The ma-R-velous tidyverse

I am currently taking a class in environmental economics, which of late seems to be doubling as a crash course in the R tidyverse. This blog post will discuss several R packages and demonstrate the application of several tidyverse sub-packages and commands that I have learned to use through this class. Two datasets will be used: the storm dataset, which contains data on the types of storms that occurred in the United States from 1975 to 2015, and the damages dataset, which is a record of the dollar cost of the damages caused by every tropical storm that occurred from 1975 to 2015. The datasets required for this example can be found in in this GitHub repository.

Tidy data and the tidyverse?

‘Tidy’ datasets provide a standardized way to link the physical layout of a dataset with the meaning associated with it [1]. A ‘tidy’ dataset has the following characteristics:

  1. Each variable forms a column
  2. Each observation forms a row
  3. Each cell contains only one value of only one data type

On the other hand, the tidyverse is a meta-package that bundles the packages facilitating data ‘tidying’, or data cleaning such that the final form of the dataset is ‘tidy’. You can check the full set of packages bundled into the tidyverse using tidyverse_packages(), which will show you the following output in the console:

Within this set, we will be using the dplyr, tidyr and lubridate packages. The dplyr and tidyr packages are “workhorse” packages – their main purpose is to clean and wrangle large datasets, and will be the two packages that we use most often in this example. The lubridate package is used to convert strings into dates, which we will use later in this example.

In this example, we will transform a ‘messy’ dataset into a ‘tidy’ one, and perform several other operations that are enabled by the R tidyverse.

The pacman package management tool

No, this is not the Toru Iwatani-created, dot-munching yellow puck released in 1980 by Namco. Instead, the pacman tool is a convenient library and package wrapper that combines the functionality of base library-related functions into intuitively named functions [2]. It was created to improve workflow by reducing the time required in calling and re-calling obscurely named functions. Ideally, it should be called at the beginning of the R script, like so:

# Run this line
if (!require("pacman")) install.packages("pacman")

This enables you to automatically install and load packages throughout the base R libraries and the R tidyverse using the p_load function. For this example, we will only require the tidyverse package. Load this package using the following script:

# Then run p_load to load the necessary packages
pacman::p_load(
 tidyverse
)

As previously mentioned, this automatically loads the dplyr, tidyr and lubridate packages that we will need for this example.

Before beginning, make sure that you are running the correct versions of dplyr and tidyr. Check the package versions using the packageVersion('package_name') function. All packages should be at least version 1.0.0.

Working with the datasets

Some key tidyr verbs

With the explanations out of the way, we begin by loading our two datasets:

# Load dataset
storms <- read.csv("storms.csv")
damages <- read.csv("damages.csv")

Note that the damages dataset is in an inconvenient ‘wide’ shape. To convert damages into ‘tidy’ format, we use pivot_longer(). pivot_longer() is one of the four key tidyr verbs, the remaining three being pivot_wider(), separate() and unite(). These verbs perform the following uses:

  1. pivot_longer(): Vertically stretches the data, increasing the number of rows and decreasing the number of columns
  2. pivot_wider(): Horizontally stretches the data, increasing the number of columns and decreasing the number of rows
  3. separate(): Turns a single-character column into multiple columns
  4. unite(): Pastes multiple columns into one

For this example, pivot_longer() is used to turn damages from its original wide shape into a long, 3-column dataframe where each column contains data on the storm’s name, type, and total dollar cost of damages. The code is as follows:

# tidy-up damages
damages <- damages %>% mutate(status = "tropical storm") %>%
  pivot_longer(-status, names_to="name", values_to="damages")

This script first categorizes all the storms within damages as tropical storms, and them assigns the names of each storm to the column ‘name’ and the cost of their damages to ‘damages’. This turns the original dataset:

into the following shape:

This results in a more readable dataset!

It’s a lubridate!

Next, we will visit out storm dataset. Note that the information regarding the date and time for each storm is inconveniently split between four columns:

We use the as_datetime() function within the lubridate package to combine these columns into one, easy-to-interpret column called ‘date’. Use the following script to:

# Paste four columns and convert to date-time format
storms <- storms %>% mutate(date=as_datetime(
  paste0(year, "-", month, "-", day, " ", hour, ":00:00")))

This pastes the data in the ‘year’, ‘month’, ‘day’ and ‘hour’ columns together and inserts formatted date and time into the newly-added ‘date’ columns:

Great! Our storm dataset is now more readable.

Some dplyr verbs and operations

There are five key dplyr verbs, some of which you have already seen:

  1. filter(): Obtains a subset of rows based on their column values
  2. arrange(): Reorders rows (in ascending or descending order) based on their values
  3. select(): Selects columns or variables of interest
  4. mutate(): Creates new columns or variables and automatically appends them to the dataframe
  5. summarize(): Collapses multiple rows into a single summary value, commonly by grouping a pre-selected variable

The dplyr package also come with a set of join operations, which enables the merging of two dataframes, x and y. The types of operations include:

  1. inner_join(): Matches pairs of observations whenever their values are equal
  2. left_join(): Keeps all x-observations that appear in either x or y
  3. right_join(): Keeps all y-observations that appear in either x or y
  4. full_join(): Keeps all x– and y-observations that appear in either x or y
  5. semi_join(): Keeps all observations in x that have a match in y
  6. anti_join(): Drops all observations in x that have a match in y

In this example, we will only be using the inner_join() function to merge a subset of storms and the whole of damages.

The verbs can be used to modify datasets using ‘pipes’, represented in R as %>%. We have previously seen the application of the mutate verb when we added the new columns ‘storm’ and ‘date’ to the damages and storms dataset respectively. Now, let’s apply the filter() verb to obtain only tropical storm data from storms:

# filter out only the tropical storms
ts <- storms %>% filter(
    stringr::str_detect(status, "tropical storm")
)

Here, we use stringr::str_detect() to detect the string ‘tropical storm’ within the ‘status’ column of storms and store it within the new ts dataframe.

Next, we use the select() function to select all columns but the columns containing the diameter of the hurricane/storm, since this is data that we will not use in this example:

# select a subset of columns
ts <- ts %>% select(!ts_diameter:hu_diameter)

Using ! indicates to the function that you would like to select the complement of the set following the exclamation point. Using : is useful when there are consecutive columns you would like to (de)select. Following this, you will have a pared-down dataset:

Now it’s time to apply our inner_join() operation to merge ts with damages in a new dataframe joined_sd that contains both geophysical information of each tropical storm, as well as its dollar cost of damages.

# joining tropical storms and damages
joined_sd <- inner_join(ts, damages)

where joined_sd has the following form:

Now, we perform some simple operations to obtain the mean dollar cost of damages per windspeed, and arrange each tropical storm in order of decreasing damage per windspeed:

# calculate average cost of damage per windspeed
joined_sd <- joined_storms_damages %>% 
  mutate(damage_per_mph = mean(damages)/wind) %>% 
  arrange(desc(damage_per_mph))

This results in the following final dataset:

From here, you can see that tropical storm Amy resulted in the most expensive damages. Finally, you can also use the summarize() verb to identify the mean of the cost of damages per windspeed of all the tropical storms:

# summarize by mean and sd of cost per windspeed
summary_sd <- joined_sd %>% summarize(
  damage_per_mph=mean(damage_per_mph))

You will find that, on average, each tropical storm costs USD$11,244,531. Plotting the cost of damages with respect to time using

# plot damages wrt time
ggplot(joined_sd, aes(year, damages)) + 
  geom_point() + ylim(min(joined_sd$damages), max(joined_sd$damages)) +
  geom_smooth(method = lm) + ggtitle("Damages ($) with respect to time")

you will find a slight increase in the cost of damages incurred by tropical storms as the years go by.

Conclusion

In this blog post, we walked through two example datasets to demonstrate three sub-packages within the tidyverse: tidyr, lubridate and dplyr. For the full version of this example, please visit the author’s GitHub repository.

References

pacman package – RDocumentation. (2021). Retrieved 14 November 2021, from https://www.rdocumentation.org/packages/pacman/versions/0.5.1

Rudik, I. (2021). AEM 6510 Chapter 10: R and the tidyverse. Presentation.

Tidy data. (2021). Retrieved 14 November 2021, from https://cran.r-project.org/web/packages/tidyr/vignettes/tidy-data.html

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s