e. larger points are heavier cars:

ggplot(mtcars2, 
       aes(x=hp, y=mpg, 
           color=cyl,
           size=wt)) +
    geom_point() +
    facet_grid(origin~transmission)

At this point we have six variables from the mtcars2 dataset in one plot with very little code.

scale_ settings

As we add more information, it is important that we tend to the clarity of the presentation. The appearance of each variable in aes() can be modified via scale_ options. First, we help distinguish between overlapping points with outlines by manually specifying shape to pull values from the upper end of ?points, which allow one to define colors for both outline and fill:

 ggplot(mtcars2, 
        aes(x=hp, y=mpg, 
            color=cyl, 
            shape=cyl,
            size=wt)) +
   geom_point() 
 
 ggplot(mtcars2, 
        aes(x=hp, y=mpg, 
            fill=cyl, 
            shape=cyl,
            size=wt)) +
   geom_point() +
   scale_shape_manual(values=c(21,22,24)) + 
   scale_size_continuous(guide=FALSE)

Let’s also shut off the size legend and add a caption to indicate that point size simply scales with vehicle weight:

 ggplot(mtcars2, 
        aes(x=hp, y=mpg, 
            fill=cyl, 
            shape=cyl,
            size=wt)) +
   geom_point() + 
   scale_shape_manual(values=c(21,22,24)) + 
   scale_size_continuous(guide=FALSE) +
   labs(caption = 'Point sizes scaled to show relative car weights')

There are actually more obvious graph components controlled by adding labs(); let’s make the axis labels more specific:

 ggplot(mtcars2, 
        aes(x=hp, y=mpg, 
            fill=cyl, 
            shape=cyl,
            size=wt)) +
   geom_point() +
   scale_shape_manual(values=c(21,22,24)) + 
   scale_size_continuous(guide=FALSE) +
   labs(x="Engine power (horsepower)",
        y="Fuel economy (miles/gallon)", 
       caption = 'Point sizes scaled to show relative car weights')

Adjusting themes

There are two uses of “theme” in ggplot-speak. First, there are a huge number of settings that relate to the size, orientation, and justification of almost every plot component (see ?theme). We’ll play with just two here to enlarge a couple bits of text:

ggplot(data=mtcars2, 
      aes(x=hp, y=mpg, 
          fill=cyl, 
          shape=cyl,
          size=wt)) +
 geom_point() +
 scale_shape_manual(values=c(21,22,24)) + 
 scale_size_continuous(guide=FALSE) +
 theme(axis.title = element_text(size=14), 
       axis.text = element_text(size=12))

Next, there are some shortcuts for various theme settings that are added via theme_. These calls generally modify the default theme and have two parts. Firstly, what comes after theme_, such as theme_bw or theme_map, must be defined via (a) ggplot2 defaults, (b) a third-party loaded library, or (c) a user-defined object. Secondly, a numeral in the parentheses, like (16), is a general modifier to the base font size of the theme. It is a shortcut for adjusting the size of all text proportionally.

Let’s reuse the earlier ggp object and review themes:

ggp + theme_bw() # No background, major + minor gridlines
ggp + theme_linedraw() # No background, no minor gridlines
ggp + theme_minimal() # No axis lines 

Here’s a final version of our information-maximized graph:

ggplot(mtcars2, 
      aes(x=hp, y=mpg)) +
 geom_smooth(method="lm", color="black", se=FALSE) +
 geom_point(aes(fill=cyl, 
                shape=cyl,
                size=wt)) +
 facet_grid(origin~transmission) +
 scale_shape_manual(name="Engine\ncylinders", 
                    values=c(21,22,24)) + 
 scale_fill_viridis_d(name="Engine\ncylinders") + 
 scale_size_continuous(guide=FALSE) +
 labs(x="Engine power (horsepower)",
      y="Fuel economy (miles/gallon)", 
      title="Fuel efficiency declines as engine power increases",
      caption="Point sizes scaled to car weight") +
 theme_bw(16) 

We won’t cover custom modification of themes here (other blogs do so), but we can check out a few that have been pre-packaged to mimic classic styles…

pacman::p_load(ggthemes)
  ggp + theme_wsj()   # Print like a Wall Street Journal graph
  ggp + theme_economist()  # Or one from the Economist
  ggp + theme_tufte() 

… and those of other stats packages:

ggp + theme_stata() 

ggp + theme_excel() 

ggp + theme_base()