3.2 ggplot2: Geoms

1. Smoothing: geom_smooth()

A smooth line is a way to see a trend. The default is a loess curve with a gray standard-error band.

ggplot(data = mpg, mapping = aes(x = displ, y = hwy)) +
  geom_smooth()

ggplot(data = mpg, mapping = aes(x = displ, y = hwy)) +
  geom_smooth(se = FALSE)

se = FALSE drops the band. A categorical mapping in aes() gives one line per level:

ggplot(data = mpg, mapping = aes(x = displ, y = hwy, color = drv)) +
  geom_smooth(se = FALSE)

ggplot(data = mpg, mapping = aes(x = displ, y = hwy, linetype = drv)) +
  geom_smooth(se = FALSE)

color is usually easier to read than linetype.

method = lm is ordinary least squares. For a binary 0/1 response, method = glm with a binomial family is a logistic curve:

ggplot(data = mpg, mapping = aes(x = displ, y = hwy)) +
  geom_point() +
  geom_smooth(se = FALSE, method = lm)

ggplot(data = mtcars, mapping = aes(x = hp, y = vs)) +
  geom_point() +
  geom_smooth(method = glm, method.args = list(family = "binomial"), se = FALSE)

2. Where the mapping lives

ggplot(aes(...)) is the default mapping. Later geoms inherit it unless they set their own.

ggplot(data = mpg, mapping = aes(x = displ, y = hwy)) +
  geom_point() +
  geom_smooth(se = FALSE)

A mapping written only inside one geom stays there. Later geoms do not see it. That is how you color the points by drv and still get one smooth line:

ggplot(data = mpg, mapping = aes(x = displ, y = hwy)) +
  geom_point(mapping = aes(color = drv)) +
  geom_smooth(se = FALSE)

If aes(x, y) is only on geom_point(), geom_smooth() has nothing to smooth and errors.

3. Overplotting

When many rows share the same rounded values, points sit on top of each other. cty and hwy in mpg do this.

Three common fixes:

1. A small data set: geom_jitter() nudges each point a little
2. A large data set: geom_point(alpha = 0.1) (or smaller)
3. A very large data set: take a random sample first

ggplot(data = mpg, mapping = aes(x = cty, y = hwy)) +
  geom_point()

ggplot(data = mpg, mapping = aes(x = cty, y = hwy)) +
  geom_jitter()

ggplot(data = mpg, mapping = aes(x = cty, y = hwy)) +
  geom_point(alpha = 0.1)

geom_hex() bins the plane into hexagons and colors by the count. Useful when jitter and alpha are not enough:

data("diamonds")

ggplot(data = diamonds, mapping = aes(x = carat, y = price)) +
  geom_hex(bins = 50) +
  viridis::scale_fill_viridis()

diamonds is in ggplot2. data("diamonds") loads it.

4. One quantitative variable: histograms and densities

A histogram bins a numeric variable and counts how many values fall in each bin. The default is 30 bins; change bins or binwidth until the shape is readable.

ggplot(data = mpg, mapping = aes(x = hwy)) +
  geom_histogram()

ggplot(data = mpg, mapping = aes(x = hwy)) +
  geom_histogram(bins = 20)

ggplot(data = mpg, mapping = aes(x = hwy)) +
  geom_histogram(binwidth = 5)

fill colors the interior of the bars. color colors the outline. Mapping a categorical variable to fill stacks the groups inside each bin:

ggplot(data = mpg, mapping = aes(x = hwy, fill = drv)) +
  geom_histogram(bins = 20)

Stacked bars are hard to compare. position changes the layout:

• "identity": all groups start at zero and can hide each other; add alpha
• "dodge": bars side by side
• "fill": each bin is a proportion (the bar goes to 1)

ggplot(data = mpg, mapping = aes(x = hwy, fill = drv)) +
  geom_histogram(bins = 20, position = "identity", alpha = 0.3)

ggplot(data = mpg, mapping = aes(x = hwy, fill = drv)) +
  geom_histogram(bins = 20, position = "dodge")

ggplot(data = mpg, mapping = aes(x = hwy, fill = drv)) +
  geom_histogram(bins = 20, position = "fill")

A density is a smoothed histogram:

ggplot(data = mpg, mapping = aes(x = hwy)) +
  geom_density()

ggplot(data = mpg, mapping = aes(x = hwy, fill = drv)) +
  geom_density(alpha = 0.3)

5. One categorical and one quantitative: boxplots

A boxplot shows the median, the 25th and 75th percentiles (the box, whose height is the IQR), whiskers out to 1.5 IQR from the hinges, and points beyond that.

ggplot(data = mpg, mapping = aes(x = drv, y = hwy)) +
  geom_boxplot()

geom_violin() shows the density of the points inside the box. ggbeeswarm::geom_beeswarm() shows every point without stacking them on a line.

Salaries, prices, and other right-skewed variables often need a log scale. Two different jobs:

• y = log(volume) changes the data. The axis ticks are logs.
• scale_y_log10() changes the axis. The ticks are still in the original units, spaced on a log scale.

txhousing |>
  ggplot(aes(x = as.factor(year), y = volume)) +
  geom_boxplot()

txhousing |>
  ggplot(aes(x = as.factor(year), y = volume)) +
  geom_boxplot() +
  scale_y_log10()

txhousing is in ggplot2.

6. Practice

1. Scatterplot petal length vs petal width. Color the points by species. Add one linear smooth (method = lm) that ignores species.

2. Plot price vs carat in diamonds. Fix the overplotting.

3. Histogram of diamond price, filled by cut.

4. Boxplot of diamond price by cut on a log-y scale. Does Ideal look cheaper? Check carat the same way.