Chapter 8 Graphics and Their Grammar

Example: Plotting Princes.

Figure 8.1 depicts the number of babies given the name “Prince” over the years.

Figure 8.1: The popularity of the name Prince over the years

BabyNames, where the Prince data resides, is not in the form needed for this plot. That is, BabyNames is not glyph-ready for this particular graphic; there are entries in BabyNames for 92599 other names than Prince. Wrangling BabyNames to remove those other names is a straightforward but necessary step to draw the graphic. In this situation, wrangling involves only one step: [Click to see note.]As you’ll see in Chapter 10, filter() is a data verb that passes through only those cases meeting the specified criterion.

Princes <-
  BabyNames %>%
  filter(name == "Prince") 

The graphic’s frame is count versus year. Frames are created with ggplot(). The mapping of the variables year and count to the x- and y-axes is specified by the aes() function.

the_frame <- ggplot(data = Princes, 
                    aes(x = year, y = count))

There are two layers, one showing the number of Princes each year and one with a vertical line at the year Prince’s first album was released. In the number-of-princes layer, shape was used to distinguish between the sexes.

layer1 <- geom_point(data = Princes, aes(shape = sex))

In the second layer, the glyph is a vertical line. This glyph is provided by the geom_vline() function:

layer2 <- geom_vline(xintercept = 1978)

Finally, put them all together with +

the_frame + layer1 + layer2

Figure 8.2: Putting together the frame, layer 1, and layer 2 into a complete graphic.

Chaining is built in to ggplot() and the other graphics functions. This provides a way to write your graphics commands without having to specify, over and over again, the source of data and the mappings of aesthetics. To illustrate, here is another way of describing the Prince graphic: [Click to see note.]Note that the aesthetics for geom_point() are inherited from a previous layer of the graphic.

Princes %>%
  ggplot(aes(x = year, y = count, shape = sex)) +
  geom_point() +
  geom_vline(xintercept = 1978)

Example: Growing children

People grow in height through childhood but then stay more or less the same height as adults. Question: Is there systematic change in height during adulthood? Does it depend on smoking?

First, wrangle NCHS into a glyph-ready form to give the mean height at each age, distinguishing between the sexes.

Heights <- 
  NCHS %>%
  filter( age > 20, ! is.na(smoker)) %>%
  group_by(sex, smoker, age) %>%
  summarise(height = mean(height, na.rm=TRUE))

Table 8.1: Glyph-ready data for displaying height versus age.

Figure 8.3 has a frame displaying height versus age. Either a point or a line could be used as the glyph. A separate line is made for each group: smokers versus non-smokers. Sex facets the graphic. Line type (dotted or not) distinguishes smokers from non-smokers smokers.

Heights %>%
  ggplot(aes(x = age, y = height)) +
  geom_line(aes(linetype = smoker)) +
  facet_wrap( ~ sex) 

Figure 8.3: Height versus age for smokers and non-smokers.

It looks like both female and male height decline with age. For women, smokers tend to be taller than non-smokers, but not for men.

The statistically savvy reader may point out that any causal relationship might go from height to smoking. Perhaps smoking is more common among taller women, rather than non-smoking causing women to be shorter. As well, NCHS does not track individuals over time.11 NCHS is from a “cross-sectional” study rather than a “longitudinal study.” It might be that an individual’s height doesn’t change with age, but the older people in NCHS grew up in a time with poorer nutrition and so are shorter for that reason.

The sharp ups and downs in Figure 8.3 for different ages suggest that there is variation in height that isn’t accounted for by age, sex, and smoking status. Such random variation can be quantified by a statistical technique called a confidence band. These will be introduced in Chapter 15.