4 Data visualisation: ggplot2

library("tidyverse")

4.1 Why visualising data?

4.1.1 The Anscombe’s Quartet

In Anscombe, F. J. (1973). “Graphs in Statistical Analysis” there was the following dataset:

quartet <- read_csv("https://raw.githubusercontent.com/agricolamz/2020.02_Naumburg_R/master/data/anscombe.csv")
quartet

quartet %>% 
  group_by(dataset) %>% 
  summarise(mean_X = mean(x),
            mean_Y = mean(y),
            sd_X = sd(x),
            sd_Y = sd(y),
            cor = cor(x, y),
            n_obs = n()) %>% 
  select(-dataset) %>% 
  round(2)

Let’s visualise those datasets:

4.1.2 The DataSaurus

In Matejka and Fitzmaurice (2017) “Same Stats, Different Graphs” there are the following datasets:

datasaurus <- read_csv("https://raw.githubusercontent.com/agricolamz/2020.02_Naumburg_R/master/data/datasaurus.csv")
datasaurus

And… all discriptive statistics are the same!

datasaurus %>% 
  group_by(dataset) %>% 
  summarise(mean_X = mean(x),
            mean_Y = mean(y),
            sd_X = sd(x),
            sd_Y = sd(y),
            cor = cor(x, y),
            n_obs = n()) %>% 
  select(-dataset) %>% 
  round(1)

4.2 Basic ggplot2

ggplot2 is a modern tool for data visualisation. There are a lot of extentions for ggplot2. There is also a cheatsheet on ggplot2. There is also a whole book about ggplot2 (Wickham 2016).

Every ggplot2 plot has three key components:

• data,
• A set of aesthetic mappings between variables in the data and visual properties, and
• At least one layer which describes how to render each observation. Layers are usually created with a geom_...() function.

4.2.1 Scatterplot

I downloaded a Polish dictionary from here. I removed all abbreviations and proper names and took only one form from the paradigm. After all this I calculated the number of syllables (simply by counting vowels, combinations of i and other vowels I counted as one), number of symbols in each word and extracted the first letter. Here is the result dataset.

Download this dataset to the variable polish_dictionary. How many words are there?

So this data could be visualised using the following code:

• ggplot2

ggplot(data = polish_dictionary, aes(x = n_char, y = n_vowels)) +
  geom_point()

• dplyr and ggplot2

polish_dictionary %>%
  ggplot(aes(x = n_char, y = n_vowels))+
  geom_point()

4.2.1.1 Layers

All commands in ggplot2 are separated by + sign (author of the package, Hadley Wickham, deeply regrets that it is not %>%), but their order matters:

polish_dictionary %>%
  ggplot(aes(n_char, n_vowels))+
  geom_point()+
  geom_smooth()

polish_dictionary %>%
  ggplot(aes(n_char, n_vowels))+
  geom_smooth()+
  geom_point()

4.2.1.2 aes()

Since every ggplot2 plot has data as a key component there is a function aes() that maps variables from dataframe into visual properties of the graph. There is a simple rule:

If values are from dataframe put them into aes(), otherwise — don’t.

polish_dictionary %>%
  ggplot(aes(n_char, n_vowels, color = first_letter))+
  geom_point()

polish_dictionary %>%
  ggplot(aes(n_char, n_vowels))+
  geom_point(color = "green")

polish_dictionary %>%
  ggplot(aes(n_char, n_vowels))+
  geom_point(aes(color = first_letter))

There are some other possibilities to mark categories:

• with shape argument

polish_dictionary %>%
  filter(first_letter == "a" |
           first_letter == "i" |
           first_letter == "u") %>% 
  ggplot(aes(n_char, n_vowels, shape = first_letter))+
  geom_point()

• with label argument and geom_text()

polish_dictionary %>%
  filter(first_letter == "a" |
           first_letter == "i" |
           first_letter == "u") %>% 
  ggplot(aes(n_char, n_vowels, label = first_letter))+
  geom_text()

• with opacity argument

polish_dictionary %>%
  filter(first_letter == "a" |
           first_letter == "i" |
           first_letter == "u") %>% 
  ggplot(aes(n_char, n_vowels))+
  geom_point(alpha = 0.1)

Sometimes annotations overlap:

polish_dictionary %>%
  slice(8400:8450) %>%  # lets pick 50 words from our dictionary
  ggplot(aes(n_char, n_vowels, label = word))+
  geom_text()

Then it is better to use geom_text_repel() from the ggrepel library (do not forget to download it using install.packages("ggrepel")):

library("ggrepel")
polish_dictionary %>%
  slice(8400:8450) %>% 
  ggplot(aes(n_char, n_vowels, label = word))+
  geom_text_repel()

It looks better, when you add some points:

polish_dictionary %>%
  slice(8400:8450) %>% 
  ggplot(aes(n_char, n_vowels, label = word))+
  geom_text_repel()+
  geom_point()

4.2.1.3 Annotate labels, axis, caption etc.

polish_dictionary %>%
  slice(8400:8450) %>% 
  ggplot(aes(n_char, n_vowels, label = word))+
  geom_text_repel()+
  geom_point()+
  labs(x = "number of characters",
       y = "number of vowels",
       title = "Correlation of number of characters and vowels",
       subtitle = "data from sjp.pl",
       caption = "this graph is created with ggplot2")

Download this dataset and create a scatterplot. What is there?

4.2.2 Barplots

The same data can be aggregated and non-aggregated:

misspelling <- read_csv("https://raw.githubusercontent.com/agricolamz/DS_for_DH/master/data/misspelling_dataset.csv")
misspelling

• variable spelling is aggregated: for each value of speeling variable there is a corresponding value in count variable.
• variable correct is non-aggregated: there isn’t any variable associated with counts of correct variable

In order to create a bar plot from aggregated data you need to use geom_col():

misspelling %>% 
  slice(1:20) %>% 
  ggplot(aes(spelling, count))+
  geom_col()

Lets flip axes:

misspelling %>% 
  slice(1:20) %>% 
  ggplot(aes(spelling, count))+
  geom_col()+
  coord_flip()

In order to create a bar plot from aggregated data you need to use geom_bar():

misspelling %>% 
  ggplot(aes(correct))+
  geom_bar()

Lets flip axes:

misspelling %>% 
  ggplot(aes(correct))+
  geom_bar()+
  coord_flip()

Non-aggregated data could be transformed into aggregated

misspelling %>% 
  count(correct)

Aggregated data could be transformed into non-aggregated

misspelling %>% 
  uncount(count)

Coloring bars actually should be done with fill argument. Compare:

misspelling %>% 
  ggplot(aes(correct))+
  geom_bar(color = "navy")+
  coord_flip()

misspelling %>% 
  ggplot(aes(correct))+
  geom_bar(fill = "navy")+
  coord_flip()

The same argument could be used in the aes() function:

misspelling %>% 
  ggplot(aes(correct, fill = correct))+
  geom_bar()+
  coord_flip()

4.2.2.1 Factors

All variables in the previous section are ordered alphabetically. In order to create your own orders we need to look at factors:

my_factor <- factor(misspelling$correct)
head(my_factor)

## [1] deschanel deschanel deschanel deschanel deschanel deschanel
## 15 Levels: deschanel galifianakis johansson kaepernick labeouf ... shyamalan

levels(my_factor)

##  [1] "deschanel"    "galifianakis" "johansson"    "kaepernick"   "labeouf"     
##  [6] "macaulay"     "mcconaughey"  "mcgwire"      "mclachlan"    "minaj"       
## [11] "morissette"   "palahniuk"    "picabo"       "poehler"      "shyamalan"

levels(my_factor) <- rev(levels(my_factor))
head(my_factor)

## [1] shyamalan shyamalan shyamalan shyamalan shyamalan shyamalan
## 15 Levels: shyamalan poehler picabo palahniuk morissette minaj ... deschanel

misspelling %>% 
  mutate(correct = factor(correct, levels = c("deschanel",
                                              "galifianakis",
                                              "johansson",
                                              "kaepernick",
                                              "labeouf",
                                              "macaulay",
                                              "mcgwire",
                                              "mclachlan",
                                              "minaj",
                                              "morissette",
                                              "palahniuk",
                                              "picabo",
                                              "poehler",
                                              "shyamalan",
                                              "mcconaughey"))) %>% 
  ggplot(aes(correct))+
  geom_bar()+
  coord_flip()

There is a package forcats for factors (it is in tidyverse, here is a cheatsheet). There are a lot of useful functions in forcats, but the one I use the most is the fct_reorder() function:

misspelling %>% 
  count(correct)

misspelling %>% 
  count(correct) %>% 
  ggplot(aes(fct_reorder(correct, n), n))+
  geom_col()+
  coord_flip()

There is an article on Pudding about English pubs. Here is an aggregated dataset, that they used. Visualise the 30 most popular pub’s names in UK.

📋 list of hints ➡

👁 How to get this counts? ➡

Use the count function.

👁 Why there are so many values? ➡

In the task I asked you to take only 30 of them. Maybe you need the slice() function in order to do it.

👁 Why there are pubs with count 1 on my graph?. ➡

By default the count function does not sort anything, so you get only pubs with frequency 1 from the slice() function. In order to sort your values you need to use the arrange() function or use an additional sort = TRUE argument in the count() function.

👁 It looks like I’ve finished. ➡

Have you removed your x and y axes’ annotation? Have you added the caption?

4.3 Faceting

Faceting – is a really powerful tool for data exploration. This function splits visualisations into subplots using some variables.

misspelling %>% 
  filter(count > 500) %>% 
  ggplot(aes(fct_reorder(spelling, count), count))+
  geom_col()+
  coord_flip()

misspelling %>% 
  filter(count > 500) %>% 
  ggplot(aes(fct_reorder(spelling, count), count))+
  geom_col()+
  coord_flip()+
  facet_wrap(~correct)

By default facet_wrap() creates the same scale for all facets. This could be changed by argument scales:

misspelling %>% 
  filter(count > 500) %>% 
  ggplot(aes(fct_reorder(spelling, count), count))+
  geom_col()+
  coord_flip()+
  facet_wrap(~correct, scales = "free")

It is also possible to add multiple variables:

diamonds %>% 
  ggplot(aes(carat, price))+
  geom_point(size = 0.3)+
  facet_wrap(~color+cut, scales = "free")

There is a way to make it more compact using the facet_grid() function instead of the facet_wrap() function:

diamonds %>% 
  ggplot(aes(carat, price))+
  geom_point(size = 0.3)+
  facet_grid(cut~color, scales = "free")

It is also possible to create a marginal summary with the margins argument of the facet_grid() function :

diamonds %>% 
  ggplot(aes(carat, price))+
  geom_point(size = 0.3)+
  facet_grid(cut~color, scales = "free", margins = TRUE)

Use the polish_dictionary and reproduce the following graph.

References

Wickham, Hadley. 2016. Ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag New York. https://ggplot2.tidyverse.org.