Introduction

In this practical you will learn about exploratory data analysis (EDA). EDA involves inspecting, transforming and visualising data with the goal of finding interesting features to substantiate one or more research questions. This practical will draw on skills you learned in week 1 and week 2, while introducing you to some new material like visualisation with ggplot().

You will need the following packages for this tutorial.

library(tidyverse)
library(ggplot2)
library(ggpubr)
library(kableExtra)
library(weathermetrics)

You will work with the mpg dataset from the ggplot2 package. mpg contains a subset of the fuel economy data that EPA makes available here. The data frame has 234 rows and 11 variables:

  • manufacturer: manufacturer name
  • model: model name
  • displ: engine displacement in litres
  • year: year of manufacture
  • cyl: number of cyclinders
  • trans: type of transmission
  • drv: type of drive tran (f = front-wheel drive; r = rear-wheel drive; 4 = four-wheel drive)
  • cty: city miles per gallon
  • hw: highway miles per gallon
  • fl: fuel type
  • class” type of car

You can also type ?mpg into the console to learn more.

Let’s get an overview of the data.

head(mpg)

In previous weeks we learned that str() can be used to get an overview of the data and the structure of a dataframe. You can also use glimpse() from the tidyverse which returns similar results in a more readable format.

glimpse(mpg)
## Rows: 234
## Columns: 11
## $ manufacturer <chr> "audi", "audi", "audi", "audi", "audi", "audi", "audi", "…
## $ model        <chr> "a4", "a4", "a4", "a4", "a4", "a4", "a4", "a4 quattro", "…
## $ displ        <dbl> 1.8, 1.8, 2.0, 2.0, 2.8, 2.8, 3.1, 1.8, 1.8, 2.0, 2.0, 2.…
## $ year         <int> 1999, 1999, 2008, 2008, 1999, 1999, 2008, 1999, 1999, 200…
## $ cyl          <int> 4, 4, 4, 4, 6, 6, 6, 4, 4, 4, 4, 6, 6, 6, 6, 6, 6, 8, 8, …
## $ trans        <chr> "auto(l5)", "manual(m5)", "manual(m6)", "auto(av)", "auto…
## $ drv          <chr> "f", "f", "f", "f", "f", "f", "f", "4", "4", "4", "4", "4…
## $ cty          <int> 18, 21, 20, 21, 16, 18, 18, 18, 16, 20, 19, 15, 17, 17, 1…
## $ hwy          <int> 29, 29, 31, 30, 26, 26, 27, 26, 25, 28, 27, 25, 25, 25, 2…
## $ fl           <chr> "p", "p", "p", "p", "p", "p", "p", "p", "p", "p", "p", "p…
## $ class        <chr> "compact", "compact", "compact", "compact", "compact", "c…

Before you begin exploring your data, you need to check that the class or mode of your variables is as expected. Remember from previous practicals, you have character/factor and numerical/integer types.

  1. Are there any variables that do not have the class you expect it to?
# All variables that are listed as 'character' are categorical and should be listed as 'factor'.
  1. Make any changes you feel are necessary.
# Here we should change all of the character variables to factor variables with levels. For now, we leave the levels as they are.
# There are a couple of methods to do this, but the quickest way is using `mutate_if()` which performs a conditional operation. Base R is also fine.
  
mpg <- mpg %>% 
  mutate_if(is.character, as.factor)

# If we inspect the data gain using glimpse(), we now see that the character variables are converted into factors
glimpse(mpg)
## Rows: 234
## Columns: 11
## $ manufacturer <fct> audi, audi, audi, audi, audi, audi, audi, audi, audi, aud…
## $ model        <fct> a4, a4, a4, a4, a4, a4, a4, a4 quattro, a4 quattro, a4 qu…
## $ displ        <dbl> 1.8, 1.8, 2.0, 2.0, 2.8, 2.8, 3.1, 1.8, 1.8, 2.0, 2.0, 2.…
## $ year         <int> 1999, 1999, 2008, 2008, 1999, 1999, 2008, 1999, 1999, 200…
## $ cyl          <int> 4, 4, 4, 4, 6, 6, 6, 4, 4, 4, 4, 6, 6, 6, 6, 6, 6, 8, 8, …
## $ trans        <fct> auto(l5), manual(m5), manual(m6), auto(av), auto(l5), man…
## $ drv          <fct> f, f, f, f, f, f, f, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, r, …
## $ cty          <int> 18, 21, 20, 21, 16, 18, 18, 18, 16, 20, 19, 15, 17, 17, 1…
## $ hwy          <int> 29, 29, 31, 30, 26, 26, 27, 26, 25, 28, 27, 25, 25, 25, 2…
## $ fl           <fct> p, p, p, p, p, p, p, p, p, p, p, p, p, p, p, p, p, p, r, …
## $ class        <fct> compact, compact, compact, compact, compact, compact, com…

Exploratory data analysis

Exploratory Data Analysis is a process whereby you examine and transform your data to extract interesting insights and understand what relationships (if any!) exist among the variables. In this way, EDA is closely related to principles of data cleaning because tidy data is easier to explore. Initial steps of EDA involve examining distributions, handling missings and outliers, removing redundant data and coding categorical variables properly.

EDA is very flexible and iterative process and gives the researcher a lot of freedom, but it is typically split into graphical and non-graphical analysis. Further within this framework we can also start to think about what kind of variation occurs within variables and what kind of covariation occurs between my variables.

Variation

Variation describes how the values of a variable change each time you measure it. A continuous variable that is measured on one occassion will have different values to that same variable measured on another occasion. This variation between measurement occasions are what we call “error”. Categorical variables also vary across measurement occasions or across individuals. Visualizing variation is a very intuitive way to understand trends or patterns within a variable.

Covariation

Covariation describes how the values of at least two variables vary together or are related. Again, visualising covariation is also the most intuitive way to understand patterns between variables.

Non-Graphical EDA

If you are interested in a particular aspect of a variable you can summarise this information in a table. We have covered how to create nicely formatted tables in the previous practical, so the example below is just to refresh your memory.

Summary tables

Below we summarise hwy by vehicle class, by asking for commonly used summary statistics.

mpg %>% 
  group_by(class) %>%         # by class
  summarise(min = min(hwy),   # minimum of hwy
            mean = mean(hwy), # mean
            sd = sd(hwy),     # standard deviation
            max = max(hwy),   # maximum
            n = n())  %>%     # nr of observations
  kbl() %>%  
  kable_styling(latex_options = c("striped", "hover"), full_width = F)
class min mean sd max n
2seater 23 24.80000 1.303840 26 5
compact 23 28.29787 3.781620 44 47
midsize 23 27.29268 2.135930 32 41
minivan 17 22.36364 2.062655 24 11
pickup 12 16.87879 2.274280 22 33
subcompact 20 28.14286 5.375012 44 35
suv 12 18.12903 2.977973 27 62

3. Create a summary of engine displacement by year, including the minimum, maximum, median, and inter quantile range.

mpg %>% 
  group_by(year) %>%                    # by year
  summarise(min = min(displ),           # minimum of engine displacement
            Q1 = quantile(displ, 0.25), # first quantile
            median = median(displ),     # median
            Q3 = quantile(displ, 0.75), # third quantile
            max = max(displ)) %>%       # maximum
  kbl() %>%  
  kable_styling(latex_options = c("striped", "hover"), full_width = F)
year min Q1 median Q3 max
1999 1.6 2.2 3.0 4.0 6.5
2008 1.8 2.5 3.6 4.7 7.0

Graphical EDA: base R vs ggplot

Plots can be made using base R functions like plot(), hist(), or barplot(). Here are some examples of how this works on the mpg data.

hist(mpg$displ)

barplot(table(mpg$cyl))

plot(x = mpg$displ, y = mpg$hwy,
     xlab = "Highway mpg",
     ylab = "Engine displacement (L)")

These plots are quick to produce and are useful for an initial understanding of the data, but the syntax used to create them is specific to the type of plot. In contrast, ggplot has a streamlined approach to plotting involving largely the same steps regardless of plot type. Specifically, in ggplot we build up visualizations layer by layer using the + operator (which is similar to the %>% operator we are already familiar with).

The main, important, layers in ggplot are:

  • Pass the data to ggplot()
  • Choose aesthetic mappings with aes()
  • Choose geometric components with geom()
  • Choose additional labels, themes, and visuals
  • Choose faceting if it is applicable with facet()

What does this layering system look like? In Chapter 3: Data Visualisation of Hadley Wickham’s R4DS book, there is a general ggplot template to give you an idea of how plots are layered.

```{r}
ggplot(data = <DATA>) + 
  <GEOM_FUNCTION>(
     mapping = aes(<MAPPINGS>)) +
  <COORDINATE_FUNCTION> +
  <FACET_FUNCTION> +
  <THEME> 
```

Below is a real example of how this looks. Remember that you do not have to / need to use all of the ggplot options (and there are many many more than presented here). However, this gives you an idea of how the layers of these plots are built up.

ggplot(iris) + # Data
  geom_point(mapping = aes(x = Sepal.Length, # Variable on the x-axis
                           y = Sepal.Width, # Variable on the y-axis
                           colour = Species)) + # Legend 
  labs(x = "Sepal Length",
       y = "Sepal Width", 
       title = "Relationship between Sepal Length and Width by Species") +
  coord_cartesian() + # Default standard for mapping x and y
  facet_wrap(~Species) + # Splits plot by the species variable
  theme_bw() # Sets the background theme

You can read about different ggplot2 options here. The reference guides by RPubs, STHDA and Data Novia will be useful once you grasp the basics and want to make more complicated visualisations.

Visualising distributions of single variables

Categorical variables

Bar charts are often used to visualise categorical variables with a finite set of values. Below is an example how to visuailse the distribution of drv - the type of drive train in the mpg dataset.

geom_bar transforms each value of drv into a count to be plotted on the y-axis. The x-axis presents the category associated with each count.

ggplot(mpg) +
  geom_bar(aes(x = drv))

  1. Create a bar chart that shows a count for the different vehicle classes in mpg.
ggplot(mpg) +              # list the dataset
  geom_bar(aes(x = class)) # geom_bar for a barplot, and the 'class' variable on the x-axis. 

# geom_bar adds counts on the y-axis by default
  1. Look up different ggplot themes, and apply one to the bar chart you just created.
ggplot(mpg) +
  geom_bar(aes(x = class)) +
  theme_bw() # Here we add the theme

# There are different `ggplot` themes and the one you choose is largely down to personal preference, but you also want one that enhances your data visualization. Here I choose `+ theme_bw()` but other examples include `+ theme_minimal()` and `+ theme_light()`.

Continuous variables

Histograms are often used to visualise the distribution of a continuous variable. Below is an example of how to visualise the distribution of cty - the number of city miles per gallon in the mpg dataset.

geom_hist transforms the x-axis into equally spaced “bins” and the height of each bar on the y-axis is the number of observations falling in each bin.

ggplot(mpg) +
  geom_histogram(aes(x = cty), 
                 binwidth = 3)

You can change the value of binwidth to adjust the width of the intervals. Different binwidths can reveal different patterns in the data.

  1. What happens when you change the value of binwidth to 1? Does the distribution change?
ggplot(mpg) +
  geom_histogram(aes(x = cty), 
                 binwidth = 1)

# The general distribution does not drastically change, but it is now much easier to spot outliers. Specifically, there are two extreme values to the right of the distribution where the city miles per gallon exceed 30 in a small number of cases.

Density plots are an alternative to histograms for continuous data. Below is an example of the density of cty from the mpg data. The argument fill = "darkseagreen" just adds colour. You can also specify colour = "<COLOUR NAME>" to change the density line.

geom_density transforms the data into smoothed kernel density estimates (basically a smooth histogram). It is useful for continuous data that come from an underlying smooth distribution.

ggplot(mpg, aes(x = cty)) +
  geom_density(fill = "darkseagreen")

You can also include raw data in a histogram in the form of “rug” marks.

  1. Add rug marks to plot above by adding the argument + geom_rug(size = 1, colour = "darkorange").
ggplot(mpg, aes(x = cty)) +
  geom_density(fill = "darkseagreen") +
  geom_rug(size = 1, 
           colour = "darkorange")
## Warning: Using `size` aesthetic for lines was deprecated in ggplot2 3.4.0.
## ℹ Please use `linewidth` instead.
## This warning is displayed once every 8 hours.
## Call `lifecycle::last_lifecycle_warnings()` to see where this warning was
## generated.

Visualing the distributions of multiple variables

Continuous - Continuous

Scatterplots are often used to explore how two continuous variables covary together. Below is an example of how to visualise covariation between displ (engine displacement) and hwy (highway miles per gallon) in the mpg data. If there is a clear pattern in how the points fall on the x- and y-axes then we can say these variables covary.

geom_point plots values of displ on the y-axis and values of hwy on the x-axis.

ggplot(mpg) +
  geom_point(aes(x = displ, 
                 y = hwy))

Some of these values might be clustered together based on another characteristic. You can explore this by adding colour to the aesthetic mappings, for example.

ggplot(mpg) +
  geom_point(aes(x = displ, 
                 y = hwy, 
                 colour = class))

  1. Repeat the plot above, but mapping the type of transmission to the colour aesthetic. Add a theme and change the titles of the x- and y-axes.
ggplot(mpg) +
  geom_point(aes(x = displ, 
                 y = hwy, 
                 colour = trans)) +
  labs(x = "Engine Displacement (L)",
       y = "Highway miles per gallon",) +
  theme_bw()

Different aesthetic mappings can be combined in one plot. For example, you may want to add a line of best fit to a scatterplot plot, which you can do using geom_smooth().

  1. Create a scatter plot showing the relationship between engine displacement and city miles, including a line of best fit.
ggplot(mpg, aes(x = displ, 
                y = hwy)) +
  geom_point() +
  geom_smooth() +
  labs(x = "Engine Displacement (L)",
       y = "Highway miles per gallon",) +
  theme_bw()
## `geom_smooth()` using method = 'loess' and formula = 'y ~ x'

# `geom_smooth()` is used to add a line of best fit. The default method is "loess" with the formula x ~ y, but you can change this to any other method. For example, you can choose `method = "lm"` for a linear best fit line.

You can fit more than one line in the same plot. This is often useful in multilevel modelling, where different hierarchies (e.g., groups, locations, individuals) have different slopes.

10. Recreate the previous plot, but adding colour = class to the aesthetic mappings. Use method = "lm" and set se = FALSE in geom_smooth(). What difference does this make?

ggplot(mpg, aes(x = displ, 
                y = hwy, 
                colour = class)) +
  geom_point() +
  geom_smooth(method = lm, 
              se = FALSE) +
  labs(x = "Engine Displacement (L)",
       y = "Highway miles per gallon",) +
  theme_bw()
## `geom_smooth()` using formula = 'y ~ x'

# Adding `colour = class` to the aesthetic mappings causes `geom_smooth()` to fit a line for each class of vehicle. Using `method = "lm"` with `se = FALSE` fits a linear line without the standard error. Dropping the standard error makes the plot more readable, since they are otherwise overlapping.

Continuous - Categorical

Boxplots are often used to explore the distribution of a continuous variable broken down by a categorical variable. below is an example of hwy broken down by drv.

geom_boxplot transforms the data to show summary statistics that are represented by the different visual features of the boxplot.

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

These features of the boxplot are:

  • Lower horizontal line
  • Thicker horizontal middle line
  • Upper horizontal line
  • Vertical whiskers
  • Points beyond the whiskers
  1. What do each of these features tell us?
# Lower horizontal line -> 25th percentile
# Thicker horizontal middle line -> median
# Upper horizontal line -> 75th percentile
# Vertical whiskers -> IQR 
# 1.5 Points beyond the whiskers ->  OUTLIERS
  1. Add + coord_flip() to the plot above. What does this do?
ggplot(mpg) +
  geom_boxplot(aes(x = drv, y = hwy)) +
  coord_flip()

# coord_flip()` swaps the x- and y-axes.
  1. What can you conclude about the highway miles per gallon of each type of drive train from the boxplot above?
# There are many outliers (points beyond the whiskers) for drive traint type "f", which represents a front-wheel drive. There are no outliers beyond the IQR for the other types of drive train, 4-wheel drive and rear-wheel drive.

The variable trans contains 10 different transmissions types. These 10 categories can be assumed under 2 broader categories: manual and auto.

  1. Use mutate() and fct_collapse() to collapse the 10 categories of trans to just these 2 categories.
mpg <- mpg %>% 
  mutate(trans = fct_collapse(trans,
                             "auto" = c("auto(av)", 
                                        "auto(l3)", 
                                        "auto(l4)", 
                                        "auto(l5)", 
                                        "auto(l6)", 
                                        "auto(s4)", 
                                        "auto(s5)", 
                                        "auto(s6)"),
                             "manual" = c("manual(m5)", 
                                          "manual(m6)")))
  1. Use the previous example to create a boxplot mapping cty on the y-axis and drv on the x-axis, this time adding the argument colour = trans to aes(). What has changed?
  ggplot(mpg) +
  geom_boxplot(aes(x = drv, 
                   y = cty, 
                   colour = trans))

# Adding the argument `colour = trans` to the aesthetic mappings further splits the drive train into auto and manual.

# Note that you can also perform the above operations in one piped step.

mpg %>% 
    mutate(trans = fct_collapse(trans,
                             "auto" = c("auto(av)", 
                                        "auto(l3)", 
                                        "auto(l4)", 
                                        "auto(l5)", 
                                        "auto(l6)", 
                                        "auto(s4)", 
                                        "auto(s5)", 
                                        "auto(s6)"),
                             "manual" = c("manual(m5)", 
                                          "manual(m6)"))) %>% 
  ggplot() +
  geom_boxplot(aes(x = drv, 
                   y = cty, 
                   colour = trans))
## Warning: There was 1 warning in `mutate()`.
## ℹ In argument: `trans = fct_collapse(...)`.
## Caused by warning:
## ! Unknown levels in `f`: auto(av), auto(l3), auto(l4), auto(l5), auto(l6), auto(s4), auto(s5), auto(s6), manual(m5), manual(m6)

Facets

Facets are a way to split your plot into many subplots according to some categorical variable.

To do this, you use the command facet_wrap() to facet by a single variable. The first argument to facet_wrap() is a formula initiated by ~ and a variable name. Below is an example of how this can work on the mpg data.

ggplot(mpg) + 
  geom_point(aes(x = displ, 
                 y = hwy)) + 
  facet_wrap(~ class, nrow = 2)

It is possible to facet on a combination of two variables using facet_grid(). Like before, the first argument is a formula, but containing two variable names separated by ~. Below is an example of how this can work.

ggplot(mpg) + 
  geom_point(aes(x = displ, 
                 y = hwy)) + 
  facet_grid(drv ~ cyl)

  1. Notice that there are empty cells in the bottom left facets above. What do you think this means?
# There are no rear-wheel drives that have 4 cyclinders, so there is nothing to plot here.
  1. Create a scatter plot of displ and hwy, faceting by the of manufacturer name.
ggplot(mpg) + 
  geom_point(aes(x = displ, 
                 y = hwy)) + 
  facet_wrap(~manufacturer, nrow = 3)

  1. Change the names on the axes to be more informative.
ggplot(mpg) + 
  geom_point(aes(x = displ, 
                 y = hwy)) + 
  labs(x = "Engine displacement (litres)", 
       y = "Highway miles per gallon") +
  facet_wrap(~manufacturer, nrow = 3)

Arranging multiple plots

If you have multiple plots that you want to arrange on the same page, you can use ggarrange() from the ggpubr package.

The first step is creating some plots. For this exercise we will only use the variables hwy, cty, and class.

  1. Create the following three plots:

A. A bar plot showing counts of class, assigning class to the fill aesthetic.

B. A box plot of class by hwy per gallon, assigning class to the colour aesthetic.

C. A jittered scatterplot of hwy and cty miles per gallon, assigning class to the colour aesthetic.

Remember to save each plot as an object to be called upon later.

# plot A
p1 <- ggplot(data = mpg, aes(class, fill = class)) +
        geom_bar() +
        labs(x = "Vehicle class",
              y = "Count") +
        coord_flip() +  
        theme_bw()

# plot B
p2 <- ggplot(data = mpg) + 
        geom_boxplot(mapping = aes(x = class, y = hwy, colour = class)) +
        labs(x = "Vehicle class",
             y = "Highway mpg") +
        coord_flip() +
        theme_bw()
# plot C
p3 <- ggplot(data = mpg) + 
        geom_jitter(mapping = aes(x = cty, y = hwy, colour = class)) +
        labs(x = "Vehicle class",
             y = "Highway mpg") +
        theme_bw()

# I create three plots names "p1", "p2", and "p3", using previous examples to guide me if necessary. For p1 and p2 I add `coord_flip()` to make the categories easier to read.

D. Use ggarrange() to arrange the three plots in one space. ggarrange() takes the plots as arguments as well as ncol or nrow to customise the arrangement

ggarrange(p1, p2, p3, 
          ncol = 2, 
          nrow = 2)

# The only necessary arguments are the plots, but specifying `ncol` and `nrow` can improve the appearance. the repetition of the legend is usually something we want to avoid.

E. Repeat the code from the previous question, adding common.legend = TRUE and legend = "bottom" to ggarrange().

ggarrange(p1, p2, p3, 
          nrow = 2, 
          ncol = 2, 
          vjust = 0.5, 
          common.legend = TRUE,
          legend = "bottom")

# Setting `common.legend = TRUE` gets rid of the repetitive legend and slightly improves the look. In this example I move the legend to the bottom position, but "top", "left", and "right" are other options.

End of practical