Data visualisation

Facets

Creating sub-plots based on the rodent_type variable uses facets. If we are faceting plots by a single variable we use facet_wrap(). The first argument is ~ followed by the variable name, here rodent_type. This argument is called a formula. There are many additional arguments that can be provided, see ?facet_wrap.

Using facet_wrap(~ rodent_type) gives us this plot:

ggplot(data = by_quarter,
       mapping = aes(x=quarter,y=captures,colour=plot_type)) +
  geom_line() +
  geom_point() +
  facet_wrap(~ rodent_type)

So this is starting to look more like the published figure, and now we can confirm that the granivore populations do increase when the kangaroo rats are excluded.

However we can improve things, and we’ll come back making the figure more polished at the end of this lesson.

If we wanted to plot the combination of two variables we’d use facet_grid() in which the formula becomes first_variable ~ second_variable.

Challenge

Try plotting by_quarter using facet_grid() with the variables plot_type and rodent_type as arguments as described above for plotting the combination of two variables

We don’t need to map the colour aesthetic. The plot should look like the one below:

Boxplots

Box-plots are a standard way to plot the distribution of a data set. Plotting the data this way means that we drop the dynamic information, the time dimension, but gain a compact summarised view of the variability of the number of captures.

In the following code we use the boxplot() geom, and this time swap the mapping such that colour maps to the rodent type, and plots are faceted according to the plot type.

ggplot(data= by_quarter,
       mapping = aes(x = rodent_type, y = captures, 
                     colour = rodent_type)) +
  geom_boxplot() +
  facet_wrap(~ plot_type)

Position adjustments

Another common adjustment we might make to our plots is to adjust the position of the data.

Recall the mpg data set for cars and our plot showing the fuel efficiency versus the size of the engine. Many of the points are actually over plotted such that we can’t see them.

By adding a bit of random noise to the position each data point, called jitter we can spread the points out and see them more clearly. This is seen in the second plot below, where the postion = "jitter" argument has been provided to the geom_point() function.

It’s a little counter intuitive to add noise to a plot, but when we’re trying to explore data for patterns it can be very useful.

# Without jitter
ggplot(data = mpg) + 
  geom_point(mapping = aes(x = displ, y = hwy, colour = class))

# With jitter
ggplot(data = mpg) + 
  geom_point(mapping = aes(x = displ, y = hwy, colour = class),
             position = "jitter")

Let’s add points to our box-plot and for the challenge try comparing the plot to one where the points are jittered.

# Boxplot without jitter
ggplot(data= by_quarter,
       mapping = aes(x = rodent_type, y = captures, 
                     colour = rodent_type)) +
  geom_boxplot() +
  geom_point() +
  facet_wrap(~ plot_type)

Challenge

Plot the previous boxplot with jittered points

The plot should look like the one below.

Statistical transformations

Next we’ll plot bar charts.

These simple plots actually reveal subtleties in plots. In the following code we take the by_month_genus subset and create a bar chart of genus, filling the bars with colour according to rodent_type.

# Create a bar plot of genus from by_month_genus 
ggplot(by_month_genus) +
  geom_bar(mapping = aes(x = genus, fill=rodent_type)) + 
  facet_grid(~ plot_type)

We see that the code has automatically created a count variable on the y-axis for each genus plotted as a bar on the x-axis. (We’ll come back to the issue of illegible labels shortly.)

This is because the geom_bar() algorithm automatically performs a statistical transformation of the mapped variable. That is to say it performs a count and bins the data according to the genus. This means we have a chart where the bar height corresponds to the number of rows for each genus.

But this isn’t a count of the captures, it’s how many rows each genus appears in of by_month_genus. Remember how we summarised this data to calculate captures per month.

Often when people talk about bar charts they are referring to when the height of the bar is a variable present in the data, here that would be the captures column. In this kind of situation we need to map the y-axis too, and change the default geom_bar() statistical transformation to stat = "identity" so that it takes the values provided in the captures column to determine the height of the bar.

To reiterate: if you are trying to create a bar chart using x and y variables contained in your data set, you need to set geom_bar(stat = "identity").

# Create a bar plot of genus from by_month_genus using y = captures and 
# stat = "identity"
ggplot(by_month_genus) +
  geom_bar(mapping = aes(x = genus, y = captures,
                         fill=rodent_type), stat = "identity") + 
  facet_grid(~ plot_type)

Challenge

Let’s compare the plot we just made with by_month_genus to one using surveys_subset

Use the code from the plot where we didn’t use the captures column and use the default statistical transformation.

As surveys_subset has one row for every capture, counting the rows should produce the same plot as when we used the calculated captures column in by_month_genus.

To learn more about the 20 statistical transformations built into ggplot2 have a look at the ggplot2 cheatsheet.

Challenge

For the next part we need to calculaute the mean monthly captures.

Recall how to do grouped summaries and create an object by_month_mean using by_month_genus, grouping by genus,plot_type,rodent_type and then summarising mean_captures = mean(captures).

Using our new summary data frame by_month_mean we’ll look at using positions with bar charts.

The next few graphs illustrate a stacked bar chart situation. Here we plot the mean monthly number of captures at each plot type, colouring by genus.

In the first plot the bars are stacked such that we can see the relative mean monthly captures per genus, but not what the actual mean was for each genus.

To get round this problem we can alter the default position setting to identity. However, this means that filled bars would overlap, so we’ve set fill = NA. Now we can see the mean values as well as the proportions.

# Create a stacked bar chart of plot type coloured by genus
ggplot(by_month_mean) +
  geom_bar(mapping = aes(x = plot_type, y = mean_captures, 
                         colour = genus),
           fill = NA, stat = "identity")

# Use position = "identitiy" to create overlapping bars
ggplot(by_month_mean) +
  geom_bar(mapping = aes(x = plot_type, y = mean_captures, 
                         colour = genus),
           fill = NA, stat = "identity", position ="identity")

Alternatively we could set postion = dodge and we get the next plot, where the bars for each genus are side by side, again allowing us to see both the proportions and the mean values.

# Use position = "dodge" to create side by side bars
ggplot(by_month_mean) +
  geom_bar(mapping = aes(x = plot_type, y = mean_captures, 
                         colour = genus),
           fill = NA, stat = "identity", position ="dodge")

For more positional adjustments, check out the help files ?postion_jitter, ?position_dodge, ?position_fill, ?position_identity, and ?position_stack.

Coordinate adjustments

Previously, we plotted by_month_genus and the genus labels we overlapping. We could change the label size, but to illustrate another function, let’s flip the plot round using coord_flip() to improve things:

# Flip the coordinates to make the labels readable 
ggplot(by_month_genus) +
  geom_bar(mapping = aes(x = genus, y = captures, fill = rodent_type), 
           stat = "identity") + 
  facet_grid(~ plot_type) +
  coord_flip()

Factors

Finally for this plot, it would be nice to put the bars in order of size. We can do this by converting the genus to a factor.

Recall that factors are how R represents categorical variables, variables with a limited number of values, such as genus.

Also recall factors look like strings, but behave like integers. That is to say they are strings of characters with associated values called levels that can be used to order catergories.

This is why factors are useful, especially when we want to place things in non-alphabetical order. We can take advantage of factor levels to order things as we wish.

Check out the forcats tidyverse package to explore the power of factors, but here to give you a taste of what is possible we’re going to use the forcats package fct_reorder() function to convert genus into a factor and in doing so set the level according to to the number of captures. In this way we can order the bars in our plot according to the number of captures and see the pattern more clearly.

In fct_reorder(), the first argument is the variable we wish to make into a factor genus, and the second argument is variable we wish to use to create the order. Here we want to order according to the number of captures, so captures is the second argument.

Note we still keep x and y mappings, but now x is now a function of both the genus and the capture columns.

This gives us a nice ordered bar chart by genus and number of captures during the experimental period.

We can clearly see that kangaroo rat exclosure works, and that this corresponds with an increase in granivore captures these plots.

# Load the forcats tidyverse package
library(forcats)

# Order the rodents according to number of captures
ggplot(by_month_genus) +
  geom_bar(mapping = aes(x = fct_reorder(genus, captures), y = captures, 
                          fill = rodent_type), stat = "identity") + 
  facet_grid(~ plot_type) +
  coord_flip()

Challenge

To illustrate using the size aesthetic and factors, recreate the by_month_mean plot using geom_point() with the additional argument of size = rodent_type.

Use fct_reorder() to change the x mapping to a factor that is a function of genus and mean_captures such that the plot is ordered by mean_captures.

Don’t forget to change fill to colour when changing from geom_bar() to geom_point().

Changing colours, labels, themes and adding titles

Recall our box-plot with points overlaid. Following the facet function, we’re going add a succession of functions to change the colours scale_colour_brewer(), change the axis labels, add a title using labs() and change the theme.

Themes deal with non-data elements e.g. the grid, the axes etc. There are a range of built-in themes and packages of themes such as ggthemes. Here we’ll use the built-in theme, theme_bw() that gives a simple layout.

ggplot(data= by_quarter,
       mapping = aes(x = rodent_type, y = captures, 
                     colour = rodent_type)) +
  geom_boxplot() +
  geom_point(position = "jitter") +
  facet_wrap(~ plot_type) +
  scale_colour_brewer(palette = "Set1") +
  labs(title = "Quarterly average captures of rodents between 1977 and 1990",
       x = "Rodent type",
       y = "Number of captures",
       colour = "Rodent type") +
  theme_bw()

Check out ?scale_colour_brewer and ?scale_fill_brewer for more information on colour schemes for discrete data. And ?scale_color_continuous for creating colour gradients for continuous data.