| title | output | ||||
|---|---|---|---|---|---|
Advancing with data visualization in R using `ggplot2` |
|
By Christine Zhang (ychristinezhang at gmail dot com and @christinezhang on Twitter)
Data + Narrative Workshop at Boston University (June 7, 2018)
GitHub repository for Data+Code: https://github.com/underthecurve/r-dataviz-ggplot2
Data visualization is a popular aspect of R, since R can create sophisticated graphics with (relatively few) lines of code. ggplot2 is one of the most popular graphics packages. ggplot2 is named for the Grammar of Graphics, a concept first popularized by Leland Wilkinson. The ggplot2 package was developed by Hadley Wickham, with the goal of providing a structured approach to graphing. As he wrote in 2012:
"The emphasis in ggplot2 is reducing the amount of thinking time by making it easier to go from the plot in your brain to the plot on the page."
You can read more about the Grammar of Graphics philosophy here: http://vita.had.co.nz/papers/layered-grammar.html
One of the advantages of ggplot2 is its versatility in creating different types of graphs. This is important because good data visualization involves making the best choices for showcasing a given dataset. Depending on the question you're trying to answer, these choices could be very different, even if you have the same data.
FlowingData blogger and statistician Nathan Yau made this point with a post earlier this year called "One Dataset, Visualized 25 Ways" (January 24, 2017). He took World Health Organization data on life expectancy by country from 2000 to 2015 and produced 25 different charts (some more useful than others). The blog post was also covered by Wired magazine in an article, "25 Visualizations Spin the Same Data Into 25 Different Tales" (February 9, 2017), which you can read here: https://www.wired.com/2017/02/25-visualizations-spin-data-25-different-tales/
In this workshop, we will:
- Examine the basic syntax of
ggplot2 - Use global life expectancy data to produce different type of plots charts using
ggplot2 - Explore advanced features of
ggplot2 - Learn to export graphics for publication
Loading tidyverse (which includes ggplot2)
First, let's load the tidyverse package, which is a collection of R packages designed for data science. Note tidyverse includes ggplot2.
# install.packages('tidyverse') # if you don't already have tidyverse
library('tidyverse') # load the tidyverse packageThe syntax of ggplot2
A ggplot2 graph is built up from a few basic elements:
- Data: the raw data you want to plot.
- Aesthetic mappings: e.g., which variable is on the x-axis? the y-axis? Should the color/size/position of the plotted data that be mapped to some variable?
- Geometries: the geometric shapes that represent the data
- Statistics: statistical transformations that are used to summarize the data
These elements can be customized to make a variety of ggplot2 graphics.
Let's take a look at a simple example. In the previous session you used qplot() to make a quick scatter plot of drivers killed by year in Great Britain. Let's do this again.
Make the data.seatbelts dataframe:
# Create a data frame
data(Seatbelts)
data.seatbelts <- data.frame(Year = floor(time(Seatbelts)),
Month = factor(cycle(Seatbelts),
labels = month.abb), Seatbelts)qplot() stands for "quick plot" and is the quick and simple plotting function of ggplot2.
# Now, let's plot the data, using the basic plotting function of ggplot2
qplot(data = data.seatbelts, x = Year, y = DriversKilled, main = 'Drivers Killed by Year')
Let's try making the same plot using the ggplot() function instead of qplot().
First, we create the ggplot object. Let's call it drivers.plot:
drivers.plot <- ggplot(data = data.seatbelts, # the data
aes(x = Year, # 'aes' stands for 'aesthetics': what's on the x- and y- axes
y = DriversKilled))Is this enough to generate the plot?
Let's take a look by calling the drivers.plot object:
drivers.plot
The graph is blank, although it does contain the correct axes for Year and DriversKilled. This is because we have mapped these using the aes() function to the x- and y- axes, respectively.
What's missing is the layer for Geometry, or what kind of geometric object we want to draw. qplot() automatically determines the geometry, but ggplot() requires you to specify it. geom_point() is the function for creating a scatter plot.
drivers.plot + geom_point() # adding geometry layer
The simplest form of a ggplot() graphic is one that has a ggplot object, with an added geometry layer. data = and aes() are the data and aesthetics mappings that tell R what data to plot and how.
Instead of geom_point() we could use geom_text()
drivers.plot +
geom_text(aes(label = Month))
We can add a title by adding a ggtitle() layer.
drivers.plot + geom_point() + ggtitle('Drivers Killed by Year')
A common practice to simplify the code is to save the ggplot() object as a separate R object, as we did above by creating drivers.plot. Then, we can add layers onto this object. We'll do this occasionally throughout this workshop.
Let's review qplot() and ggplot():
# These produce the exact same plot:
qplot(data = data.seatbelts, x = Year, y = DriversKilled, main = 'Drivers Killed by Year')
ggplot(data = data.seatbelts,
aes(x = Year, y = DriversKilled)) +
geom_point() +
ggtitle('Drivers Killed by Year') 
So the full form, ggplot(), is a lot more verbose than qplot(). But it is cleaner and allows for greater customization.
To see what I mean, let's move on to a real-life dataset that showcases some more interesting trends.
Life expectancy by country, 2000 to 2015
Let's load in a dataset of life expectancy by country (and the global average) between 2000 and 2015, downloaded from the World Health Organization.
life <- read_csv('life.csv')## Parsed with column specification:
## cols(
## year = col_integer(),
## region.code = col_character(),
## region = col_character(),
## country.code = col_character(),
## country = col_character(),
## `both sexes` = col_double(),
## female = col_double(),
## male = col_double()
## )
We can take a look at the data using head():
head(life)## # A tibble: 6 x 8
## year region.code region country.code country `both sexes` female male
## <int> <chr> <chr> <chr> <chr> <dbl> <dbl> <dbl>
## 1 2000 AFR Africa AGO Angola 45.3 46.8 43.8
## 2 2000 AFR Africa BDI Burundi 50.8 53.0 48.6
## 3 2000 AFR Africa BEN Benin 55.4 56.6 54.0
## 4 2000 AFR Africa BFA Burkina… 50.1 51.3 48.8
## 5 2000 AFR Africa BWA Botswana 47.8 47.7 47.9
## 6 2000 AFR Africa CAF Central… 46.0 47.1 44.9
Scatter plots
Exercise: fill in the blanks to create a scatter plot of female life expectancy (y-axis) versus male life expectancy (x-axis)
ggplot(_____, aes(x = ______, y = _____)) +
_____Let's take a look at just the United States. We can use the filter() function in the dplyr package (also included within tidyverse) to filter the life dataframe accordingly. We'll call it life.usa.
life.usa <- life %>% filter(country.code == 'USA')We can see the life expectancy of males versus females in the USA via a scatter plot:
ggplot(life.usa, aes(x = male, y = female)) +
geom_point() 
Line plots
Let's add a line:
ggplot(life.usa, aes(x = male, y = female)) +
geom_point() +
geom_line()
This is line that literally connects the points. But we might want to draw a fitted line instead. We can do this using the geom_smooth() layer rather than geom_line():
ggplot(life.usa, aes(x = male, y = female)) +
geom_point() +
geom_smooth()## `geom_smooth()` using method = 'loess' and formula 'y ~ x'
geom_smooth() by default uses a method called loess (local regression) to create the fitted curve. A more straightforward method for this data would be a straight-line, or a linear model. This is done using method = lm within the geom_smooth() layer. Also, you may have noticed a thin grey band around the line, the 95% confidence interval. We can remove this by setting se = F within geom_smooth():
ggplot(life.usa, aes(x = male, y = female)) +
geom_point() +
geom_smooth(method = 'lm', se = F)
Now let's go back to the life dataframe of all countries. What if we wanted to plot the total (both sexes) life expectancy for every country by year?
ggplot(life, aes(x = year, y = `both sexes`)) +
geom_point()
Let's fit a line for each country.
ggplot(life, aes(x = year, y = `both sexes`)) +
geom_point() +
geom_line() # what if I had run geom_smooth() instead?
Why didn't that work correctly?
ggplot(life, aes(x = year, y = `both sexes`, group = country)) +
geom_point() +
geom_line() 
ggplot(life, aes(x = year, y = `both sexes`, group = country)) +
# geom_point() +
geom_line() 
Facetted plots
It's hard to see which country is which—especially the country whose life expectancy dipped below 40 in 2010. When we have so many countries, facetting is often a better way to explore the trends in the data. This way of visualizaing information is also called "small multiples" in data viz lingo. (Warning: the following code will take a while to run, and you'll need to zoom in on the plot in R to see it in full.)
ggplot(life, aes(x = year, y = `both sexes`, group = country)) +
geom_line() +
facet_wrap(~ country)
What do we see in this plot (the zoomed in one on RStudio, not the one you actually see above) that we can't see in the one before it? Note you wouldn't publish the above graph, but it's useful as an exploratory exercise.
Reordering a plot
Haiti stands out. Take a look at the WHO metadata for life expectancy here http://apps.who.int/gho/data/node.wrapper.imr?x-id=65 as well as an alternative source, the World Bank: http://data.worldbank.org/indicator/SP.DYN.LE00.IN?end=2015&locations=HT&start=2010&year_low_desc=true. Also this paper on the HIV epidemic and the 2010 earthquake in Haiti: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2862993/
Let's see how Haiti in 2015 compares to the rest of the countries in the region. First, let's create a new dataframe, americas, for the countries within the Americas region.
americas <- life %>% filter(region == 'Americas' & year == 2015)Now, let's plot all the countries in the Americas. We could use a bar graph for this, but since we've done scatter plots so far, let's try one now.
ggplot(data = americas, aes(x = `both sexes`, y = country)) +
geom_point()
Why didn't I put the countries on the x-axis?
What would make this plot look better?
Right now, the countries are ordered from top to bottom in reverse alphabetical order. We want ggplot2 to order the countries such that the country with the highest life expectancy appears first (on the top), and the country with the lowest appears last (on the bottom).
In order to do this, we need to create a factor variable. Factors are categorical variables, and they can be assigned levels that determine the order they are plotted in.
To illustrate what I mean, let's create a variable, country.factor, which is the factor version of the existing country (character) variable:
class(americas$country) # a character## [1] "character"
americas$country.factor <- as.factor(americas$country)
class(americas$country.factor) # a factor## [1] "factor"
We can see the order of items in the factor variable, or the levels, using the levels() function. Think of it this way: the countries in country.factor are "secretly" numbers deep down, even though they look the same as the countries in country.
levels() tells us which number is assigned to each country in country.factor:
levels(americas$country.factor)## [1] "Antigua and Barbuda"
## [2] "Argentina"
## [3] "Bahamas"
## [4] "Barbados"
## [5] "Belize"
## [6] "Bolivia (Plurinational State of)"
## [7] "Brazil"
## [8] "Canada"
## [9] "Chile"
## [10] "Colombia"
## [11] "Costa Rica"
## [12] "Cuba"
## [13] "Dominican Republic"
## [14] "Ecuador"
## [15] "El Salvador"
## [16] "Grenada"
## [17] "Guatemala"
## [18] "Guyana"
## [19] "Haiti"
## [20] "Honduras"
## [21] "Jamaica"
## [22] "Mexico"
## [23] "Nicaragua"
## [24] "Panama"
## [25] "Paraguay"
## [26] "Peru"
## [27] "Saint Lucia"
## [28] "Saint Vincent and the Grenadines"
## [29] "Suriname"
## [30] "Trinidad and Tobago"
## [31] "United States of America"
## [32] "Uruguay"
## [33] "Venezuela (Bolivarian Republic of)"
We see that the default is to assign things alphabetically, so "Antigua and Barbuda" is 1 and "Venezuela (Bolivarian Republic of)" is 33.
Why are the countries plotted in reverse alphabetical order (from top to bottom) on the y-axis of the ggplot graph above?
We want to reorder the countries in country.factor such that they are ranked according to the both sexes variable. This is also called "releveling" the factor variable. The forcats package, another package included in the tidyverse set of packages, makes it easy to to this using the fct_reorder() function. Let's create another variable, country.factor.reorder, for the reordered factor:
library('forcats') #load forcats package
americas$country.factor.reorder <- fct_reorder(americas$country.factor, # factor variable to reorder
americas$`both sexes`) # variable to reorder it byWe can check to see that country.factor.reorder is indeed reordered by using the levels() function on it:
levels(americas$country.factor.reorder)## [1] "Haiti"
## [2] "Guyana"
## [3] "Belize"
## [4] "Bolivia (Plurinational State of)"
## [5] "Trinidad and Tobago"
## [6] "Suriname"
## [7] "Guatemala"
## [8] "Saint Vincent and the Grenadines"
## [9] "El Salvador"
## [10] "Grenada"
## [11] "Dominican Republic"
## [12] "Paraguay"
## [13] "Venezuela (Bolivarian Republic of)"
## [14] "Honduras"
## [15] "Colombia"
## [16] "Nicaragua"
## [17] "Brazil"
## [18] "Saint Lucia"
## [19] "Barbados"
## [20] "Peru"
## [21] "Bahamas"
## [22] "Ecuador"
## [23] "Jamaica"
## [24] "Argentina"
## [25] "Antigua and Barbuda"
## [26] "Mexico"
## [27] "Uruguay"
## [28] "Panama"
## [29] "Cuba"
## [30] "United States of America"
## [31] "Costa Rica"
## [32] "Chile"
## [33] "Canada"
How can we check that that worked? (hint: is the country assigned 1 the one with the lowest 2015 life expectancy in the Americas? Is the country assigned 6 the country with the sixth lowest 2015 life expectancy in the Americas?)
Now, instead of y = country in our ggplot() object, let's set y = country.factor.reorder and see what happens:
ggplot(data = americas, aes(x = `both sexes`, y = country.factor.reorder)) +
geom_point()
To save time (and once you're more familiar with factors and the forcats library), you can type y = fct_reorder(country.factor, both sexes) within the ggplot() object itself instead of creating a new variable.
Let's save the plot into an object called p:
p <- ggplot(data = americas,
aes(x = `both sexes`, y = fct_reorder(country.factor, `both sexes`))) +
geom_point()
p
Why didn't we need the dollar signs (e.g., fct_reorder(americas$country.factor, americas$both sexes) this time)?
Axes labels and scales
We can change the x-axis and y-axis titles using the labs() layer:
p + labs(x = 'Life expectancy at birth in 2015, years', y = '')
We can change the scale of the x-axis by adding a scale_x_continuous() layer. Let's say we want the x-axis to start at 50 and go to 100. We do this using limits = c(60, 100) within scale_x_continuous():
p + labs(x = 'Life expectancy at birth in 2015, years', y = '') +
scale_x_continuous(limits = c(50, 100))
That doesn't look too great. Let's set limits = c(60, 85) instead.
Maybe we want the x-axis labels to be every 5 years rather than every 10 years. We can add an argument to scale_x_continuous() called breaks() in order to do this. Note the use of the shortcut function seq(from = 50, to = 100, by = 5):
p + labs(x = 'Life expectancy at birth in 2015, years', y = '') +
scale_x_continuous(limits = c(60, 85), # minimum and maximum for the x-axis
breaks = seq(from = 60, to = 85, by = 5)) # x-axis labels
Maybe we want to remove the minor breaks, too. Let's go ahead and do that, as well as adding a title using the ggtitle() layer saving the plot as an R object called plot1:
# ?scale_x_continuous()
plot1 <- p + labs(x = 'Life expectancy at birth in 2015, years', y = '') +
scale_x_continuous(limits = c(60, 85), # minimum and maximum for the x-axis
breaks = seq(from = 50, to = 100, by = 5), # x-axis labels
minor_breaks = NULL) +
ggtitle("Haiti's life expectancy is the lowest in the region") # why did I use double quotes here?
plot1
Adjusting themes/appearance
We can further customize the look and feel of plot1 using theme as well as the ggthemes() library:
More on these themes here and here
# plot1 + theme_bw()
# install.packages('ggthemes') # if you don't already have ggthemes
library('ggthemes') # load the ggthemes pacakge
plot1 + theme_few()
The above themes are pre-made themes. You can also do some customization with the theme() layer. For example,
plot1 + theme(panel.background = element_blank())
The theme() layer can be a bit tricky to figure out. Take a look at the "Theme elements" section of this page to figure out the different parts.
Saving a plot
Finally, we can save the plot using ggsave():
ggsave('plot1.png', plot1 + theme_few())## Saving 7 x 5 in image
ggsave('plot1.png', plot1 + theme_few(), width = 8, height = 6)Other attributes of ggplot: sizes, scales, colors
Let's use the gapminder package to load in a different slice of the life expectancy data, GDP per capita, and population by country. Find out more about Gapminder here: http://www.gapminder.org/data/. And if you haven't seen it already, take some time to watch founder Hans Rosling's TED talk, "The best stats you've ever seen": https://www.ted.com/talks/hans_rosling_shows_the_best_stats_you_ve_ever_seen. I'd also encourage you to check out "Factfulness", a book written by the late Rosling and his family members/collaborators.
# install.packages('gapminder') # if you don't already have gapminder
library('gapminder') # load the gapminder package
gapminder## # A tibble: 1,704 x 6
## country continent year lifeExp pop gdpPercap
## <fct> <fct> <int> <dbl> <int> <dbl>
## 1 Afghanistan Asia 1952 28.8 8425333 779.
## 2 Afghanistan Asia 1957 30.3 9240934 821.
## 3 Afghanistan Asia 1962 32.0 10267083 853.
## 4 Afghanistan Asia 1967 34.0 11537966 836.
## 5 Afghanistan Asia 1972 36.1 13079460 740.
## 6 Afghanistan Asia 1977 38.4 14880372 786.
## 7 Afghanistan Asia 1982 39.9 12881816 978.
## 8 Afghanistan Asia 1987 40.8 13867957 852.
## 9 Afghanistan Asia 1992 41.7 16317921 649.
## 10 Afghanistan Asia 1997 41.8 22227415 635.
## # ... with 1,694 more rows
Exercise: create a scatter plot of GDP per capita in 2007 on the x-axis and life expectancy in 2007 on the y-axis and put it into an R object called p
p <- ggplot(data = gapminder %>% filter(year == 2007), aes(x = gdpPercap, y = lifeExp)) +
geom_point()
p
What if we wanted the size of the points to be equal to the population? We can use the size attribute in the aes() mapping function to do this.
p <- ggplot(data = gapminder %>% filter(year == 2007), aes(x = gdpPercap, y = lifeExp)) +
geom_point(aes(size = pop))An additional layer is required here, since we tend to perceive the size of points by area and not radius. Say we have cirlce 1 and circle 2. We want circle 2 to be twice as big as circle 1. In order to achieve this, we would want circle 2's area to be double the area of circle 2. If we set circle 2's radius to be twice as large, the area would be four times the area of circle 2 (since the area of a circle is proportional to the squared radius via pi * r^2).
We accomplish this by adding a scale_size_area() layer.
p
p + scale_size_area()
We could also color the plots according to continent by using the color attribute in the aes() function.
p <- ggplot(data = gapminder %>% filter(year == 2007), aes(x = gdpPercap, y = lifeExp)) +
geom_point(aes(size = pop, color = continent))
p + scale_size_area()
What if we just wanted the colors of the points to be, say, blue rather than mapped to the continent variable?
p <- ggplot(data = gapminder %>% filter(year == 2007), aes(x = gdpPercap, y = lifeExp)) +
geom_point(aes(size = pop), color = 'blue')
p + scale_size_area(max_size = 10) 
# max_size = 10 sets the maximum size of the points (in this case, it makes them larger)The difference between putting color = ___ inside or outside of the aes() function is that placing it inside makes the color an aesthetic mapping—that is, a color based on another variable in the data (in our case, the continent variable), and placing int outside makes the color a value in its own right, not tied to a variable in the data (in our case, 'blue'). Here are some more colors in R. Note I added max_size = 10 in order to make all the points a little bigger.
The legend
Earlier, we talked about the theme() layer. Customizing the legend requires making additions to the theme() layer.
Take a look here and see if you can figure out how to rename the legend "population" rather than "pop."
Sometimes it's nice to put the legend at the bottom rather than on the right, which is the default. This is another thing you can do within theme():
p + scale_size_area(max_size = 10) + theme(legend.position = 'bottom')
In this case, I think the legend should actually go inside the plot, in the bottom right hand corner, since I have some empty space there. Can you use this page to figure out how to do this?
Exercise: try to figure out how to make the population variable show up in non-scientific notation (hint: why do you think it shows up in scientific notation in the first place?)
Bar plots
Bar plots are another type of geometry that we can set using geom_bar() or geom_col().
Let's go back to our 2015 data on country and life expectancy in the Americas, the americas dataframe.
Let's try to make it into a bar graph
ggplot(data = americas, aes(x = country.factor, y = `both sexes`)) +
geom_bar()Why didn't this work?
Try this: (also ?geom_bar())
ggplot(data = americas, aes(x = country.factor)) +
geom_bar()
ggplot(data = americas, aes(x = country.factor, y = `both sexes`)) +
geom_bar(stat = 'identity')
In order for the heights of the bars to represent values in the data, we need to either specify stat = 'identity' within the geom_bar() layer, or, better yet, just use geom_col().
p <- ggplot(data = americas, aes(x = country.factor, y = `both sexes`)) +
geom_col()We can flip the coordinates by adding coord_flip() as a layer:
p + coord_flip()
Exercise: reorder the bars by life expectancy from highest to lowest using the fct_reorder() function in the forcats package
What else?
-
Change the look and feel of the bar graph above to your liking using the tools we've learned today and save it as a
.pngfile. -
What other visualizations could you make using the
gapminderdataset? -
What are some other customizations of
ggplot2plots would you like to see that we haven't covered today (additional resources are in the GitHub repo: https://github.com/underthecurve/r-dataviz-ggplot2)