group_by() and summarize()We’ve been working with the dplyr package for the past few classes and we’ve seen just how powerful it is when manipulating/transforming data. So far you’ve used the dplyr functions summarized in the table below
dplyr function |
Description |
|---|---|
filter() |
Subset by row values |
arrange() |
Sort rows by column values |
select() |
Subset columns |
mutate() |
Add new columns |
rename() |
Rename columns |
top_n() |
Select and order the top n entries according to a column (variable) |
summarize() |
Summarize columns |
Today we are going to introduce a few more dplyr functions that will help you with data manipulation and data analysis. In particular we are going to see how the functions group_by and summarize allow you do yield rapid insight into your data. Then we’ll work on a bunch of exercises to reinforce/test the concepts that you’ve learned.
Note: Before starting make sure to clear your Environment so that we can get rid of any data objects from last class. To do this you can go to your Environment tab and click the icon that looks like a broom.
dplyr package and the gapminder packageWe’ll load in tidyverse which contains dplyr (as well as many other packages). We will also load in the gapminder library so we can continue to work with the gapminder data we’ve been using in the past few lectures.
library(tidyverse)
library(readr) # we'll use this package later in the lecture to load in files from our class website
library(gapminder)We’ll create our own copy of the gapminder data that we’ll use in the upcoming sections.
my_gap <- gapmindersummarize() refresherIf you got to the summarize() section of last lecture, then work through this section below as a refresher, otherwise this will be new.
When analyzing a dataset, we are often interested in generating a table with statistics that summarize that data. As the name suggests the summarize() function helps us do just that.
Let’s compute average life expectancies and per capita GDP on our gap_data. Before doing this, let’s filter our data so we are just looking at year 2007.
my_gap_2007 <- filter(my_gap, year == 2007)Now, let’s use the summarize() function. The basic syntax is the summarize(dataset, variable_name_1 = statistic, variable_name_2 = statistic,...).
Note: both the American English spelling summarize() and British English spelling summarise() will work. If you want to make your code a bit classier, I suggest you use the British English.
summarize(my_gap_2007, avg_life = mean(lifeExp), avg_gdp_per_cap = mean(gdpPercap) )You can use a ton of other summary statistics functions (see your dplyr cheatsheet).
Create a few more summary tables using your my_gap data (note you may want to filter your data first as we did with year 2007). Try to test out some of the additional summary statistics functions from the dplyr cheatsheet.
# Your code here Did you learn anything interesting? If so, feel free to share what you found with the class or your neighbor.
group_by() and summarize()As you’ve seen, the summarize() function is really powerful. However, when we first group our data and then summarize we can often do so much more. Let’s see just how powerful summarize() is when we’ve first employed the group_by() function.
The group_by() function will create a “grouped” copy of a table and subsequent dplyr operations will manipulate each group separately and then the results will be combined.
Let’s try out an example to help make this clearer. You want to determine the minimum, mean, and maximum life expectancy observed on each continent in the year 2007. So first let’s group our my_gap_2007 data by continent
my_gap_2007 <- group_by(my_gap_2007, continent) # group the data by continentNow, let’s apply the summarize() function to our “grouped” dataset
summarize(my_gap_2007, min_life = min(lifeExp), mean_life = mean(lifeExp), max_life = max(lifeExp))## `summarise()` ungrouping output (override with `.groups` argument)Look at that! We’ve now got a summary table telling us the minimum, average, and maximum life expectancies observed on each continent in the year 2007! We did this with just a few lines of code! Really beats, creating a for loop to loop over each continent and compute the statistics.
You can even group by multiple variables. This is often incredibly useful. For instance, we might want to see how the life expectancy statistics by continent have changed over time. In this case we would group by continent and year before applying the summarize() function.
my_gap <- group_by(my_gap, continent, year) # group by continent then yearsummarize(my_gap, min_life = min(lifeExp), mean_life = mean(lifeExp), max_life = max(lifeExp))## `summarise()` regrouping output by 'continent' (override with `.groups` argument)Pretty cool right!
FYI, if you want to ungroup a dataset that you’ve grouped, you can use the ungroup() function. You can of course always regroup the data if you want.
Also remember that there are tons of statistics functions that you can use with summarize(). Take a look the the Summary Functions section of your dplyr cheatsheet for more info.
my_gap <- ungroup(my_gap) # ungroup the my_gap datagroup_by() with other dplyr functionsWhile group_by() is often used along with summarize(), you can use group_by() with other dplyr functions as well.
We can grab the top three per capita GDPs for each of the years that data was collected. To do this we’ll need to group the data by year and then apply the top_n() function. I’m going to use the pipe operator to do all of this in a single line of code.
my_gap %>%
group_by(year) %>%
top_n(3, gdpPercap)You can see that this worked, but the data was sorted in alphabetical order by country. It would be more useful to have the data sorted by year. Let’s modify the code above to sort the data too.
my_gap %>%
group_by(year) %>%
top_n(3, gdpPercap) %>%
arrange(year)Take a look at the summary table. Did you find anything interesting?
You’ve seen how group_by can be powerfully combined with dplyr functions, in particular the summarize() function. Now try out several interesting things of your own. Think of a few interesting question that you’d like to answer and answer them below.
# Your code heregapminder dataAt this point in the term we’ve established a pretty solid toolkit for programming in R and doing some data wrangling and analysis. Below are some exercises that will allow you to test out your skills, with a specific focus on the dplyr tools that you are now familiar with. Remember you can use the pipe %>% operator to easily string together many operations.
my_gap dataset
# Your code heremy_gap dataset
Save the table to a new object called gap_summary_table and sort this table in descending order by the change in life expectancy. Did you notice anything interesting? Think about what might explain the observed changes.
Next sort this same table in ascending order based on life expectancy. Do you observe anything interesting?
Once you’ve done the above, try sorting the table according to the ratio of per capita GDP
Hint: Look at your dplyr cheatsheet in the “Summary Functions” section for some functions that will be useful for the exercise above
# Your code hereLet’s take a look at some precipitation data from the National Oceanic and Atmospheric Agency (NOAA) to gain insight into annual and seasonal variability in precipitation for a US state of your choosing.
The dataset that we will use has monthly precipitation data from 1895 through 2017 for each US state. We are going to generate a summary of annual precipitation for each year on record for your state of interest. For example if you choose NY, you will create a dataframe that has the total annual precipitation for NY for each year from 1895 through 2017. If you didn’t have the tools from the dplyr package you would have to do this using a loop and indexing the dataframe with base R (cumbersome, requires more coding, harder to read, more prone to making a coding error). Thankfully we’ve learned how to use dplyr.
First we’ll load in dataset from last week
precip_data <- read_csv("https://stahlm.github.io/ENS_215/Data/NOAA_State_Precip_LabData.csv")##
## -- Column specification --------------------------------------------------------
## cols(
## Year = col_double(),
## Month = col_double(),
## Precip_inches = col_double(),
## state_cd = col_character()
## )Take a few minutes to look at the dataset to familiarize yourself with it (e.g. view it, use the summary() function,…).
Now let’s select a state of interest. Remember that if we weren’t using dplyr we would have to select the state of interest using code like what is shown directly below.
state_2_get <- "NY" # Abbreviation code for state I want to select
state_data <- precip_data[precip_data$state_cd == state_2_get, ] # get the rows with desired state, and get all columnsThat code works, but it is cumbersome to write, easy to make an error, and sort of difficult to decipher. We can do much better now.
Use dplyr to create a state_data object that has all of the data for your state of choosing.
# Your code hereOk, now let’s proceed to the exercises below which will allow you to demonstrate/test your dplyr skills and will highlight just how powerful the dplyr package is.
Some of these exercises will be challenging, so make sure to consult your dplyr cheatsheet, discuss with your classmates and me.
Some helpful advice: Remember to think step-by-step. Test each step as you go. Even for an experienced programmer/scientist it is often necessary to break the task into smaller chunks. Piping with %>% can make your task easier by allowing you to combine many individual steps together. If you finish all of the exercises you should go back and take more time to examine your results and think about the environmental siginificane/implications.
Now use the tools from dplyr to do the following:
# Your code here# Your code hereprecip_data, which has data for all of the states, create a single table that has
# Your code here# Your code hereprecip_data, which has data for all of the states, create a table that has
Hint: There are many ways to do this, however it might help to make an “intermediate” table of stats that you then operate on to get your final table (this isn’t necessary, but might help)
# Your code hereseason to your state_data that has a “flag” (categorical value) that takes the value “season-1” if the month is March through August and takes the value “season-2” otherwisestate_data create a table that reports for each year of record the ratio of precipitation in season-1 to the ratio of the total precipitation in that same year.This table will allow us to see how the seasonal distribution of precipitation may have changed (or not) over the years on record.
# Your code here