In astamm/teachr: Tutorials for Teaching Tidyverse-Based R Basics

library(learnr)
library(testwhat)
library(magrittr)
library(ggplot2)
starwars <- dplyr::starwars

options(repos = "https://cloud.r-project.org")
tutorial_options(
  exercise.timelimit = 60,
  exercise.checker = testwhat::testwhat_learnr
)
knitr::opts_chunk$set(comment = NA)

Purpose

Disclaimer

This tutorial is in parts built from RStudio 2-day internal R bootcamp.

Visualisation

The simple graph has brought more information to the data analyst’s mind than any other device. — John Tukey

• Data visualization is the creation and study of the visual representation of data.
• There are many tools for visualizing data (R is one of them), and many approaches/systems within R for making data visualizations (ggplot2 is one of them, and that is what we are going to use).

The ggplot2 package

Introduction

Powerful and complex data visualisations can be achieved in a systematic fashion using the ggplot2 package, which implements a grammar of graphics that enables us to concisely describe the components of a graphic.

Notes:

• Contrary to what we prescribed, we will deliberately import all functions from ggplot2 into our R session so that we do not have to repeatedly write ggplot2:: in front of each function from ggplot2 that we call. This is because, when it comes to plotting, an enormous quantity of ggplot2 functions are used so the code would become unnecessarily complex if we decided not to import ggplot2;
• Similarly, we will illsutrate the features of ggplot2 using the dplyr::starwars, so we will extract it from dplyr to be able to access it by simply calling starwars.

Both of these tasks are performed by the following lines of code:

library(ggplot2)
starwars <- dplyr::starwars

Let us start by looking at this grammar syntax from an example. We first plot the mass of each character against their height:

ggplot(data = starwars, mapping = aes(x = height, y = mass)) +
  geom_point()

What does the following modification achieve?

ggplot(data = starwars, mapping = aes(x = height, y = mass)) +
  geom_point() +
  geom_smooth()

Note: When constructing a plot, you do not apply consectutive processes to an original data set. Rather, you want to add layers for plotting different data representations. The operator to add layers is therefore the natural + addition operator and not the %>% pipe operator.

Visual identification of outliers

Look at the list of available geom_*() functions here to find a function that allows you to add labels to each point in order to visually spot who is the fattest character who seems on their own on the top of the plot. Modify the following code accordingly:

ggplot(data = starwars, mapping = aes(x = height, y = mass)) +
  geom_point()

Are you satisfied with the way names are added to the plot? Investigate the package ggrepel to see if you can find functions that might help obtaining a better visual representation. Again, modify the following code accordingly:

ggplot(data = starwars, mapping = aes(x = height, y = mass)) +
  geom_point()

Is it enough? How about adding just the name of the fattest character then? Again, modify the following code accordingly:

ggplot(data = starwars, mapping = aes(x = height, y = mass)) +
  geom_point()

The ggplot2 structure

• ggplot() is the main function in ggplot2
• Plots are constructed in layers
• Structure of the code for plots can be summarized by the following scheme:

knitr::include_graphics("images/ggplot-structure.png")

Visualising Star Wars

Let us use what we learnt so far to take a look at the summary of the data:

skimr::skim(starwars)

Substantially, a number of features are reported for each character appearing in the Star Wars movies. The following image illustrates on Luke Skywalker the measured variables:

knitr::include_graphics("images/luke-skywalker.png")

Note: Data sets are also documented in R packages. This means that you can also invoke the help function on data sets and not only on functions. You can thus have information about a package data set using, e.g., ?dplyr::starwars.

Your turn: The following code plots the mass against the height. Change the variables assigned to the x and y axes to see the distributions of other variables and see what happens:

ggplot(data = starwars, mapping = aes(x = height, y = mass)) +
  geom_point()

Specifying labels

You can give your plot a proper title and subtitle, specify axis labels and rename the variables mapped to aesthetics with more human-readable names. This is the job of the labs() function:

ggplot(data = starwars, mapping = aes(x = height, y = mass)) +
  geom_point() +
  labs(
    title = "Mass vs. height of Starwars characters", 
    x = "Height (cm)", 
    y = "Weight (kg)"
  )

Aesthetics

Visual characteristics of plotting characters that can be mapped to a specific variable in the data are:

• linetype
• label
• shape
• colour
• fill
• size
• alpha (transparency)

There are two types of variables:

• Continuous variable are measured on a continuous scale
• Discrete variables are measured (or often counted) on a discrete scale

Mapping continuous or discrete variables to the previous aesthetics will behave differently:

aesthetics | discrete | continuous ------------- | ------------------------------------ | ------------ linetype | type of line for each (small number) | shouldn't be used label | value for each (converted to text) | shouldn't be used shape | different shape for each | shouldn't (and doesn't) work color | rainbow of colors | gradient fill | rainbow of colors | gradient size | discrete steps | linear mapping between radius and value alpha | shoudn't be used | linear mapping between radius and value

You might also want to set these features to fixed values, not depending on a variable:

• Use aesthetics for mapping features of a plot to a variable;
• Define the features directly in the geom_*() function for customization not mapped to a variable.

Color

ggplot(
  data = starwars, 
  mapping = aes(
    x = height, 
    y = mass, 
    color = gender
  )
) +
  geom_point()

Color + Size

• Using a fixed point size:

ggplot(
  data = starwars, 
  mapping = aes(
    x = height, 
    y = mass, 
    color = gender
  )
) +
  geom_point(size = 2)

• Using the variable birth_year mapped to thesize aesthetic:

ggplot(
  data = starwars, 
  mapping = aes(
    x = height, 
    y = mass, 
    color = gender, 
    size = birth_year
  )
) +
  geom_point()

Let us prep this last plot for reporting using a more complex piece of code:

ggplot(
  data = starwars, 
  mapping = aes(
    x = height, 
    y = mass, 
    color = gender, 
    size = birth_year
  )
) +
  geom_point(alpha = 0.7) +
  labs(
    title = "Mass vs. height of Starwars characters",
    subtitle = "by gender and birth year",
    x = "Height (cm)", 
    y = "Weight (kg)",
    color = "Gender",
    size = "Birth year"
  ) +
  theme_minimal() +
  theme(
    legend.direction = "horizontal", 
    legend.position = "bottom", 
    legend.box = "vertical"
  )

Faceting

• Smaller plots that display different subsets of the data
• Useful for exploring conditional relationships and large data

You have two functions in ggplot2 to do that:

• facet_grid(): Lay out panels in a grid;
• facet_wrap(): Wrap a 1d ribbon of panels into 2d.

ggplot(data = starwars, mapping = aes(x = height, y = mass)) +
  facet_grid(cols = vars(gender)) +
  geom_point() +
  labs(
    title = "Mass vs. height of Starwars characters", 
    subtitle = "Faceted by gender"
  ) + 
  theme(aspect.ratio = 1)

Important: Take a close look at the syntax to specify the facetting variables. They need to be specified within a call to the function vars(). This is how ggplot2 knows that it should find the variables in the data set associated with the current plot.

Your turn. Now check out the online help of the facet_grid() and facet_wrap() functions to solve the following exercises.

Exercise 1. Modify the code to have each subplot with its own x-axis scale:

ggplot(data = starwars, mapping = aes(x = height, y = mass)) +
  facet_grid(cols = dplyr::vars(gender)) +
  geom_point() +
  labs(
    title = "Mass vs. height of Starwars characters", 
    subtitle = "Faceted by gender"
  ) + 
  theme(aspect.ratio = 1)

Exercise 2. Modify the code to have each subplot displayed in a single column rather than in a single row:

ggplot(data = starwars, mapping = aes(x = height, y = mass)) +
  facet_grid(cols = dplyr::vars(gender)) +
  geom_point() +
  labs(
    title = "Mass vs. height of Starwars characters", 
    subtitle = "Faceted by gender"
  ) + 
  theme(aspect.ratio = 1)

Exercise 3. Modify the code to use eye_color instead of gender for facetting. What is the difference between facet_grid and facet_wrap? Which one is it better to use in this case?

ggplot(data = starwars, mapping = aes(x = height, y = mass)) +
  facet_grid(cols = dplyr::vars(gender)) +
  geom_point() +
  labs(
    title = "Mass vs. height of Starwars characters", 
    subtitle = "Faceted by gender"
  ) + 
  theme(aspect.ratio = 1)

Exercise 4. Modify the code to use hair_color and gender for facetting.

ggplot(data = starwars, mapping = aes(x = height, y = mass)) +
  facet_grid(cols = dplyr::vars(gender)) +
  geom_point() +
  labs(
    title = "Mass vs. height of Starwars characters", 
    subtitle = "Faceted by gender"
  ) + 
  theme(aspect.ratio = 1)

Visualizing variation

Numerical data

Histogram

ggplot(data = starwars, mapping = aes(x = height)) + 
  geom_histogram(binwidth = 10)

Density plot

ggplot(data = starwars, mapping = aes(x = height)) +
  geom_density()

Box plot

ggplot(data = starwars, mapping = aes(y = height)) +
  geom_boxplot()

Categorical data: the bar plot

ggplot(data = starwars, mapping = aes(x = gender)) +
  geom_bar()

Visualizing co-variation

Numerical and categorical data

Color-filled histograms?

ggplot(data = starwars, mapping = aes(x = height, fill = gender)) +
  geom_histogram(
    binwidth = 10, 
    position = position_dodge2(), 
    color = "black"
  )

This representation is not really helpful because adjusting a unique binning parameter appropriate for all histograms is rarely possible.

Color-filled densities

ggplot(data = starwars, mapping = aes(x = height, fill = gender)) +
  geom_density(color = "black", alpha = 0.3)

Side-by-side box plots

ggplot(data = starwars, mapping = aes(y = height, x = gender)) +
  geom_boxplot()

Scatter plot?

ggplot(data = starwars, mapping = aes(y = height, x = gender)) +
  geom_point()

This is not a great representation of these data. Why?

Violin plots

ggplot(data = starwars, mapping = aes(y = height, x = gender)) +
  geom_violin()

Both categorical data

Segmented bar plots, counts

ggplot(
  data = starwars, 
  mapping = aes(x = gender, fill = hair_color)
) +
  geom_bar(position = position_stack())

Recode hair color

Using the fct_other() function from the forcats package, which is also part of the tidyverse.

starwars <- starwars %>%
  dplyr::mutate(
    hair_color2 = forcats::fct_other(
      f = hair_color, 
      keep = c("black", "blond", "brown", "white")
    )
  )

Segmented bar plots, counts

ggplot(
  data = starwars, 
  mapping = aes(x = gender, fill = hair_color2)
) +
  geom_bar(position = position_stack()) +
  coord_flip()

Segmented bar plots, proportions

ggplot(
  data = starwars, 
  mapping = aes(x = gender, fill = hair_color2)
) +
  geom_bar(position = position_fill()) +
  coord_flip() + 
  labs(y = "proportion")

Nested bar plots

Version 1: Explore how the hair color is distributed within each gender category.

ggplot(
  data = starwars, 
  mapping = aes(x = gender, fill = hair_color2)
) +
  geom_bar(position = position_dodge2(preserve = "single"))

Version 2: Explore how gender is distributed within each hair color category.

ggplot(
  data = starwars, 
  mapping = aes(x = hair_color2, fill = gender)
) +
  geom_bar(position = position_dodge2(preserve = "single"))

Your turn: Why all bar plots do not have the same number of bars? How could you remedy?

**Hint:** 1. Combine [dplyr::count()](https://dplyr.tidyverse.org/reference/tally.html) and [tidyr::complete()](https://tidyr.tidyverse.org/reference/complete.html) to manually perform the counting and avoid dropping combinations with zero count. 2. Use `stat = "identity"` as an argument to the `geom_bar()` since you do not need [**ggplot2**](https://ggplot2.tidyverse.org) to perform the counting behind the scene as you already did it.

Visualization in higher dimensions

ggplot2 extensions

The ggplot2 package is now widely used worldwide. As a result, a number of people have been developing extensions to perform all kinds of plots. It is highly probable that, if you do not manage to draw a plot you have on your mind easily with your knowledge of ggplot2, then someone probably has published an extension already that perfectly fits your needs. You can consult at any time the up-to-date list of ggplot2 extensions.

Heatmaps

You can visualize a third dimension by color-coding a 2D plane. This is what does a heatmap. There are several ways to draw heatmaps in R. We recommend the heatmaply package which provides interactive heatmaps.

Data visualization of 3D data in 3D

Lots of packages offer 3D visualization tools. We recommend the plot3D package, which is particularly well implemented and easy to use. A nice introductory tutorial can be found here.

Visualizing nD data

Visualization of nD data seems impossible on paper. One way to actually do this is to visualize projections of the nD data into $\mathbb{R}$, $\mathbb{R}^2$ or $\mathbb{R}^3$. You then create a tour of your data by generating a smooth sequence of projections and display your data as a movie. In R, this is easily achieved by the tourr package.

To demonstrate this technique we will use a data set in the tourr package called flea, which contain measurements of flea beetle characteristics for 3 different species:

skimr::skim(tourr::flea)

Displaying data as a movie cannot be performed within a tutorial, so let's open another RStudio application to run the following code:

# macOS and Linux
quartz()
# Windows
# X11()
tourr::animate_xy(
  data = tourr::flea[, -7],
  col = tourr::flea[, 7]
)

Exercise

The values of two numerical variables x and y has been collected in 13 different situations. The data is available for you under the name experiment as a tibble.

experiment <- datasauRus::datasaurus_dozen_wide

1. Transform the data so as to make it tidy.

1. Calculate the mean and standard deviation of both x and y as well as their correlation, for each of the 13 situations.

**Hint:** Use `mean()` to compute the mean, `sd()` to compute the standard deviation and `cor()` to compute the correlation.

1. Create a visualization of y vs. x, for each of the 13 situations. How similar or different are the data collected in the different situations?

astamm/teachr documentation built on Jan. 12, 2023, 7:21 a.m.