Dataframe Manipulation with dplyr

Overview

Teaching: 40 min
Exercises: 15 min

Questions

• How can I manipulate dataframes without repeating myself?

Objectives

• To be able to use the six main dataframe manipulation ‘verbs’ with pipes in dplyr.

• To understand how group_by() and summarize() can be combined to summarize datasets.

• Be able to analyze a subset of data using logical filtering.

Manipulation of dataframes means many things to many researchers, we often select certain observations (rows) or variables (columns), we often group the data by a certain variable(s), or we even calculate summary statistics. We can do these operations using the normal base R operations:

mean(gapminder[gapminder$continent == "Africa", "gdpPercap"])

[1] 2193.755

mean(gapminder[gapminder$continent == "Americas", "gdpPercap"])

[1] 7136.11

mean(gapminder[gapminder$continent == "Asia", "gdpPercap"])

[1] 7902.15

But this isn’t very nice because there is a fair bit of repetition. Repeating yourself will cost you time, both now and later, and potentially introduce some nasty bugs.

The dplyr package

Luckily, the dplyr package provides a number of very useful functions for manipulating dataframes in a way that will reduce the above repetition, reduce the probability of making errors, and probably even save you some typing. As an added bonus, you might even find the dplyr grammar easier to read.

Here we’re going to cover 5 of the most commonly used functions as well as using pipes (%>%) to combine them.

1. select()
2. filter()
3. group_by()
4. summarize()
5. mutate()

If you have have not installed this package earlier, please do so:

install.packages('dplyr')

Now let’s load the package:

library("dplyr")

Using select()

If, for example, we wanted to move forward with only a few of the variables in our dataframe we could use the select() function. This will keep only the variables you select.

year_country_gdp <- select(gapminder,year,country,gdpPercap)

If we open up year_country_gdp we’ll see that it only contains the year, country and gdpPercap. Above we used ‘normal’ grammar, but the strengths of dplyr lie in combining several functions using pipes. Since the pipes grammar is unlike anything we’ve seen in R before, let’s repeat what we’ve done above using pipes.

year_country_gdp <- gapminder %>% select(year, country, gdpPercap)

To help you understand why we wrote that in that way, let’s walk through it step by step. First we summon the gapminder dataframe and pass it on, using the pipe symbol %>%, to the next step, which is the select() function. In this case we don’t specify which data object we use in the select() function since in gets that from the previous pipe. Fun Fact: There is a good chance you have encountered pipes before in the shell. In R, a pipe symbol is %>% while in the shell it is | but the concept is the same!

Using filter()

If we now wanted to move forward with the above, but only with European countries, we can combine select and filter

year_country_gdp_euro <- gapminder %>%
    filter(continent == "Europe") %>%
    select(year, country, gdpPercap)

Challenge 1

Write a single command (which can span multiple lines and includes pipes) that will produce a dataframe that has the African values for lifeExp, country and year, but not for other Continents. How many rows does your dataframe have and why?

Solution to Challenge 1

year_country_lifeExp_Africa <- gapminder %>%
                           filter(continent == "Africa") %>%
                           select(year, country, lifeExp)

As with last time, first we pass the gapminder dataframe to the filter() function, then we pass the filtered version of the gapminder dataframe to the select() function. Note: The order of operations is very important in this case. If we used ‘select’ first, filter would not be able to find the variable continent since we would have removed it in the previous step.

Using group_by() and summarize()

Now, we were supposed to be reducing the error prone repetitiveness of what can be done with base R, but up to now we haven’t done that since we would have to repeat the above for each continent. Instead of filter(), which will only pass observations that meet your criteria (in the above: continent=="Europe"), we can use group_by(), which will essentially use every unique criteria that you could have used in filter.

str(gapminder)

'data.frame':	1704 obs. of  6 variables:
 $ country  : Factor w/ 142 levels "Afghanistan",..: 1 1 1 1 1 1 1 1 1 1 ...
 $ year     : int  1952 1957 1962 1967 1972 1977 1982 1987 1992 1997 ...
 $ pop      : num  8425333 9240934 10267083 11537966 13079460 ...
 $ continent: Factor w/ 5 levels "Africa","Americas",..: 3 3 3 3 3 3 3 3 3 3 ...
 $ lifeExp  : num  28.8 30.3 32 34 36.1 ...
 $ gdpPercap: num  779 821 853 836 740 ...

str(gapminder %>% group_by(continent))

Warning: The `printer` argument is deprecated as of rlang 0.3.0.
This warning is displayed once per session.

Classes 'grouped_df', 'tbl_df', 'tbl' and 'data.frame':	1704 obs. of  6 variables:
 $ country  : Factor w/ 142 levels "Afghanistan",..: 1 1 1 1 1 1 1 1 1 1 ...
 $ year     : int  1952 1957 1962 1967 1972 1977 1982 1987 1992 1997 ...
 $ pop      : num  8425333 9240934 10267083 11537966 13079460 ...
 $ continent: Factor w/ 5 levels "Africa","Americas",..: 3 3 3 3 3 3 3 3 3 3 ...
 $ lifeExp  : num  28.8 30.3 32 34 36.1 ...
 $ gdpPercap: num  779 821 853 836 740 ...
 - attr(*, "vars")= chr "continent"
 - attr(*, "drop")= logi TRUE
 - attr(*, "indices")=List of 5
  ..$ : int  24 25 26 27 28 29 30 31 32 33 ...
  ..$ : int  48 49 50 51 52 53 54 55 56 57 ...
  ..$ : int  0 1 2 3 4 5 6 7 8 9 ...
  ..$ : int  12 13 14 15 16 17 18 19 20 21 ...
  ..$ : int  60 61 62 63 64 65 66 67 68 69 ...
 - attr(*, "group_sizes")= int  624 300 396 360 24
 - attr(*, "biggest_group_size")= int 624
 - attr(*, "labels")='data.frame':	5 obs. of  1 variable:
  ..$ continent: Factor w/ 5 levels "Africa","Americas",..: 1 2 3 4 5
  ..- attr(*, "vars")= chr "continent"
  ..- attr(*, "drop")= logi TRUE

You will notice that the structure of the dataframe where we used group_by() (grouped_df) is not the same as the original gapminder (data.frame). A grouped_df can be thought of as a list where each item in the listis a data.frame which contains only the rows that correspond to the a particular value continent (at least in the example above).

Using summarize()

The above was a bit on the uneventful side but group_by() is much more exciting in conjunction with summarize(). This will allow us to create new variable(s) by using functions that repeat for each of the continent-specific data frames. That is to say, using the group_by() function, we split our original dataframe into multiple pieces, then we can run functions (e.g. mean() or sd()) within summarize().

gdp_bycontinents <- gapminder %>%
    group_by(continent) %>%
    summarize(mean_gdpPercap = mean(gdpPercap))

continent mean_gdpPercap
     <fctr>          <dbl>
1    Africa       2193.755
2  Americas       7136.110
3      Asia       7902.150
4    Europe      14469.476
5   Oceania      18621.609

That allowed us to calculate the mean gdpPercap for each continent, but it gets even better.

Challenge 2

Calculate the average life expectancy per country. Which has the longest average life expectancy and which has the shortest average life expectancy?

Solution to Challenge 2

lifeExp_bycountry <- gapminder %>%
   group_by(country) %>%
   summarize(mean_lifeExp = mean(lifeExp))
lifeExp_bycountry %>%
   filter(mean_lifeExp == min(mean_lifeExp) | mean_lifeExp == max(mean_lifeExp))

# A tibble: 2 x 2
 country      mean_lifeExp
 <fct>               <dbl>
1 Iceland              76.5
2 Sierra Leone         36.8

Another way to do this is to use the dplyr function arrange(), which arranges the rows in a data frame according to the order of one or more variables from the data frame. It has similar syntax to other functions from the dplyr package. You can use desc() inside arrange() to sort in descending order.

lifeExp_bycountry %>%
   arrange(mean_lifeExp) %>%
   head(1)

# A tibble: 1 x 2
 country      mean_lifeExp
 <fct>               <dbl>
1 Sierra Leone         36.8

lifeExp_bycountry %>%
   arrange(desc(mean_lifeExp)) %>%
   head(1)

# A tibble: 1 x 2
 country mean_lifeExp
 <fct>          <dbl>
1 Iceland         76.5

The function group_by() allows us to group by multiple variables. Let’s group by year and continent.

gdp_bycontinents_byyear <- gapminder %>%
    group_by(continent, year) %>%
    summarize(mean_gdpPercap = mean(gdpPercap))

That is already quite powerful, but it gets even better! You’re not limited to defining 1 new variable in summarize().

gdp_pop_bycontinents_byyear <- gapminder %>%
    group_by(continent, year) %>%
    summarize(mean_gdpPercap = mean(gdpPercap),
              sd_gdpPercap = sd(gdpPercap),
              mean_pop = mean(pop),
              sd_pop = sd(pop))

count() and n()

A very common operation is to count the number of observations for each group. The dplyr package comes with two related functions that help with this.

For instance, if we wanted to check the number of countries included in the dataset for the year 2002, we can use the count() function. It takes the name of one or more columns that contain the groups we are interested in, and we can optionally sort the results in descending order by adding sort=TRUE:

gapminder %>%
    filter(year == 2002) %>%
    count(continent, sort = TRUE)

# A tibble: 5 x 2
  continent     n
  <fct>     <int>
1 Africa       52
2 Asia         33
3 Europe       30
4 Americas     25
5 Oceania       2

If we need to use the number of observations in calculations, the n() function is useful. It will return the total number of observations in the current group rather than counting the number of observations in each group within a specific column. For instance, if we wanted to get the standard error of the life expectency per continent:

gapminder %>%
    group_by(continent) %>%
    summarize(se_le = sd(lifeExp)/sqrt(n()))

# A tibble: 5 x 2
  continent se_le
  <fct>     <dbl>
1 Africa    0.366
2 Americas  0.540
3 Asia      0.596
4 Europe    0.286
5 Oceania   0.775

You can also chain together several summary operations; in this case calculating the minimum, maximum, mean and se of each continent’s per-country life-expectancy:

gapminder %>%
    group_by(continent) %>%
    summarize(
      mean_le = mean(lifeExp),
      min_le = min(lifeExp),
      max_le = max(lifeExp),
      se_le = sd(lifeExp)/sqrt(n()))

# A tibble: 5 x 5
  continent mean_le min_le max_le se_le
  <fct>       <dbl>  <dbl>  <dbl> <dbl>
1 Africa       48.9   23.6   76.4 0.366
2 Americas     64.7   37.6   80.7 0.540
3 Asia         60.1   28.8   82.6 0.596
4 Europe       71.9   43.6   81.8 0.286
5 Oceania      74.3   69.1   81.2 0.775

Using mutate()

We can also create new variables prior to (or even after) summarizing information using mutate().

gdp_pop_bycontinents_byyear <- gapminder %>%
    mutate(gdp_billion = gdpPercap*pop/10^9) %>%
    group_by(continent,year) %>%
    summarize(mean_gdpPercap = mean(gdpPercap),
              sd_gdpPercap = sd(gdpPercap),
              mean_pop = mean(pop),
              sd_pop = sd(pop),
              mean_gdp_billion = mean(gdp_billion),
              sd_gdp_billion = sd(gdp_billion))

Connect mutate with logical filtering: ifelse

When creating new variables, we can hook this with a logical condition. A simple combination of mutate() and ifelse() facilitates filtering right where it is needed: in the moment of creating something new. This easy-to-read statement is a fast and powerful way of discarding certain data (even though the overall dimension of the data frame will not change) or for updating values depending on this given condition.

## keeping all data but "filtering" after a certain condition
# calculate GDP only for people with a life expectation above 25
gdp_pop_bycontinents_byyear_above25 <- gapminder %>%
    mutate(gdp_billion = ifelse(lifeExp > 25, gdpPercap * pop / 10^9, NA)) %>%
    group_by(continent, year) %>%
    summarize(mean_gdpPercap = mean(gdpPercap),
              sd_gdpPercap = sd(gdpPercap),
              mean_pop = mean(pop),
              sd_pop = sd(pop),
              mean_gdp_billion = mean(gdp_billion),
              sd_gdp_billion = sd(gdp_billion))

## updating only if certain condition is fullfilled
# for life expectations above 40 years, the gpd to be expected in the future is scaled
gdp_future_bycontinents_byyear_high_lifeExp <- gapminder %>%
    mutate(gdp_futureExpectation = ifelse(lifeExp > 40, gdpPercap * 1.5, gdpPercap)) %>%
    group_by(continent, year) %>%
    summarize(mean_gdpPercap = mean(gdpPercap),
              mean_gdpPercap_expected = mean(gdp_futureExpectation))

Combining dplyr and ggplot2

First install and load ggplot2:

install.packages('ggplot2')

library("ggplot2")

In the plotting lesson we looked at how to make a multi-panel figure by adding a layer of facet panels using ggplot2. Here is the code we used (with some extra comments):

# Get the start letter of each country
starts.with <- substr(gapminder$country, start = 1, stop = 1)
# Filter countries that start with "A" or "Z"
az.countries <- gapminder[starts.with %in% c("A", "Z"), ]
# Make the plot
ggplot(data = az.countries, aes(x = year, y = lifeExp, color = continent)) +
  geom_line() + facet_wrap( ~ country)

This code makes the right plot but it also creates some variables (starts.with and az.countries) that we might not have any other uses for. Just as we used %>% to pipe data along a chain of dplyr functions we can use it to pass data to ggplot(). Because %>% replaces the first argument in a function we don’t need to specify the data = argument in the ggplot() function. By combining dplyr and ggplot2 functions we can make the same figure without creating any new variables or modifying the data.

gapminder %>%
   # Get the start letter of each country
   mutate(startsWith = substr(country, start = 1, stop = 1)) %>%
   # Filter countries that start with "A" or "Z"
   filter(startsWith %in% c("A", "Z")) %>%
   # Make the plot
   ggplot(aes(x = year, y = lifeExp, color = continent)) +
   geom_line() +
   facet_wrap( ~ country)

Using dplyr functions also helps us simplify things, for example we could combine the first two steps:

gapminder %>%
    # Filter countries that start with "A" or "Z"
	filter(substr(country, start = 1, stop = 1) %in% c("A", "Z")) %>%
	# Make the plot
	ggplot(aes(x = year, y = lifeExp, color = continent)) +
	geom_line() +
	facet_wrap( ~ country)

Advanced Challenge

Calculate the average life expectancy in 2002 of 2 randomly selected countries for each continent. Then arrange the continent names in reverse order. Hint: Use the dplyr functions arrange() and sample_n(), they have similar syntax to other dplyr functions.

Solution to Advanced Challenge

lifeExp_2countries_bycontinents <- gapminder %>%
   filter(year==2002) %>%
   group_by(continent) %>%
   sample_n(2) %>%
   summarize(mean_lifeExp=mean(lifeExp)) %>%
   arrange(desc(mean_lifeExp))

Other great resources

• R for Data Science
• Data Wrangling Cheat sheet
• Introduction to dplyr
• Data wrangling with R and RStudio

Key Points

• Use the dplyr package to manipulate dataframes.

• Use select() to choose variables from a dataframe.

• Use filter() to choose data based on values.

• Use group_by() and summarize() to work with subsets of data.

• Use mutate() to create new variables.