In this tutorial we look at some of the data on wealth and life expectancy of countries over time used by Hans Rosling in his famous TED talk. The goal is to introduce some simple 1D and 2D plots constructed using ggplot2
.
In these notes, I supply some initial plots for a topic, and then pose some questions or extensions as Your turn.
You will learn best if you try to reproduce them yourself as you go along. You can click Code || Hide for individual chunks. When code is shown, the icon at the upper left will copy it to the clipboard.
Open a new R script in R Studio, say, gapminder-tut.R
.
There are a lot of steps here, and you may not complete them in this lab session. I urge you to make as much progress as you can today, and treat the rest as homework. Save your script somewhere you can access it later.
library(ggplot2)
library(dplyr) # data munging
library(scales) # nicer axis scale labels
The data has conveniently been packaged in the R package gapminder, a subset of the original data set from (http://gapminder.org). For each of 142 countries, the package provides values for life expectancy, GDP per capita, and population, every five years, from 1952 to 2007.
It is unlikely that you have this package installed, so the first step is to install it and then load it.
install.packages("gapminder")
library(gapminder)
Tech note: Here is a way to make a script more portable by testing whether a package is available before installing it.
if(!require(gapminder)) {install.packages("gapminder"); library(gapminder)}
What variables are available, and what are their names? str()
is your friend here.
str(gapminder)
## tibble [1,704 x 6] (S3: tbl_df/tbl/data.frame)
## $ country : Factor w/ 142 levels "Afghanistan",..: 1 1 1 1 1 1 1 1 1 1 ...
## $ continent: Factor w/ 5 levels "Africa","Americas",..: 3 3 3 3 3 3 3 3 3 3 ...
## $ year : int [1:1704] 1952 1957 1962 1967 1972 1977 1982 1987 1992 1997 ...
## $ lifeExp : num [1:1704] 28.8 30.3 32 34 36.1 ...
## $ pop : int [1:1704] 8425333 9240934 10267083 11537966 13079460 14880372 12881816 13867957 16317921 22227415 ...
## $ gdpPercap: num [1:1704] 779 821 853 836 740 ...
Tech note: The gapminder
data set was constructed as a tibble
, a generalization of a data.frame
. The print()
method gives an abbreviated printout. Normally, you would use head(gapminder)
for this.
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
Normally, when starting with a new data set, it is useful to get some overview of the variables. The simplest one is summary()
.
summary(gapminder)
## country continent year lifeExp pop gdpPercap
## Afghanistan: 12 Africa :624 Min. :1952 Min. :23.60 Min. :6.001e+04 Min. : 241.2
## Albania : 12 Americas:300 1st Qu.:1966 1st Qu.:48.20 1st Qu.:2.794e+06 1st Qu.: 1202.1
## Algeria : 12 Asia :396 Median :1980 Median :60.71 Median :7.024e+06 Median : 3531.8
## Angola : 12 Europe :360 Mean :1980 Mean :59.47 Mean :2.960e+07 Mean : 7215.3
## Argentina : 12 Oceania : 24 3rd Qu.:1993 3rd Qu.:70.85 3rd Qu.:1.959e+07 3rd Qu.: 9325.5
## Australia : 12 Max. :2007 Max. :82.60 Max. :1.319e+09 Max. :113523.1
## (Other) :1632
We will want to look at trends over time by continent
. How many countries are in this data set in each continent? There are 12 years for each country. Are the data complete? table()
gives an answer.
table(gapminder$continent, gapminder$year)
##
## 1952 1957 1962 1967 1972 1977 1982 1987 1992 1997 2002 2007
## Africa 52 52 52 52 52 52 52 52 52 52 52 52
## Americas 25 25 25 25 25 25 25 25 25 25 25 25
## Asia 33 33 33 33 33 33 33 33 33 33 33 33
## Europe 30 30 30 30 30 30 30 30 30 30 30 30
## Oceania 2 2 2 2 2 2 2 2 2 2 2 2
Tech note: table()
doesn’t have a data=
argument, so you have to qualify the names of variables using data$variable
notation. Another way to do this is to use the with()
function, that makes variables in a data set available directly. The same table can be obtained using:
with(gapminder, {table(continent, year)})
Bar plots are often used to visualize the distribution of a discrete variable, like continent
. With ggplot2
, this is relatively easy:
x
variable to continent
geom_bar()
layer, that counts the observations in each category and plots them as bar lengths.ggplot(gapminder, aes(x=continent)) + geom_bar()
To make this (and other plots) more colorful, you can also map the fill
attribute to continent
. Note that adding the fill
attribute automatically adds a legend.
ggplot(gapminder, aes(x=continent, fill=continent)) + geom_bar()
As you progress with ggplot2
, you will also learn to:
The next step shows several of these features:
geom_bar
we transform the computed ..count..
to ..count../12
so it represents the number of countries. 12 = length(unique(gapminder$year))
is the number of years of data for each countryy
axiscontinent
, which is redundant here, because the continents are labeled on the x axis.ggplot(gapminder, aes(x=continent, fill=continent)) +
geom_bar(aes(y=..count../12)) +
labs(y="Number of countries") +
guides(fill=FALSE)
Another ggplot2
feature is that every plot is a ggplot
object. If you like a given plot, you can save it to some name by using mybar <- ggplot() + ...
, or after you’ve done it, using mybar <- last_plot()
# record a plot for future use
mybar <- last_plot()
Then, we can add other layers or transform the coordinates with coord_
functions.
coord_trans(y="sqrt")
plots the frequencies on a square root scalecoord_flip()
interchanges the X, Y axes, to make a horizontal bar chart.coord_polar()
maps (X,Y) to polar coordinates (radius, angle); a coxcomb chart rather than a pie chart.mybar + coord_trans(y="sqrt")
mybar + coord_flip()
mybar + coord_polar()
There are several continuous variables in this data set: life expectancy (lifeExp
), population (pop
) and gross domestic product per capita (gdpPercap
) for each year
and country
. For such variables, density plots provide a useful graphical summary.
We will start with lifeExp
. The simplest plot uses this as the horizontal axis, aes(x=lifeExp)
and then adds geom_density()
to calculate and plot the smoothed frequency distribution.
ggplot(data=gapminder, aes(x=lifeExp)) +
geom_density()
Tech note: The smoothness of the density curve is controlled with a bandwidth, bw =
or adjust =
argument. Larger bw
values give smoother curves. The default is calculated from the data. The adjust
argument multiplies the default by that factor. Try running the following lines:
ggplot(data=gapminder, aes(x=lifeExp)) + geom_density(adjust = 0.25)
ggplot(data=gapminder, aes(x=lifeExp)) + geom_density(adjust = 4)
To make these plots prettier, you can:
size=
),fill=""
),alpha=
): you can therefore see the grid lines underneath the following plots because the fill color is partially transparent.ggplot(data=gapminder, aes(x=lifeExp)) +
geom_density(size=1.5, fill="pink", alpha=0.3)
If you want, you can also add a histogram. This is a little more complicated to get right, because histograms are computed differently and need some additional arguments. In particular, the aesthetic y = ..density..
refers to the density that is calculated internally by geom_histogram
.
ggplot(data=gapminder, aes(x=lifeExp)) +
geom_density(size=1.5, fill="pink", alpha=0.5) +
geom_histogram(aes(y=..density..),
binwidth=4, color="black", fill="lightblue", alpha=0.5)
I don’t recommend such plots in general, because the density plot usually gives most of the information, and the histogram doesn’t add much.
The plot of lifeExp
is bimodal, and looks wierd. Maybe this is hiding a difference among countries in different continents. The easy way to see this in ggplot2
is to add another aesthetic attribute, fill=continent
, which is inherited in geom_density()
. Using transparent colors (alpha=
) makes it easier to see the different distributions across continent
.
ggplot(data=gapminder, aes(x=lifeExp, fill=continent)) +
geom_density(alpha=0.3)
It is easy to see that African countries differ markedly from the rest.
Alternatively, you might want to view the distributions of life expectancy by another visual summary, grouped by continent
. All you need to do is change the aesthetic to show continent
on one axis, and life expectancy (lifeExp
) on the other.
gap1 <- ggplot(data=gapminder, aes(x=continent, y=lifeExp, fill=continent))
Then, add a geom_boxplot()
layer:
gap1 +
geom_boxplot(outlier.size=2)
geom_violin()
The continents are a factor, and are ordered alphabetically by default. It might be more useful to order them by the mean or median life expectancy.
In this example, I use the dplyr
“pipe” notation (%>%
) to send the gapminder
data to the dplyr::mutate()
function, and within that, reorder()
the continents by their median life expectancy.
gapminder %>%
mutate(continent = reorder(continent, lifeExp, FUN=median))
(In other situations, you could use FUN=mean
, FUN=sd
, or FUN=max
to sort the levels by their means, standard deviatons, maximums, or any other function.)
Continuing from here, you can pipe the result of this right into ggplot
:
gapminder %>%
mutate(continent = reorder(continent, lifeExp, FUN=median)) %>%
ggplot(aes(x=continent, y=lifeExp, fill=continent)) +
geom_boxplot(outlier.size=2)
Let’s look at the distribution of gdpPercap
in a similar way, starting with the unconditional distribution.
ggplot(data=gapminder, aes(x=gdpPercap)) +
geom_density()
Hmm, that is extremely positively skewed.
lifeExp
plot the distributions separately for each continentx
axis to log10(gdpPercap)
.gdpPercap
by continent
How has life expectancy changed over time? The simplest way to to plot a line for each country over year
. To do this, we use the group
aesthetic.
ggplot(gapminder, aes(x=year, y=lifeExp, group=country)) +
geom_line()
What a mess! But there is also something strange here: Several countries show a very jagged pattern.
One way to examine this is to find the countries with the largest measures of variability over time. For this, we can group the data by country and summarise()
with sd()
and IQR()
. Note the use of top_n()
to find the largest values and arrange()
for sorting.
gapminder %>%
group_by(country) %>%
summarise(sd=sd(lifeExp), IQR=IQR(lifeExp)) %>%
top_n(8) %>%
arrange(desc(sd))
## # A tibble: 8 x 3
## country sd IQR
## <fct> <dbl> <dbl>
## 1 Oman 14.1 25.5
## 2 Vietnam 12.2 21.2
## 3 Saudi Arabia 11.9 20.3
## 4 Indonesia 11.5 18.4
## 5 Libya 11.4 19.8
## 6 Yemen, Rep. 11.0 19.7
## 7 West Bank and Gaza 11.0 19.3
## 8 Tunisia 10.7 19.1
A more complete analysis would investigate the data for those countries further, or omit them for some purposes.
A better look at trends over time is to find the mean or median for each year
and continent
and plot those.
gapminder %>%
group_by(continent, year) %>%
summarise(lifeExp=median(lifeExp)) %>% head()
## # A tibble: 6 x 3
## # Groups: continent [1]
## continent year lifeExp
## <fct> <int> <dbl>
## 1 Africa 1952 38.8
## 2 Africa 1957 40.6
## 3 Africa 1962 42.6
## 4 Africa 1967 44.7
## 5 Africa 1972 47.0
## 6 Africa 1977 49.3
One nice feature of the dplyr
and tidyverse
framework, is that you can pipe the result of such a summary directly to ggplot()
:
gapminder %>%
group_by(continent, year) %>%
summarise(lifeExp=median(lifeExp)) %>%
ggplot(aes(x=year, y=lifeExp, color=continent)) +
geom_line(size=1) +
geom_point(size=1.5)
Alternatively, if you want to make several plots of such a summarized data set, save the result in a new object. Here I’m using a handy trick: assign a result at the end of a computation using -> gapyear
# save in a new data set
gapminder %>%
group_by(continent, year) %>%
summarise(lifeExp=median(lifeExp)) -> gapyear
Then, rather than joining all the points, we could fit linear regression lines for each continent
.
ggplot(gapyear, aes(x=year, y=lifeExp, color=continent)) +
geom_point(size=1.5) +
geom_smooth(aes(fill=continent), method="lm")
Use a loess
smooth rather than a linear regression line. Which do you prefer as a description of changes in life expectancy over year?
Another possibility is a quadratic regression, using y ~ poly(x, 2)
as the formula
argument in geom_smooth()
.
I don’t like the default use of legends in such plots.
Now let’s explore the relationship between life expectancy and GDP with a scatterplot, which was the subject of Rosling’s TED talk. (Actually, he did more than this, with a “moving bubble plot”, using a bubble symbol ~ population, and animating this over time.)
A basic scatterplot is set up by assigning two variables to the x
and y
aesthetic attributes. The following just creates an empty plot frame.
plt <- ggplot(data=gapminder,
aes(x=gdpPercap, y=lifeExp))
plt
Add the points in another layer. Because I saved the basic plot, I can just use + geom_point()
.
plt + geom_point()
Or, color them by continent. Note that we can map the color
attribute in this layer.
plt + geom_point(aes(color=continent))
Add a smoothed curve for all the data:
plt + geom_point(aes(color=continent)) +
geom_smooth(method="loess")
From what we saw earlier about GDP, this variable is so skewed that it is better plotted on a log scale:
plt + geom_point(aes(color=continent)) +
geom_smooth(method="loess") +
scale_x_log10()
scale_x_log10(labels=scales::comma)
+ theme(legend.position = c(0.8, 0.2))
.+ theme_bw()
loess
smoothed curve with a separate regression line for each continent.size
of each point to population (pop
)Wouldn’t it be nice to animate the relationship between life expectancy and GDP over time? The plotly package, by Carson Sievert, adds a frame
aesthetic to ggplot, and allows interactive, linked views of a series of frames over time.
The plot is also interactive, in that you can hover the mouse over a point and see a pop-up window giving details about the country.
library(plotly)
g <- crosstalk::SharedData$new(gapminder, ~continent)
gg <- ggplot(g, aes(gdpPercap, lifeExp, color = continent, frame = year)) +
geom_point(aes(size = pop, ids = country)) +
geom_smooth(se = FALSE, method = "lm") +
scale_x_log10()
ggplotly(gg) %>%
highlight("plotly_hover")
There is an online book describing plotly. The image above is described in Section 5.2.