Introduction to forcats

In forcats: Tools for Working with Categorical Variables (Factors)

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

The goal of the forcats package is to provide a suite of useful tools that solve common problems with factors. Factors are useful when you have categorical data, variables that have a fixed and known set of values, and when you want to display character vectors in non-alphabetical order. If you want to learn more, the best place to start is the chapter on factors in R for Data Science.

Ordering by frequency

library(dplyr)
library(ggplot2)
library(forcats)

Let's try answering the question, "what are the most common hair colors of star wars characters?" Let's start off by making a bar plot:

#| fig.alt: >
#|   A bar chart of hair color of starwars characters. The bars are
#|   alphabetically ordered, making it hard to see general patterns.
ggplot(starwars, aes(y = hair_color)) + 
  geom_bar()

That's okay, but it would be more helpful the graph was ordered by count. This is a case of an unordered categorical variable where we want it ordered by its frequency. To do so, we can use the function fct_infreq():

#| fig.alt: >
#|   The bar chart of hair color, now ordered so that the least
#|   frequent colours come first and the most frequent colors come last.
#|   This makes it easy to see that the most common hair color is none 
#|   (~35), followed by brown (~18), then black (~12). Surprisingly,
#|   NAs are at the top of the graph, even though there are ~5 NAs and 
#|   other colors have smaller values.
ggplot(starwars, aes(y = fct_infreq(hair_color))) + 
  geom_bar()

Note that fct_infreq() it automatically puts NA at the top, even though that doesn't have the smallest number of entries.

It's a little surprising that the NA bar isn't ordered by frequency. To understand why we need to make a brief digression to discuss NAs in values vs. NAs in levels

NAss in levels and values

There are two ways to represent a missing value in a factor:

• You can include it in the values of the factor; it does not appear in the levels and is.na() reports it as missing. This is how missing values are encoded in a factor by default:

  r f <- factor(c("x", "y", NA)) levels(f) is.na(f)

• You can include it in the levels of the factor and is.na() does not report it as missing. This requires a little more work to create, because, by default, factor() uses exclude = NA, meaning that missing values are not included in the levels. You can force NA to be included by setting exclude = NULL:

  r f <- factor(c("x", "y", NA), exclude = NULL) levels(f) is.na(f)

NAs in the values tend to be best for data analysis, since is.na() works as you'd expect. NAs in the levels can be useful for display if you need to control where they appear in a table or a plot.

To fix the issue above, we can use fct_na_value_to_level() to convert the NA in the value to an NA in the levels. Then they appear where you'd expect:

#| fig.alt: >
#|   The bar chart of hair color, now ordered so that NAs are 
#|   ordered where you'd expect: in between white (4) and black (12).
ggplot(starwars, aes(y = fct_infreq(fct_na_value_to_level(hair_color)))) + 
  geom_bar() + 
  labs(y = "Hair color")

(If you need the opposite operation, you can use fct_na_level_to_value().)

Combining levels

Let's take a look at skin color now:

starwars %>%
  count(skin_color, sort = TRUE)

We see that there's 31 different skin colors - if we want to make a plot this would be way too many to display! Let's reduce it to only be the top 5. We can use fct_lump() to "lump" all the infrequent colors into one factor, "other." The argument n is the number of levels we want to keep.

starwars %>%
  mutate(skin_color = fct_lump(skin_color, n = 5)) %>%
  count(skin_color, sort = TRUE)

We could also have used prop instead, which keeps all the levels that appear at least prop of the time. For example, let's keep skin colors that at least 10% of the characters have:

starwars %>%
  mutate(skin_color = fct_lump(skin_color, prop = .1)) %>%
  count(skin_color, sort = TRUE)

Only light and fair remain; everything else is other.

If you wanted to call it something than "other", you can change it with the argument other_level:

starwars %>%
  mutate(skin_color = fct_lump(skin_color, prop = .1, other_level = "extra")) %>%
  count(skin_color, sort = TRUE)

What if we wanted to see if the average mass differed by eye color? We'll only look at the 6 most popular eye colors and remove NAs.

avg_mass_eye_color <- starwars %>%
  mutate(eye_color = fct_lump(eye_color, n = 6)) %>%
  group_by(eye_color) %>%
  summarise(mean_mass = mean(mass, na.rm = TRUE))

avg_mass_eye_color

Ordering by another variable

It looks like people (or at least one person) with orange eyes are definitely heavier! If we wanted to make a graph, it would be nice if it was ordered by mean_mass. We can do this with fct_reorder(), which reorders one variable by another.

#| fig-alt: >
#|   A column chart with eye color on the x-axis and mean mass on the 
#|   y-axis. The bars are ordered by mean_mass, so that the tallest bar
#|   (orange eye color with mean mass of ~275) is at the far right.

avg_mass_eye_color %>%
  mutate(eye_color = fct_reorder(eye_color, mean_mass)) %>%
  ggplot(aes(x = eye_color, y = mean_mass)) + 
  geom_col()

Manually reordering

Let's switch to using another dataset, gss_cat, the general social survey. What is the income distribution among the respondents?

gss_cat %>%
  count(rincome)

Notice that the income levels are in the correct order - they start with the non-answers and then go from highest to lowest. This is the same order you'd see if you plotted it as a bar chart. This is not a coincidence. When you're working with ordinal data, where there is an order, you can have an ordered factor. You can examine them with the base function levels(), which prints them in order:

levels(gss_cat$rincome)

But what if your factor came in the wrong order? Let's simulate that by reordering the levels of rincome randomly with fct_shuffle():

reshuffled_income <- gss_cat$rincome %>%
  fct_shuffle()

levels(reshuffled_income)

Now if we plotted it, it would show in this order, which is all over the place! How can we fix this and put it in the right order?

We can use the function fct_relevel() when we need to manually reorder our factor levels. In addition to the factor, you give it a character vector of level names, and specify where you want to move them. It defaults to moving them to the front, but you can move them after another level with the argument after. If you want to move it to the end, you set after equal to Inf.

For example, let's say we wanted to move Lt $1000 and $1000 to 2999 to the front. We would write:

fct_relevel(reshuffled_income, c("Lt $1000", "$1000 to 2999")) %>%
  levels()

What if we want to move them to the second and third place?

fct_relevel(reshuffled_income, c("Lt $1000", "$1000 to 2999"), after = 1) %>%
  levels()

forcats documentation built on Feb. 16, 2023, 8:57 p.m.