Exercises

The following packages are required for this practical:

library(dplyr)
library(magrittr)
library(mice)
## Warning: package 'mice' was built under R version 3.5.1
library(ggplot2)

and if you’d like the same results as I have obtained, you can fix the random seed

set.seed(123)
  1. Function plot() is the core plotting function in R. Find out more about plot(): Try both the help in the help-pane and ?plot in the console. Look at the examples by running example(plot).

The help tells you all about a functions arguments (the input you can specify), as well as the element the function returns to the Global Environment. There are strict rules for publishing packages in R. For your packages to appear on the Comprehensive R Archive Network (CRAN), a rigorous series of checks have to be passed. As a result, all user-level components (functions, datasets, elements) that are published, have an acompanying documentation that elaborates how the function should be used, what can be expected, or what type of information a data set contains. Help files often contain example code that can be run to demonstrate the workings.

?plot
## starting httpd help server ... done
example(plot)
## 
## plot> require(stats) # for lowess, rpois, rnorm
## 
## plot> plot(cars)

## 
## plot> lines(lowess(cars))
## 
## plot> plot(sin, -pi, 2*pi) # see ?plot.function

## 
## plot> ## Discrete Distribution Plot:
## plot> plot(table(rpois(100, 5)), type = "h", col = "red", lwd = 10,
## plot+      main = "rpois(100, lambda = 5)")

## 
## plot> ## Simple quantiles/ECDF, see ecdf() {library(stats)} for a better one:
## plot> plot(x <- sort(rnorm(47)), type = "s", main = "plot(x, type = \"s\")")

## 
## plot> points(x, cex = .5, col = "dark red")

There are many more functions that can plot specific types of plots. For example, function hist() plots histograms, but falls back on the basic plot() function. Packages lattice and ggplot2 are excellent packages to use for complex plots. Pretty much any type of plot can be made in R. A good reference for packages lattice that provides all R-code can be found at http://lmdvr.r-forge.r-project.org/figures/figures.html. Alternatively, all ggplot2 documentation can be found at http://docs.ggplot2.org/current/

  1. Create a scatterplot between age and bmi in the mice::boys data set

With the standard plotting device in R:

mice::boys %$% plot(bmi ~ age)

or, with ggplot2:

p <- ggplot(mice::boys, aes(age, bmi))
p + geom_point()
## Warning: Removed 21 rows containing missing values (geom_point).

Package ggplot2 offers far greater flexibility in data visualization than the standard plotting devices in R. However, it has its own language, which allows you to easily expand graphs with additional commands. To make these expansions or layers clearly visible, it is advisable to use the plotting language conventions. For example,

mice::boys %>% 
  ggplot(aes(age, bmi)) +
  geom_point()

would yield the same plot as

ggplot(mice::boys, aes(age, bmi)) + geom_point()

but the latter style may be less informative, especially if more customization takes place and if you share your code with others.

  1. Now recreate the plot with the following specifications:

Hint: it may help to expand the data set with a new variable.

It may be easier to create a new variable that creates the specified categories. We can use the cut() function to do this quickly

boys2 <- 
  boys %>%
  mutate(class = cut(bmi, c(0, 18.5, 25, 30, Inf),
                    labels = c("underweight",
                               "healthy",
                               "overweight",
                               "obese")))

by specifying the boundaries of the intervals. In this case we obtain 4 intervals: 0-18.5, 18.5-25, 25-30 and 30-Inf. We used the %>% pipe to work with bmi directly. Alternatively, we could have done this without a pipe:

boys3 <- boys
boys3$class <- cut(boys$bmi, c(0, 18.5, 25, 30, Inf), 
                   labels = c("underweight",
                              "healthy",
                              "overweight",
                              "obese"))

to obtain the same result.

With the standard plotting device in R we can now specify:

plot(bmi ~ age, subset = class == "underweight", col = "light blue", data = boys2, 
     ylim = c(10, 35), xlim = c(0, 25))
points(bmi ~ age, subset = class == "healthy", col = "light green", data = boys2)
points(bmi ~ age, subset = class == "overweight", col = "orange", data = boys2)
points(bmi ~ age, subset = class == "obese", col = "red", data = boys2)

and with ggplot2 we can call

boys2 %>%
  ggplot() +
  geom_point(aes(age, bmi, col = class))
## Warning: Removed 21 rows containing missing values (geom_point).

Although the different classifications have different colours, the colours are not conform the specifications of this exercise. We can manually override this:

boys2 %>%
  ggplot() +
  geom_point(aes(age, bmi, col = class)) +
  scale_color_manual(values = c("light blue", "light green", "orange", "red"))
## Warning: Removed 21 rows containing missing values (geom_point).

Because there are missing values, ggplot2 displays a warning message. If we would like to not consider the missing values when plotting, we can simply exclude the NAs by using a filter():

filter(boys2, !is.na(class)) %>%
  ggplot() +
  geom_point(aes(age, bmi, col = class)) +
  scale_color_manual(values = c("light blue", "light green", "orange", "red"))

Specifying a filter on the feature class is sufficient: age has no missings and the missings in class directly correspond to missing values on bmi. Filtering on bmi would therefore yield an identical plot.

  1. Create a histogram for age in the boys data set

With the standard plotting device in R:

boys %$%
  hist(age, breaks = 50)

The breaks = 50 overrides the default breaks between the bars. By default the plot would be

boys %$%
  hist(age)

Using a pipe is a nice approach for this plot because it inherits the names of the objects we aim to plot. Without the pipe we might need to adjust the main title for the histogram:

hist(boys$age, breaks = 50)

With ggplot2:

boys %>%
  ggplot() + 
  geom_histogram(aes(age), binwidth = .4)

Please note that the plots from geom_histogram() and hist use different calculations for the bars (bins) and hence may look slightly different.

  1. Create a bar chart for reg in the boys data set With a standard plotting device in R:
boys %$%
  table(reg) %>%
  barplot()

With ggplot2:

boys %>%
  ggplot() + 
  geom_bar(aes(reg))

Note that geom_bar by default plots the NA’s, while barplot() omits the NA’s without warning. If we would not like to plot the NAs, then a simple filter() (see exercise 2) on the boys data is efficient.

  1. Create a box plot for hgt with different boxes for reg in the boys data set With a standard plotting device in R:
boys %$%
  boxplot(hgt ~ reg)

With ggplot2:

boys %>%
  ggplot(aes(reg, hgt)) +
  geom_boxplot()
## Warning: Removed 20 rows containing non-finite values (stat_boxplot).

  1. Create a density plot for age with different curves for boys from the city and boys from rural areas (!city). With a standard plotting device in R:
d1 <- boys %>%
  subset(reg == "city") %$%
  density(age)
d2 <- boys %>%
  subset(reg != "city") %$% 
  density(age)

plot(d1, col = "red", ylim = c(0, .08)) 
lines(d2, col = "blue")

The above plot can also be generated without pipes, but results in an ugly main title

plot(density(boys$age[!is.na(boys$reg) & boys$reg == "city"]), 
     col = "red", 
     ylim = c(0, .08))
lines(density(boys$age[!is.na(boys$reg) & boys$reg != "city"]), 
      col = "blue")

With ggplot2 everything looks much nicer:

boys %>%
  mutate(area = ifelse(reg == "city", "city", "rural")) %>%
  filter(!is.na(area)) %>%
  ggplot(aes(age, fill = area)) +
  geom_density(alpha = .3) #some opacity

  1. Create a diverging bar chart for hgt in the boys data set, that displays for every age year that year’s mean height in deviations from the overall average hgt

Let’s not make things too complicated and just focus on ggplot2:

boys %>%
  mutate(Hgt = hgt - mean(hgt, na.rm = TRUE),
         Age = cut(age, 0:22, labels = 0:21)) %>%
  aggregate(Hgt ~ Age, data = ., mean) %>% #specify data = . to allow formula
  mutate(Diff = cut(Hgt, c(-Inf, 0, Inf),
                    labels = c("Below Average", "Above Average"))) %>%
  ggplot(aes(x = Age, y = Hgt, fill = Diff)) + 
  geom_bar(stat = "identity") +
  coord_flip()

We can clearly see that the average height in the group is reached just before age 7. The aggregate() function is used to return the mean() of deviation Hgt for every group in Age.

For example, if we would like the mean height hgt for every region reg in the boys data, we could call:

boys %>%
  aggregate(hgt ~ reg, data = ., FUN = mean)
##     reg      hgt
## 1 north 151.6316
## 2  east 133.9648
## 3  west 130.2783
## 4 south 128.0022
## 5  city 125.8577

We have to specify data = . in order to allow for the formula-style call to aggregate() - where the method is of class formula. However, the data set boys is parsed down the pipe as an object of class data.frame. The default evaluation of argument would therefore be

aggregate(x, by, FUN)

and in the pipe this is by default evaluated as

aggregate(., hgt ~ reg, mean)

where . is the object parsed down the pipe. This . is automatically evaluated as the first argument, unless otherwise specified by the user. In most cases this works because the data is usually the first argument that is evaluated in a function. However, the result is not a valid call to aggregate() because the object we’re parsing down the pipe has class data.frame. aggregate() would therefor try to run aggregate.data.frame(), but our code dictates an evaluation of the formula call. By assigning the . to data = ., we specifically call for a formula evaluation of aggregate(). This solves the mismatch and forces aggregate() to conform to class formula:

aggregate(formula, data, FUN)

which is evaluated as

aggregate(hgt ~ reg, data = ., mean)

in the pipe. Problem solved!

The specifics about calling functions and evaluating their arguments can always be found in the help. Try ?aggregate to see all forms this function’s call may take.

End of Practical

Useful References

The ggplot2 reference page