1.3 — Data Visualization with ggplot2 — Class Notes
Tuesday, August 25, 2020
Today we start the fun stuff - data visualization. We will cover how to build plots with the package
ggplot2 (part of the
tidyverse): I will lecture for the first half (again, it will be a lot of information that you can refer back to as needed over the semester) and you will practice making plots.
Please see today’s readings for more information and helpful resources about all of your data visualization questions. The last slide today also contains a list of helpful online resources.
Live Class Session on Zoom
The live class Zoom meeting link can be found on Blackboard (see
LIVE ZOOM MEETINGS on the left navigation menu), starting at 11:30 AM.
If you are unable to join today’s live session, or if you want to review, you can find the recording stored on Blackboard via Panopto (see
Class Recordings on the left navigation menu).
Statistics and Software Survey
“Answers” for the statistics survey have been posted, including some summaries of how your classmates feel.
Additional Useful Information
New Packages Mentioned
Learn more about each of these in context in the slides, but below are links to each packages’s online help pages, which include descriptions of how each function works, examples, and cheatsheets to help you
ggplot2a grammar of graphics for ploting
Check back here later for more tips on plotting with ggplot2!
Extensions to ggplot2
ggplot2, being one of the most popular packages, has a lot of user-made extensions that allow you to do lots of neat things with your plots, from plotting networks, Parliaments, dendrograms, and other types of graphs, to formatting and visual tools that help improve your figures.
For the following demonstrations, we will use the
gapminder data. Let’s start just by making a basic graph and saving it as an object called
p. I have decided to map (
Working with Scales
I don’t like the default choices
ggplot2 has made for my point sizes for population, or the way it depicts them (in scientific notation) on the legend.
I will set my own scale by setting the
breaksThe cut offs for using one size circle vs. the next/previous size circle, depending on population.
manually, according to a vector I define as:
c(100000, 1000000, 100000000, 1000000000). So, I will use one point size for populations of 100 thousand, a bigger one for a million, a bigger one for 100 millions, and the biggest for 1 billion.
I am going to
label these (on my legend) as the following vector:
c("<1 million","1 million","100 million", "1 billion").
Lastly, I don’t think the size of the billion circle is big enough, a billion is a lot of people! So I will set the
range of sizes from size
1 point to size
To do this, I include all of this inside the
scale_size command (because I am scaling the
size of points):
This is also very hard to see the relationship (because it is nonlinear!). So I will rescale the
x_axis logarithmically with base 10:
Doing this already gives me a much clearer view of the relationship! But I don’t like the labels, or the breaks it has chosen, so I can customize them again:
scales package has a nice command to format labels, in this case I am calling the
scales::dollar function to print dollar signs in front of my axes numbers. I could have done it manually instead by setting
labels = c("$1,000", "$10,000", "$100,000").
We learn more about this in class 1.4 using
tidyverse, but let’s only look at one year of data (there’s too much going on in this plot, especially with the large points, some points are covering other points). So let’s only look at the year 2007. I can do this in two ways:
- Subset the data, save the subsetted data as an object (I’ll call
gap2007), plot with that object as my data:
- Subset data and pipe it directly into
Now let’s clean up the graph with the same things I did before, hide the legends (set the
size, my two
guides equal to
FALSE), add some labels, and change the theme:
p3<-p2007+scale_size(breaks = c(100000, 1000000, 100000000, 1000000000), labels=c("<1 million","1 million","100 million", "1 billion"), range=c(1,10))+ scale_x_log10( breaks = c(10^3, 10^4, 10^5), labels = scales::dollar)+ labs(x = "GDP per Capita (USD)", y = "Life Expectancy (years)")+ guides(color = FALSE, size = FALSE)+ theme_classic() p3
If I were to try to label some countries, with either
geom_text (just word) or
geom_label() (text in a box), setting the
country, see what would happen:
The labels, which are plotted right on top of each point, cover the points!
Someone figured out a clever way to let us do both, and it is a package called
ggrepel, which allows you to plot labels that are “repelled” away from the point they are labelling in an intelligent way. This is a separate package, which you must first install and then load with library to use it!
This is much better, but for this particular chart, since a lot of observations are close together, it would be unwise to label everything, perhaps only label a subset of important points.
One alternative is instead of points, to use some other marking. Someone created the
ggflags package to let you plot flags of countries. This creates a new type of geom, called
geom_flag, that requires you to map the
aesthetic to a variable in your data with the country name (incidentally, in
gapminder that variable is also called
country). Let’s try that out instead (and add my same customizations as above):
We can also make our plot a bit more interactive (on web only of course!) using the
ggplotly package, which allows
plotly.I am plotting our graph from above, not the flags, since
ggflags has not been configured for
Another major type of plot that we may use often is a barplot. Suppose we want to show the GDP per Capita of the top 20 countries in 2007. If I were to plot
country on the
x axis and
gdpPercap on the
y axis with
geom_col() allows you to specify an
x and a
geom_bar() only plots the
counts of each value on the
x axis. If we had done
+aes(x = country)+geom_bar(), it would plot the number of observations of each
country in the data, not what we want!
I get the following mess:
filterYou learn about this in class 1.4
arrange our data in
descending order by
## # A tibble: 142 x 6 ## country continent year lifeExp pop gdpPercap ## <fct> <fct> <int> <dbl> <int> <dbl> ## 1 Norway Europe 2007 80.2 4627926 49357. ## 2 Kuwait Asia 2007 77.6 2505559 47307. ## 3 Singapore Asia 2007 80.0 4553009 47143. ## 4 United States Americas 2007 78.2 301139947 42952. ## 5 Ireland Europe 2007 78.9 4109086 40676. ## 6 Hong Kong, China Asia 2007 82.2 6980412 39725. ## 7 Switzerland Europe 2007 81.7 7554661 37506. ## 8 Netherlands Europe 2007 79.8 16570613 36798. ## 9 Canada Americas 2007 80.7 33390141 36319. ## 10 Iceland Europe 2007 81.8 301931 36181. ## # … with 132 more rows
We only want the top 20 observations, so lets
slice to extract just rows
1:20. Then we’ll pipe it into our plot:
Now that’s closer to what we wanted! But here are a few more tips and tricks to make it better. First, let’s flip the axes to be able to read the countries better, using
One other useful thing to know would be how to display the bars in numerical order (from largest
gdpPercap to smalleset
gdpPercap) so we can get a clear ranking of countries. To do this, we are going to make use of another
tidyverse package called
forcats (dealing with
factors), specifically the function
fct_reorder(), which reorders a
factor variable (our
country) by the values of some other variable (our
gdpPercap). We need to do this to our
This is already looking good. Here’s another creative way to depict the same thing in a more visually-striking way. Instead of using
geom_bar(), let’s combine
geom_flag (to serve as end points) and
geom_segment()Note because we are defining an
aesthetic again for this, we need to make sure
x is also the reordered list of countries, so note I am doing the whole
(line segments) to make the following version:
bar3<-gap2007 %>% arrange(desc(gdpPercap)) %>% slice(1:20) %>% ggplot(data = .)+ aes(x = forcats::fct_reorder(country, gdpPercap), #<< y = gdpPercap, fill = continent)+ geom_segment(aes(x = forcats::fct_reorder(country, gdpPercap), #<< y = 0, #<< xend = country, #<< yend = gdpPercap, #<< color = continent), #<< size = 1)+ #<< geom_flag(aes(country = country))+ #<< coord_flip() bar3