library(ggplot2)

2.1 Amount of sugar per McDonald’s category

For some of the workshop, I’ll work with the McDonald’s menu items dataset, which you can access from my wesite. To simplify things a little bit, I’ll take a subset—just four of the nine categories. To make that subset, I’ll use the subset function.

menu <- read.csv("http://joeystanley.com/downloads/menu.csv")
menu_subset <- subset(menu, Category %in% c("Smoothies & Shakes", "Desserts", "Beverages", "Snacks & Sides"))

The default plot will just be the distribution of the number of sugars in each of the four remaining categories. For the color, I’ll use Paul Tol’s themes, which I access using the package ggthemes.

m <- ggplot(menu_subset, aes(Category, Sugars, fill = Category)) +
    geom_boxplot(size = 0.75) +
    geom_jitter(color = "gray15") + 
    ggthemes::scale_fill_ptol()

So now, to plot it, all I need to do is call m.

m

Conveniently for us, if I want to make changes to the plot, I can just add additional lines of ggplot2 code to the p and it’ll work out like normal. In other words…

m + ggtitle("A Default Plot")

…is shorthand for…

ggplot(menu_subset, aes(Category, Sugars, fill = Category)) +
    geom_boxplot(size = 0.75) +
    geom_jitter(color = "gray15") + 
    ggthemes::scale_fill_ptol() + 
    ggtitle("A Default Plot")

Both will produce this plot:

2.2 Stranger Things ratings

Another dataset I’ll use is the Stranger Things dataset that was used in a previous workshop. Like before, I’ll change the season column to a factor too. For the default plot here, s, I’ll make the same scatterplot of the number of votes the episode got on IMDB by the average rating. I’ll make the dots bigger so they’re easier to see. And for fun, I’ll use a Wes Anderson theme inspired by the movie Fantastic Mr. Fox.

stranger <- read.csv("http://joeystanley.com/data/stranger.csv")
stranger$season = factor(stranger$season)
s <- ggplot(stranger, aes(votes, rating, color = season)) + 
    geom_point(size = 4) + 
    scale_color_manual(values = wesanderson::wes_palette("FantasticFox1")) + 
    labs(title = "Stranger Things episodes",
         subtitle = "Average rating by number of votes on IMDB")
s

2.3 Girlnames

The last is a dataset of the top 25 most common baby girl names in the US in 2017. I’ll read it in and prep it as I did in the intro workshop, except I’ll reverse the order of the girlnames and will make a horizontal bar chart. I’ll also only keep the top 15 just so there aren’t so many bars and color it using Color Brewer.

girlnames <- read.delim("http://joeystanley.com/data/girlnames.txt", sep = "\t")
girlnames <- girlnames[1:15,]
girlnames$name = factor(girlnames$name, levels = rev(girlnames$name))
g <- ggplot(girlnames, aes(name, n, fill = n)) + 
    geom_bar(stat = "identity") + 
    scale_fill_distiller(type = "seq", palette = "Purples") + 
    coord_flip() + 
    labs(title = "Top 15 baby girl names in the US in 2017",
         caption = "Data Source: Social Security data via the babynames package")
g

2.4 Cereal nutritional facts

Finally, the last dataset tht we’ll use—which was also made available through Kaggle.com—contains nutritional information from about 80 different kinds of cereal. We’ll load it in, make a few changes as before, and make a faceted plot showing the amount of fiber per cereal and its rating, split up by brand.

cereal <- read.csv("http://joeystanley.com/data/cereal.csv")
cereal$mfr <- forcats::fct_recode(cereal$mfr,
                                  "Kellogg's" = "K",
                                  "General Mills" = "G",
                                  "Post" = "P",
                                  "Quaker Oats" = "Q",
                                  "Ralston Purina" = "R",
                                  "Nabisco" = "N")
c <- ggplot(cereal, aes(fiber, rating)) + 
    geom_text(aes(label = name),
              check_overlap = TRUE, vjust = "inward", hjust = "inward") + 
    facet_wrap(~mfr)
c

Note that within facet_wrap, I’ve added a couple extra arguments (check_overlap, vjust, and hjust). I recently learned about these from Hadley Wichkham’s book-in-progress, ggplot2 (version 3) which you can view here. They ensure that labels don’t overlap (removing some if necessary), and then make sure they don’t spill off over the edges of the plot. Very handy.

3.1 theme_gray()

By default, ggplot2 will display using a theme that is generally pretty good. The creators have taken some liberties to decide a few things, some founded upon principles of good data visualization and others that are more opinionated. In Hadley Wickham’s words:

The theme is designed to put the data forward while supporting comparisons, following the advice of (Tufte 2006; Brewer 1994; Carr 2002, 1994; Carr and Sun 1999). We can still see the gridlines to aid in the judgement of position (Cleveland 1993a), but they have little visual impact and we can easily ‘tune’ them out. The grey background gives the plot a similar typographic colour to the text, ensuring that the graphics fit in with the flow of a document without jumping out with a bright white background. Finally, the grey background creates a continuous field of colour which ensures that the plot is perceived as a single visual entity.

Using the defaut theme will produce beautiful plots. Here’s what that would look like:

m
s
g
c

3.2 Other themes in ggplot2

Fortunately, if you don’t like the look of the gray background, you can easily switch to one of the other seven these that ggplot2 has bult in.

My go-to theme is called theme_bw() for just “black and white.” The biggest difference is that now the background is white instead of gray. But it also adds a thin black border around the whole thing and has faint grey grid lines instead of white ones.

m + theme_bw()
g + theme_bw()
s + theme_bw()
c + theme_bw()

The linedraw theme uses only pure black lines, rather than any gray ones, which makes it kind of look a little vintage. In facets is where I see the biggest change: the boxes at the top (called strips) are black.

m + theme_linedraw()
g + theme_linedraw()
s + theme_linedraw()
c + theme_linedraw()

The light theme is very similar to bw. The biggest difference is the outside box is lighter. In facets, the strips are darker and the text is colored white.

m + theme_light()
g + theme_light()
s + theme_light()
c + theme_light()

The classic theme has no grid and the top and right parts of the box are gone too, just leaving the x and y axes. The facet strips are empty white boxes now.

m + theme_classic()
g + theme_classic()
s + theme_classic()
c + theme_classic()

If you really like the gray, you can go for the dark theme, which has a darker background and a darker gray for the grid lines. In the girl names plot, you can see that the lighter colors pop out more effectively, but the middle ones blend in with the background a little bit, which is something to keep in mind when you add color to your plots.

m + theme_dark()
g + theme_dark()
s + theme_dark()
c + theme_dark()

The minimal theme is even simpler than light. It doens’t have the outside border but it does retain the inside grid. The strips are gone entirely, leaving just the text behind, which is a little confusing with this plot.

m + theme_minimal()
g + theme_minimal()
s + theme_minimal()
c + theme_minimal()

Finally, you can go completely blank with void. This may seem a little weird, but it does have useful purposes, like when creating maps.

m + theme_void()
g + theme_void()
s + theme_void()
c + theme_void()

Something to keep in mind with themes is that they override any other theme layer you might have included. For example, if you want to remove the legend and then use the classic theme, it’s still there:

m + theme(legend.position = "none") + 
    theme_classic()

No, this isn’t a bug. The reason for this is because theme_* is actually a shortcut for a whole bunch of other theme elements. As a result, there are two theme functions, so the second one overrides the first one. Fortunately, we can fix this by just reversing the order of theme and theme_classic:

m + theme_classic() +
    theme(legend.position = "none")

If you’re still relatively new to ggplot2, my recommendation is to try out one of these default themes for a while. Once you’ve gotten some experience with creating your own plots, which will help you spot things you like and dislike in others’ plots, you may get a sense of what you like in a plot and what you want to see changed. If that’s the case, then it might be time to start moving on to creating a custom theme.

3.3 Customizing existing themes

Out of the box, these themes already come with some customization. If you look at the help file for theme_bw, you can see what kinds of arguments it takes: base_size, base_family, base_line_size, and base_rect_size. These arguments have default values, but they can be modified.

With base_size, you can change the overall size of all text from the default of 10. Note that not all text is the same size: the title is bigger than the axis titles, wich are bigger than the axis labels. Changing this value will change them all proportionally. This is a useful change for making pretty big or pretty small graphs.

m + theme_bw(base_size = 5)
g + theme_bw(base_size = 5)
s + theme_bw(base_size = 5)
c + theme_bw(base_size = 5)

m + theme_bw(base_size = 15)
g + theme_bw(base_size = 15)
s + theme_bw(base_size = 15)
c + theme_bw(base_size = 15)

The argument you might change the most is the base_family. If you’re unsatisfied with the fonts, you can choose from a somewhat small set of R fonts and all the text will change.

m + theme_bw(base_family = "Palatino")
g + theme_bw(base_family = "Palatino")
s + theme_bw(base_family = "Palatino")
c + theme_bw(base_family = "Palatino")

With base_line_size, you can change how thick all the lines in the plot are, and with base_rect_size you can change how thick all the rectangle lines are.

m + theme_bw(base_line_size = 5)
g + theme_bw(base_line_size = 5)
s + theme_bw(base_line_size = 5)
c + theme_bw(base_line_size = 5)

m + theme_bw(base_rect_size = 5)
g + theme_bw(base_rect_size = 5)
s + theme_bw(base_rect_size = 5)
c + theme_bw(base_rect_size = 5)

So if you’re mostly satisfied with the themes but just want to tweak something small like the font and/or one of these other options, you can do that quick change and it’ll already look great.

4.1 Drawing lines with element_line

One of the most fundamental concepts in a plot is a line. Assuming we have a starting and stopping point set already by the data we’re plotting, we can think of properties of a line that are purely aesthetic: color, thickness, and what type of line it is (solid, dotted, dashed, etc). In ggplot2, we modify these attributes (and a few others, with element_line). One examlpe of a line in a ggplot is the faint white grid underneath the data. We can modify that grid with the panel.grid argument. We’ll get into more detail about modifying the grid later, but for now we’ll use it as an example of a line. If we want to modify the color of the lines, we add element_line and specify that the color is light blue

m + theme(panel.grid = element_line(color = "lightblue"))

We can add other arguments too:

m + theme(panel.grid = element_line(color = "lightblue", size = 4, linetype = "dotted"))

In fact, if you inspect the element_line function itself, you can see that you can also add things like lineend. With this argument, you can change the ends of the line to can make the ends "round", "butt", or "square". In this example, it’s not very illustrative since the lines extend past the edge of the plotting area, but later when we come across other lines you’d like to modify, you can use that option if you’d like.

However, if you like arrows, you’ll also see that you can add them with the arrow argument. But if you thought you were done with basic ggplot2 elements, it turns out the arrow argument takes the output of the arrow() function, which allows you to modify things like the angle, length, type, and which sides of the line should get the arrow. And if that wasn’t Inception enough for you, the length argument requires the output of the unit function, in which you specify the length and the units.

m + theme(panel.grid = element_line(color = "lightblue", 
                                    arrow = arrow(angle = 20, 
                                                  length = unit(0.2, "inches"), 
                                                  ends = "both", 
                                                  type = "closed")))

This may seem like an overly complex network of functions, but the benefit is that it allows you to modify whatever you want.

4.2 Drawing rectangles with element_rect

One step up from drawing lines is to draw rectanges with element_rect. Because they consist of four lines, many of the arguments are the same as element_line.

c + theme(panel.border = element_rect(color = "darkred", size = 2, linetype = "dashed"))

Oops. Where did the plot go? So turns out rectangles aren’t just lines anymore because they create the space in the middle. So in element_rect we need to worry about what goes in the middle. By default, it seems like the color is white. What we probably want to do is remove it, or rather, make it transparent. We can do that with the fill argument:

c + theme(panel.border = element_rect(color = "darkred", size = 2, linetype = "dashed",
                                      fill = "transparent"))

So in the theme function, there are several things that are rectangles that you can modify. You can probably think of a few now like the legend or the strips in facets. We’ll use element_rect to modify those.

4.3 Drawing text with element_text

The last major element in ggplot is to draw text, which we do with element_text. As you can imagine, there are a lot of things you can change with text, like the font (“family”), it’s “face” ("bold", "italic", "bold.italic"), color, and size. In this example, we’ll modify a plot’s title.

g + theme(plot.title = element_text(family = "Palatino", face = "bold", color = "mediumpurple4", size = 20))

In addition to these attributes, we can also adjust other things. I’m also going to add the extremely handy debug = TRUE argument. This will make the plot title area yellow and will show a yellow dot where the text is anchored. So, for example, if I want to modify the angle to something onconventional, I can do so, but the result might not be very pretty.

g + theme(plot.title = element_text(family = "Palatino", face = "bold", color = "mediumpurple4", size = 20,
                                    angle = 3, debug = TRUE))

You can see that the anchor point is the top left corner of the plotting area. We may want to adjust the vertical position of this title. Both the hjust and vjust arguments take a number, ranging from 0 (left/bottom) to 1 (right/top). When we set the vertical adjustment to 0, it’ll anchor it to the bottom, which is probably what we wanted.

g + theme(plot.title = element_text(family = "Palatino", face = "bold", color = "mediumpurple4", size = 20,
                                    angle = 3, vjust = 0, debug = TRUE))

If you want to center your plot, you can do so by setting hjust to 0.5.

g + theme(plot.title = element_text(family = "Palatino", face = "bold", color = "mediumpurple4", size = 20,
                                    hjust = 0.5, debug = TRUE))

Note that you can use values outside of the range [0,1] and they’ll move around just as you expect:

g + theme(plot.title = element_text(family = "Palatino", face = "bold", color = "mediumpurple4", size = 20,
                                    vjust = -2, hjust = 2, debug = TRUE))

Finally, you may want to adjust the spacing around the text. To illustrate lineheight, I’ll change the title to something with two lines so you can get a better picture of what it’s doing. So if I really want to add double spacing, I can set it to 2. If I want to do something more subtle, maybe like 1.15.

g + labs(title = "Here is a\nTwo-lined Title",
         subtitle = "This is the subtitle") + 
    theme(plot.title = element_text(family = "Palatino", face = "bold", color = "mediumpurple4", size = 20,
                                    lineheight = 2, debug = TRUE))

For titles, this often won’t be necessary, but for other text elements in your plot, the lineheight may be useful.

The last thing is the margin, which takes the output of the margin function. In it, you specify how large the margins should be around your title area. The order is top, right, bottom, left, which you can maybe remember with the mneumonic “trouble”. For the title, left and right margins don’t mean much, but you can adjust how far the title is from the top of the graph or from the plotting area. Of course, if you want it right up against the plotting area, you can set the bottom margin to zero.

g + theme(plot.title = element_text(family = "Palatino", face = "bold", color = "mediumpurple4", size = 20,
                                    margin = margin(1, 0, 0, 0, unit = "cm"), debug = TRUE))

So that’s the element_text function. It’s powerful because you can control a lot of things with it.

4.4 Drawing nothing with element_blank

The final category of ggplot elements is element_blank and it’s pretty simple: it draws nothing. An important detail though is that not only does draw nothing, but it doesn’t even reserve space for that thing. So in the Stranger Things plot, the x- and y-axis labels are useful, but if we wanted to purposely make the plot confusing, we could remove them.

s + theme(axis.title = element_blank())

So if there’s something in your plot by default and you want to remove it, you just simply use element_blank and it’ll zap it from existence.

6.1 Adding with + and replacing with %+replace%

It’s time to start creating your own custom theme During the course of the workshop, I’ve made some changes that I kind of liked, and I’d like to try a theme that incporates those elements.

When building a theme, the simplest way to create your own is to add to an existing theme. Using the +, which is what we’ve been doing this whole time, you can create a theme based on an existing one. Recall that theme_bw has the legend position to the right of the plotting area (i.e. legend.position = "right"). When we add a theme function to theme_bw, it’ll overwrite whatever is there with the elements we specify. Here, I’m changing it so that the position is in the top right. I’m also adding a couple other subtle things like making the background gray, italicizing the caption, and making the title bigger.

g + theme_bw() + 
    theme(legend.background = element_rect(fill = "white", linetype = "solid", color = "gray50", size = 0.5),
          legend.position = c(1,1),
          legend.justification = c(1,1),
          axis.title = element_text(color = "gray10", size = 12, face = "bold"),
          plot.title = element_text(size = 15),
          plot.caption = element_text(face = "italic"),
          plot.background = element_rect(fill = "gray90"))

As an alternative to +, you can use the funny-looking %+replace% function. On the surface, they may seem pretty similar, but there are some important differences.

Let’s say I want create a theme where the y-axis title text is red. When we use +, it behaves as expected. When we use %+replace% there are some additional changes. What’s happening?

g + labs(title = "Standard black and white theme") + theme_bw()

g + labs(title = "Adding theme elements with +", 
         subtitle = "Added to theme_bw()") + 
    theme_bw() + 
    theme(axis.title.y = element_text(color = "red"))

g + labs(title = "Adding theme elements with %+replace%", 
         subtitle = "Added to theme_bw()") + 
    theme_bw() %+replace%
    theme(axis.title.y = element_text(color = "red"))

Why did it rotate? As it turns out, when you add theme elements with +, it simply adds whatever changes you specify to the existing complete theme. Part of the underlying code in theme_bw (actually theme_gray, which theme_bw is based on) is axis.title.y = element_text(angle = 90) among other things. So since the black color is specified already elsewhere in the code, changing it to red just alters that one specification, leaving everything else intact.

However, when you use %+replace%, everything relating to axis.title.y gets replaced by whatever you specify; the crucial difference is that if you don’t specify something, rather than falling back on whatever is defined in the base theme, it’ll remove it entirely. So in the case of the angle, because we didn’t specify an angle after doing %+replace%, it was set to its defalt (0) instead of falling back on what the base theme uses (90). This can be problematic it not used carefully because failing to specify some things may cause an error in the code.

To be on the safe side, it’s probably best to use +, unless you’re pretty confident in what you’re doing.

6.2 Complete themes

One more argument to theme is complete, which can either be TRUE or FALSE. Normally it’s set to FALSE. This means that whatever elements you do not specify in theme will fall back on the values in the existing theme. (Keep in mind that even if you don’t specify a theme, it falls back on theme_gray.) If you set complete to true, it will not do this. So, you only want to set complete to TRUE if you have explicitly stated everything you need for a complete theme. That is, it’s not based on or a modification of an existing theme, but rather an entirely new theme built from scratch My guess is that most ggplot2 users will not need this function unless you are investing quite a bit of time creating something very different from existing themes.

6.3 Create the function

At long last, it’s time we create the theme function. I’ll call my new theme theme_joey. For simplicity, I’ll restict my theme to just modifying where the legend goes and I’ll center the title.

theme_joey <- function() {
    theme_bw() + 
    theme(legend.background = element_rect(fill = "white", linetype = "solid", color = "gray50", size = 0.5),
          legend.position = c(1,1),
          legend.justification = c(1,1),
          axis.title = element_text(color = "gray10", size = 12, face = "bold"),
          plot.title = element_text(size = 15),
          plot.caption = element_text(face = "italic"),
          plot.background = element_rect(fill = "gray90"))
}
g + theme_joey()

For a small bit of flexibliity, I may want to take advantage of the arguments to the base themes, as shown in section 3.3. A shortcut for making all those arguments accessible to theme_joey is to use .... Now I can change the font family and size without making any modifications to the function.

theme_joey <- function(...) {
    theme_bw(...) + 
    theme(legend.background = element_rect(fill = "white", linetype = "solid", color = "gray50", size = 0.5),
          legend.position = c(1,1),
          legend.justification = c(1,1),
          axis.title = element_text(color = "gray10", size = 12, face = "bold"),
          plot.title = element_text(size = 15),
          plot.caption = element_text(face = "italic"),
          plot.background = element_rect(fill = "gray90"))
}
g + theme_joey(base_family = "Palatino")

And that’s it! I can add this to all my plots and it’ll be formatted consistently every time. Furthermore, I can add to the theme just one time in the function, and it’ll show up on all the plots without any other extra work.

m + theme_joey()
g + theme_joey()
s + theme_joey()
c + theme_joey()

Now, as you may have seen already, moving the legend to the top right corner may not be the best idea to incorporate into a theme. There’s no guarantee that that space won’t be used by some data. In fact, in the bottom left plot, it covers the end of the longest bar in the bar plot. So, when creating your themes, you either need to make them robust enough to handle all sorts of different plots. Or, if you know you’ll be using your theme in a finite set of plots (like all 10 in preparation for a paper or something), you can more more aggressive in specifying theme elements, since you know your data.