4  Data visualisation with ggplot2

For this part we use the FalseBeginners dataset. The dataset is based on De Wilde et al. (2019), which is available via: https://osf.io/ndr47/. I have selected and renamed the variables for our purposes here. Table 4.1 provides an overview of the variables.

Table 4.1: Codebook Falsebeginners

ID	Variable	Description
1	Student	A unique identifier for each student
2	School	A unique identifier for each school
3	Class	A unique identifier for each class
4	PPVT	Score for the Peabody Picture Vocabulary Test (PPVT). Maximum score = 120
5	Speaking	Score for the speaking test. Maximum score = 20
6	Listening	Score for the listening test. Maximum score = 25
7	ReadingWriting	Score for the reading and writing tests. Maximum score = 50
8	Attitude	Attitude of student towards English: Positive vs. Negative
9	L1	L1: Dutch vs. Multilingual
10	Sex	Sex of student: Male vs. Female

The file is a tab delimited .txt-file. We open the file with read.delim(). The dataset is organised in long data format.

library(tidyverse)

1fb <- read.delim("FalseBeginners.txt",
2                 header = TRUE,
3                 na.strings = NA,
4                 stringsAsFactors=TRUE)

1

the text file to open,

2

the first row of the file contains the variables names,

3

missing values are marked as “NA”,

4

strings are interpreted as factors.

As always, we start with a global summary.

options(width = 70) 
summary(fb)

    Student          School        Class          PPVT      
 Min.   :  2.0   S12    : 51   9A     : 29   Min.   : 31.0  
 1st Qu.:219.8   S34    : 49   12A    : 27   1st Qu.: 69.0  
 Median :445.5   S33    : 37   11A    : 24   Median : 78.0  
 Mean   :441.1   S38    : 36   12B    : 24   Mean   : 78.6  
 3rd Qu.:660.2   S51    : 36   42A    : 24   3rd Qu.: 88.0  
 Max.   :867.0   S8     : 35   21A    : 21   Max.   :116.0  
                 (Other):536   (Other):631   NA's   :1      
    Speaking        Listening     ReadingWriting      Attitude  
 Min.   : 0.000   Min.   : 0.00   Min.   : 0.00   negative: 27  
 1st Qu.: 2.000   1st Qu.:10.00   1st Qu.:13.00   positive:733  
 Median : 5.000   Median :15.00   Median :18.00   NA's    : 20  
 Mean   : 6.786   Mean   :14.95   Mean   :21.16                 
 3rd Qu.:10.000   3rd Qu.:20.00   3rd Qu.:29.00                 
 Max.   :20.000   Max.   :25.00   Max.   :50.00                 
 NA's   :13       NA's   :2       NA's   :1                     
     L1          Sex     
 dutch:567   female:378  
 multi:207   male  :402  
 NA's :  6               
                         
                         
                         
                         

We omit all rows with NAs (this means that a row is dropped as soon as it contains one missing value).

fb <- na.omit(fb)

4.1 Histogram

Figure 4.1 shows a histogram for the vocabulary score PPVT.

2ggplot(data = fb,
1       mapping =  aes(x=PPVT)) +
3       geom_histogram(bins = 25)

1

the dataset that we wish to use. Remember that you need to open this data first. In other words, if the dataframe fb does not exist, this code won’t work,

2

mapping & aes indicates the variables for the x- and y-axis (and perhaps other variables for other aesthetics),

3

geom is the kind of visualisation that we wish to use.

Figure 4.1: A histogram of PPVT

The key idea in all ggplot figures is to start from a dataframe or tibble object, choose the variables for the x- and y-axis, and choose a geom, the kind of figure for the visualisation.

Extending the basic code with aesthetic features is quite straightforward. In Figure 4.2, we add some color, a label for the x-axis and use a minimalistic theme.

1ggplot(fb,
2       aes(x=PPVT)) +
3  geom_histogram(bins = 25,
4                 fill = "#F1A42B") +
5  theme_minimal() +
6  xlab("Score PPVT (120)")

1

the data,

2

the variable that we wish to visualise,

3

the visualisation method,

4

with a colour,

5

a general theme (there are many others available to accomodate everyone’s taste),

6

a better label for the x-axis.

Figure 4.2: A histogram for PPVT with some additional aesthetics.

4.2 Boxplot

The code for the boxplot in Figure 4.3 is very similar to that of the histogram. The only things that we have to change is the use of the y-axis rather than the x-axis and the geom.

1ggplot(fb,
2       aes(y=PPVT)) +
3  geom_boxplot() +
4  scale_x_discrete( ) +
5  theme_grey()

1

the data,

2

the variable for the y-axis (one could also create a horizontal boxplot on the x-axis),

3

use a boxplot,

4

change the x-axis for a nicer appearance,

5

use a grey theme.

Figure 4.3: A vertical boxplot of PPVT

4.3 Density curve

We change change the geom and we get the density plot in Figure 4.4.

ggplot(fb, aes(x = PPVT)) +
  geom_density()  +
  theme_minimal()

Figure 4.4: A density curve of PPVT

4.4 Barplot

Before we can create a barplot we need to tabulate the data, which we can do with count(). Then we select the categorical variable levels on the x-axis and their respective counts on the y-axis, and we get Figure 4.5.

1fb |>
2  count(Sex)  |>
3ggplot(aes(x = Sex,
4           y = n)) +
  geom_col()

1

start with fb,

2

count the observations for each level of Sex,

3

use ggplot, put Sex on the x-axis and n (the result of count()) on the y-axis,

4

visualise with a barplot (“col” = column).

Figure 4.5: A barplot of Sex.

We can again add some aesthetics to get Figure 4.6

fb |>                           
  count(Sex)  |>                
ggplot(aes(x = Sex,             
           y = n)) +            
  geom_col(fill = "#1E64C8") +  
  xlab("Sex")  +           
  theme_minimal()                

Figure 4.6: A barplot of Sex with some additional features.

4.5 Overlapping histograms

So far, we have only looked at univariate visualisations. Let’s add an extra dimension. We add Sex as a categorical variable by means of two different colours, as in Figure 4.7. The alpha argument argument adds transparency to the colours. Note that we use “fill” rather than “colour”, because the latter adds colour to the border of the histogram.

ggplot(fb,
       aes(x=PPVT,
           fill = Sex)) +
  geom_histogram(bins = 20,
                 position = "identity",
                 alpha = 0.4) +
  theme_classic()

Figure 4.7: PPVT by Sex.

4.6 Overlapping density plots

Figure 4.8 shows two overlapping density plots. The code is similar except to the one for the overlapping histograms except for the geom.

ggplot(fb,                                    
       aes(x=PPVT,                            
           fill = Sex)) +                     
  geom_density(position = "identity",       
                 alpha = 0.4) +             
  theme_classic()                           

Figure 4.8: PPVT by Sex.

4.7 Multiple boxplots

A clustered boxplot, as in Figure 4.9, is created by adding an extra categorical variable on the x-axis. Here we added some extra colour with the fill argument.

ggplot(fb, aes(x = Sex,           
               y = PPVT,            
               fill = Sex)) +         
  geom_boxplot() +                
  theme_minimal() +                
1  theme(legend.position = "none")

1

A legend is created by default for the colours, but this is redundant because of the labels on the x-axis.

Figure 4.9: PPVT by Sex.

4.8 Boxplots with a facet

Figure 4.10 visualises three variables: the continuous variable PPVT by both Sex and L1. L1 is here added as a facet.

ggplot(fb, aes(x = Sex,           
               y = PPVT,            
               fill = Sex)) +         
  geom_boxplot() +                
  theme_minimal() +   
  facet_wrap(L1 ~ .) +            
  theme_minimal() + 
  theme(legend.position = "none") 

Figure 4.10: PPVT by Sex and L1.

Figure 4.11 replaces L1 by School, which conveniently allows us to compare the Sex-differences over the different schools.

ggplot(fb, aes(x = Sex,           
               y = PPVT,            
               fill = Sex)) +         
  geom_boxplot() +                
  theme_minimal() +   
  facet_wrap( ~ School) +            
  theme_minimal() + 
  theme(legend.position = "none") 

Figure 4.11: PPVT by Sex and School.

The plots suggest that in most schools boys tend to have a higher PPVT than girls.

4.9 Clustered barplot

Is there an association between Sex and Attitude? We can examine the association between two categorical variables by means of clustered barplots, for which we first need to aggregate the data.

sa <- fb |> count(Sex, Attitude) 
sa

     Sex Attitude   n
1 female negative  13
2 female positive 352
3   male negative  13
4   male positive 368

Then we can use the dataframe to create the barplot in Figure 4.12.

ggplot(data=sa, aes(x=Sex,             
                    y=n,               
                    fill=Attitude)) +  
  geom_col(position = "dodge") +       
  theme_minimal()

Figure 4.12: A clustered barplot of Sex and Attitude

We can aggregate the attitudes for both Sexes by School and add a facet for School to the clustered barplot, as in Figure 4.13

ssa <- fb |> group_by(School, Sex) |> count(Attitude)
ggplot(data=ssa, aes(x=Sex,             
                    y=n,                
                    fill=Attitude)) +   
  geom_col(position = "dodge") +        
  facet_wrap( ~ School) + 
  theme_minimal()

Figure 4.13: Attitude by Sex for each School

Overall, both sexes show a very positive attitude towards English.

4.10 Scatterplot

The scatterplot in Figure 4.14 visualises the relation between the scores for PPVT and Speaking.

ggplot(fb, aes(x = PPVT,          
               y = Speaking)) +   
  geom_point()   +                
  theme_minimal()

Figure 4.14: A scatterplot of PPVT and Speaking.

We can add a regressionline with geom_smooth() as in Figure 4.15.

ggplot(fb, aes(x = PPVT,          
               y = Speaking)) +   
  geom_point()   +              
  geom_smooth(method = "lm", col="blue")  +
  theme_minimal()

`geom_smooth()` using formula = 'y ~ x'

Figure 4.15: A scatterplot of PPVT and Speaking with a regression line

Or with a non-parametric smoother, as in Figure 4.16.

ggplot(fb, aes(x = PPVT,          
               y = Speaking)) +   
  geom_point()   +        
  geom_smooth(method = "loess", col="blue")  +
  theme_minimal()

`geom_smooth()` using formula = 'y ~ x'

Figure 4.16: A scatterplot of PPVT and Speaking with a smoother.

And again, it’s straightforward to add extra variables a facet, as in Figure 4.17:

ggplot(fb, aes(x = PPVT,          
               y = Speaking)) +   
  geom_point()   +  
  geom_smooth(method = "loess", col="blue")  +
  facet_grid(Sex ~ L1) + 
  theme_minimal()

`geom_smooth()` using formula = 'y ~ x'

Figure 4.17: A scatterplot of PPVT and Speaking, by Sex and L1.

4.11 Exercises

Exercise 4.1

Use the ggplot2 package to create the following visualisations for the Multitask dataset:

1. Create a histogram of Time
2. Create a boxplot of Time
3. Create a barplot of Location

Exercise 4.2

Use the ggplot2 package to create the following visualisations for the Multitask dataset:

1. Visualise Time by Task as a clustered boxplot
2. Visualise Time by Task and University as overlapping density plots
3. Add a vertical line to visualize the mean Time for both Tasks.
4. Use ggplot() to visualise the paired differences in the multitask dataset
5. Create a new variable Difference, based on the difference in Time between the two Tasks.
6. Visualise the relation between Speaking and Listening.
7. Visualise the relation between Speaking and Listening for both Sexes and L1s separately.

De Wilde, Vanessa, Marc Brysbaert, and June Eyckmans. 2019. “Learning English Through Out-of-School Exposure. Which Levels of Language Proficiency Are Attained and Which Types of Input Are Important?” Bilingualism: Language and Cognition 23 (1): 171–85. https://doi.org/10.1017/s1366728918001062.