Rapid data visualization with ggplot2

Charles Martin

September 2018

Clarifications

Everything you’ll see in this post could also be done with base graphics in R.

The advantage of using ggplot2 is that you’ll spend much less time programming (i.e. doing loops, managing arrays, conditions, etc.) and much more time actually visualising data.

Also, this workshop is not about plot customization. The plots might not be exactly to your liking (e.g. color choice, symbols, etc.). It is only about data exploration.

Tips to enjoy your time with ggplot2

Our dataset

library(ggplot2)
data(msleep)
summary(msleep)

     name              genus               vore          
 Length:83          Length:83          Length:83         
 Class :character   Class :character   Class :character  
 Mode  :character   Mode  :character   Mode  :character  




    order           conservation        sleep_total      sleep_rem    
 Length:83          Length:83          Min.   : 1.90   Min.   :0.100  
 Class :character   Class :character   1st Qu.: 7.85   1st Qu.:0.900  
 Mode  :character   Mode  :character   Median :10.10   Median :1.500  
                                       Mean   :10.43   Mean   :1.875  
                                       3rd Qu.:13.75   3rd Qu.:2.400  
                                       Max.   :19.90   Max.   :6.600  
                                                       NA's   :22     
  sleep_cycle         awake          brainwt            bodywt        
 Min.   :0.1167   Min.   : 4.10   Min.   :0.00014   Min.   :   0.005  
 1st Qu.:0.1833   1st Qu.:10.25   1st Qu.:0.00290   1st Qu.:   0.174  
 Median :0.3333   Median :13.90   Median :0.01240   Median :   1.670  
 Mean   :0.4396   Mean   :13.57   Mean   :0.28158   Mean   : 166.136  
 3rd Qu.:0.5792   3rd Qu.:16.15   3rd Qu.:0.12550   3rd Qu.:  41.750  
 Max.   :1.5000   Max.   :22.10   Max.   :5.71200   Max.   :6654.000  
 NA's   :51                       NA's   :27

83 observations about mammal sleep habits.

A first plot

ggplot(data = msleep) +
  geom_point(mapping = aes(x = sleep_rem, y = awake))

Warning: Removed 22 rows containing missing values (geom_point).

ggplot: creates a graphic objet, and associates it with the given dataset. Graphic layers are then added upon it.

mapping: created links between variables in the dataset and visual properties from the plot

N.B. ggplot warns us that some rows were removed because they contained missing values for either sleep_rem or awake. We’ll see how to filter them in a future workshop.

What other visual properties are availble?

Among others…

To view all visual properties for a graphic layer, look for the Aesthetics section in the layer’s help page (?geom_point)

Example

We improve the above graphic by adding a color representing the feeding type (vore) and by linking the point size to the animal body size (bodywt)

ggplot(data = msleep) +
  geom_point(mapping = aes(
    x = sleep_rem,
    y = awake,
    color = vore,
    size = bodywt
  )
)

To modify all points at once

To change some graphic properties of all points at once (instead of associating them with values from their respective observations), one needs to specify the association outside of the aes block:

ggplot(data = msleep) +
  geom_point(
    mapping = aes(
      x = sleep_rem,
      y = awake,
      size = bodywt
    ),
    color = "blue",
    shape = 17
)

Exercise #1

Create a plot of mammal brain size (brainwt) in relationship with the animal size (bodywt)

Replace points with squares

The color of every square must show the conservation status of the species (conservation)

Where ggplot2 really shines

Let’s say you want to explore the relationship between the time spent awake by each species and the REM sleep duration…

ggplot(data = msleep) +
  geom_point(mapping = aes(
    x = sleep_rem,
    y = awake
  )
)

This looks interesting, but you are interested to know if the sleep habits vary in the same way between feeding types. So, you add color to explore that.

ggplot(data = msleep) +
  geom_point(mapping = aes(
    x = sleep_rem,
    y = awake,
    color = vore
  )
)

Right at this moment, a co-worker sees the plot above your shoulder, and suggests you to place each feeding type in it’s own panel to clarify things up.

No problem, a single line of ggplot2 does the trick:

ggplot(data = msleep) +
  geom_point(mapping = aes(
    x = sleep_rem,
    y = awake
  )) +
  facet_wrap(~vore)

Ok, no, your previous plot was better, but can you add a smoothing curve per feeding type to see if the trend is the same?

No problem, it’s no biggie to remove you panels, just remove that single ggplot2 command. Then you add a smoothing layer above your points:

ggplot(data = msleep) +
  geom_point(mapping = aes(
    x = sleep_rem,
    y = awake,
    color = vore
  )) +
  geom_smooth(mapping = aes(
    x = sleep_rem,
    y = awake,
    color = vore
  )
  )

`geom_smooth()` using method = 'loess' and formula 'y ~ x'

N.B. Here you’ll have many warnings because our dataset is way too small to calculate smoothing curves correctly for each feeding type. It’s only a toy example.

Well, that is clearly unreadble right? Can you just put linear regressions in there instead?

Yep, no problem, just edit your smoothing layer to use to lm function. You also strategically remove the standard error bands because they are all stacked up one above the other…

ggplot(data = msleep) +
  geom_point(mapping = aes(
    x = sleep_rem,
    y = awake,
    color = vore
  )) +
  geom_smooth(
    mapping = aes(
      x = sleep_rem,
      y = awake,
      color = vore
    ),
    method = "lm",
    se = FALSE
  )

To remove code duplication, use global mappings

If you need to reuse the same graphic property-dataset variable associations in many layers as we did above, you can put all your common associations inside the initial ggplot call:

ggplot(data = msleep, mapping = aes(
    x = sleep_rem,
    y = awake,
    color = vore
  )) +
  geom_point() +
  geom_smooth(
    method = "lm",
    se = FALSE
  )

To densify your code even more

You can take advantage of the fact that in R, the name of the arguments given to a function is optional, as long as you keep your arguments in the same order as specified in the help file (?ggplot):

Usage

ggplot(data = NULL, mapping = aes(), ..., environment = parent.frame())

Which allows you to do:

ggplot(msleep, aes(
    x = sleep_rem,
    y = awake,
    color = vore
  )) +
  geom_point() +
  geom_smooth(
    method = "lm",
    se = FALSE
  )

A bestiary of ggplot graphic layers

Visualizing a single continuous variable

ggplot(msleep) +
  geom_histogram(aes(x = awake))

`stat_bin()` using `bins = 30`. Pick better value with `binwidth`.

Notice the warning about the number of bins used. In contrary to the base graphics hist function, histograms in ggplot don’t use algorithms to determine the ideal number of bins. It is your job to try different values to determine what fits your particular case.

Visualizing a categorical / discrete variable

ggplot(msleep) +
  geom_bar(aes(x = vore))

Note that, if your totals are already calculated, you need to use an alternate geom:

sums <- data.frame(
  total = c(4,1,2),
  labels = c("Control", "pH+", "pH-")
)
ggplot(sums) +
  geom_col(aes(x = labels, y = total))

Visualizing the relationship between two continous variables

ggplot(msleep) +
  geom_point(aes(x = sleep_rem, y = awake))

You can also replace points with text

ggplot(msleep) +
  geom_text(aes(x = sleep_rem, y = awake, label = genus))

Visualizing the relationship between a continuous and a categorical variable

ggplot(msleep) +
  geom_boxplot(aes(x = vore, y = awake))

Note that a boxplot only expresses data about 5 of your data points (median, 1st quartile, 3rd quartile and biggest data within 1.5 IQR).

A more informative view about the whole distribution can be obtained with a violin plot

ggplot(msleep) +
  geom_violin(
    aes(x = vore, y = awake)
  )

Visualizing the relationship between two categorical variables

ggplot(msleep) +
  geom_bar(aes(x = vore, fill = conservation))

Position modifiers

You can use position modifiers to organize this bar plot in alternate ways:

ggplot(msleep) +
  geom_bar(
    aes(x = vore, fill = conservation),
    position = "dodge"
  )


ggplot(msleep) +
  geom_bar(
    aes(x = vore, fill = conservation),
    position = "fill"
  )

Visualizing error / uncertainty

For this section, we’ll need an additional dataset, where we have observed values in x, predicted values in y, and the standard error of these predicted values.

df <- data.frame(
  x = c(1,2,3,4),
  y = c(1.1,1.9,3.4,4),
  se = c(0.4, 0.5, 0.7, 0.5)
)

ggplot(df, aes(x = x, y = y)) +
  geom_point() +
  geom_errorbar(
    aes(ymin = y - se, ymax = y + se),
    width = 0.3
  )

You’ll notice that we need to manually construct the size of the error bar, using the information we have about the predicted values and their error. We could as well use 1.96 * se, etc.

ggplot(df, aes(x = x, y = y)) +
  geom_point() +
  geom_linerange(
    aes(ymin = y - se, ymax = y + se)
  )


ggplot(df, aes(x = x, y = y)) +
  geom_ribbon(
    aes(ymin = y - se, ymax = y + se),
    fill = "tomato"
  ) +
  geom_line()

The ultimate data exploration tool

If you want to explore all variables and relationships in a dataset at once, the ggpairs function from the GGally package does just that.

Note that you need to filter out columns/variables that are not numerical or categorical (i.e. species or site names, etc.)

library(GGally)
ggpairs(msleep[,-c(1,2,4)])


ggpairs(msleep[,-c(1,2,4)], aes(col = vore))

NB Please be patient with ggpairs, it might need a minute or two to compute your plot, as it is effectively doing tens or even sometimes hundred of plots depending on the number of variables

Before you tweak your plots, learn how to properly save them…

First of all, use the proper ggplot function (ggsave) to save your plots. It offers much more control over the quality and features of the ouput than the Export feature in RStudio or related functionalities.

The ggsave function sends the last plot to a file, with the correct format based on the file extension given (e.g. gif, png, jpg, pdf, eps)

ggplot(msleep) +
  geom_point(aes(x = sleep_rem, y = sleep_cycle))

ggsave(filename = "Resultats/Fig1.jpg")

Saving 7 x 5 in image

Note that to save your files in a folder, ggsave insists that you manually create the folder first.

Which dimensions should I use?

There is no easy answer here. It is a trial and error process. ggplot changes the relative size of points, text, lines, etc. based on the given dimensions. You’ll need to experiment.

Here, for example, the same plot is saved in a 2x2 file and and 8x8 file

ggsave(filename = "Resultats/2x2.jpg", width = 2, height = 2)
ggsave(filename = "Resultats/8x8.jpg", width = 8, height = 8)


Default units are inches, but you can change them to cm with the units="cm" argument

How to change the file resolution

For pixel-based files (e.g. jpg, png, gif), you can also specify image quality, in number of pixels per inches (dot per inches; dpi)

For example, the same 2x2 plot can be extremely pixelized or ultra-sharp depending or the resolution selected:

ggsave(filename = "Resultats/72.jpg", width = 2, height = 2, dpi = 72)
ggsave(filename = "Resultats/1200.jpg", width = 2, height = 2, dpi = 1200)


For printed publications, a minimum of 300 dpi is usually recommended.

On last thing… the gray background!

One of the most controversial aspects of ggplot is the use of a gray background with white grid lines. For a rapid overview of the reasons Hadley Wickham used such a color scheme, you can read a free excerpt of its excellent R for Data Science book (from which the flow of this workshop was heavily borrowed)

Nonetheless, you can easily change the theme of a ggplot by adding a theme layer.

ggplot(msleep) +
  geom_point(aes(x = sleep_rem, y = sleep_cycle)) +
  theme_classic()


ggplot(msleep) +
  geom_point(aes(x = sleep_rem, y = sleep_cycle)) +
  theme_dark()

And my personnal favorite, the minimal theme:

ggplot(msleep) +
  geom_point(aes(x = sleep_rem, y = sleep_cycle)) +
  theme_minimal()