Chapter 3: Data Wrangling with tidyverse

Transform, reshape, and summarize data using dplyr and tidyr pipes.

3.1 The Pipe Operator

Pipes chain operations left-to-right, making code readable. R now has a native pipe (|>, R 4.1+) alongside the magrittr pipe (%>%).

library(tidyverse)

# Without pipes: nested and hard to read
round(mean(c(3.1, 4.7, 2.9)), 1)

# With pipe: reads left-to-right
c(3.1, 4.7, 2.9) |> mean() |> round(1)
# Both return 3.6

3.2 dplyr: Core Verbs

We will use the mpg dataset (bundled with ggplot2) to demonstrate each verb.

filter() — Select Rows

# Cars with highway mpg > 30 and 4 cylinders
mpg |>
  filter(hwy > 30, cyl == 4)

select() — Choose Columns

mpg |>
  select(manufacturer, model, year, hwy)

# Helper functions: starts_with, ends_with, contains
mpg |>
  select(starts_with("c"))  # cty, cyl, class

mutate() — Create/Modify Columns

mpg |>
  mutate(
    kpl_city = cty * 0.4251,     # mpg to km/L
    efficient = hwy > 30         # logical flag
  )

arrange() — Sort Rows

mpg |>
  arrange(desc(hwy)) |>        # highest hwy first
  head(5)

summarize() + group_by() — Aggregate

mpg |>
  group_by(class) |>
  summarize(
    n       = n(),
    avg_hwy = mean(hwy),
    sd_hwy  = sd(hwy)
  ) |>
  arrange(desc(avg_hwy))

3.3 Combining Verbs

The real power of dplyr is chaining multiple verbs together:

# Which manufacturer has the most fuel-efficient 6-cylinder cars?
mpg |>
  filter(cyl == 6) |>
  group_by(manufacturer) |>
  summarize(avg_hwy = mean(hwy), n = n()) |>
  filter(n >= 3) |>
  arrange(desc(avg_hwy))

3.4 across() for Column-wise Operations

When you need to apply the same transformation to multiple columns, across() eliminates repetitive code. It works inside mutate() and summarize().

# Compute the mean of all numeric columns by group
mpg |>
  group_by(drv) |>
  summarize(across(where(is.numeric), mean, .names = "avg_{.col}"))

# Round specific columns to 1 decimal place
mpg |>
  mutate(across(c(cty, hwy), \(x) round(x * 0.4251, 1),
         .names = "{.col}_kpl"))

# Apply multiple functions at once
mpg |>
  summarize(across(c(cty, hwy),
            list(mean = mean, sd = sd),
            .names = "{.col}_{.fn}"))

3.5 case_when() for Complex Conditionals

case_when() is the tidyverse equivalent of a multi-branch if/else. It is far more readable than nested ifelse() calls.

# Categorize highway mpg into efficiency tiers
mpg |>
  mutate(
    efficiency = case_when(
      hwy >= 35 ~ "excellent",
      hwy >= 25 ~ "good",
      hwy >= 18 ~ "average",
      .default   = "poor"
    )
  ) |>
  count(efficiency)

# Create academic letter grades from numeric scores
students <- tibble(name = c("Ana","Ben","Cal"), score = c(92,78,65))
students |>
  mutate(grade = case_when(
    score >= 90 ~ "A",
    score >= 80 ~ "B",
    score >= 70 ~ "C",
    score >= 60 ~ "D",
    .default   = "F"
  ))

3.6 Counting and Slicing

Several dplyr helpers make quick counting and row selection painless.

# n() counts rows in the current group
# n_distinct() counts unique values
mpg |>
  group_by(manufacturer) |>
  summarize(
    total_models = n(),
    unique_models = n_distinct(model)
  )

# slice_max() / slice_min() — top/bottom rows by a variable
mpg |>
  slice_max(hwy, n = 5)        # top 5 by highway mpg

mpg |>
  group_by(class) |>
  slice_min(cty, n = 1)        # worst city mpg per class

# count() is a shortcut for group_by + summarize(n = n())
mpg |> count(class, sort = TRUE)

3.7 relocate() and rowwise()

# relocate() — move columns to a new position
mpg |>
  relocate(hwy, cty, .before = manufacturer)

mpg |>
  relocate(where(is.character), .after = last_col())

# rowwise() — operate across columns within each row
scores <- tibble(
  student = c("Ana", "Ben", "Cal"),
  exam1 = c(88, 75, 92),
  exam2 = c(91, 80, 85),
  exam3 = c(78, 88, 90)
)
scores |>
  rowwise() |>
  mutate(avg = mean(c(exam1, exam2, exam3))) |>
  ungroup()

3.8 nest() / unnest() and Reading Multiple Files

Nested data frames store entire tibbles inside cells. Combined with purrr::map(), this pattern is powerful for batch operations.

library(tidyr)
library(purrr)

# Nest: create a list-column of data frames
nested <- mpg |>
  group_by(drv) |>
  nest()
nested    # 3 rows (f, 4, r), each with a 'data' list-column

# Apply a model to each nested group
nested |>
  mutate(model = map(data, \(df) lm(hwy ~ displ, data = df)))

# Unnest: flatten back to a regular tibble
nested |> unnest(data)

# Read multiple CSV files at once using purrr::map
# Suppose you have: data/sales_2020.csv, data/sales_2021.csv, etc.
files <- list.files("data/", pattern = "sales_.*\\.csv", full.names = TRUE)

all_sales <- files |>
  map(read_csv) |>
  list_rbind()    # stack all data frames into one

# Alternative: use purrr::map with an id column
all_sales <- files |>
  set_names() |>
  map(read_csv) |>
  list_rbind(names_to = "source_file")

3.9 Joining Data Frames

orders <- tibble(
  id = c(1, 2, 3),
  product = c("A", "B", "C"),
  amount = c(100, 200, 150)
)
customers <- tibble(
  id = c(1, 2, 4),
  name = c("Alice", "Bob", "Dana")
)

left_join(orders, customers, by = "id")   # keep all orders
inner_join(orders, customers, by = "id")  # only matching rows
full_join(orders, customers, by = "id")   # keep everything

3.10 tidyr: Reshaping Data

pivot_longer() — Wide to Long

wide <- tibble(
  country = c("US", "UK"),
  y2020 = c(331, 67),
  y2021 = c(333, 67)
)
wide |>
  pivot_longer(cols = starts_with("y"),
               names_to = "year",
               values_to = "population")

pivot_wider() — Long to Wide

long <- tibble(
  name = c("A","A","B","B"),
  metric = c("sales","profit","sales","profit"),
  value = c(100,20,150,35)
)
long |>
  pivot_wider(names_from = metric, values_from = value)

separate() and unite()

df <- tibble(date = c("2024-01-15", "2024-03-22"))

df |> separate(date, into = c("year","month","day"), sep = "-")

# unite() does the reverse
tibble(first = "John", last = "Doe") |>
  unite("full_name", first, last, sep = " ")

Exercises