Mastering 'group_by' in R with dplyr: A Comprehensive Guide

Introduction

dplyr::group_by() changes the groups over which later verbs operate; it does not itself summarize rows. This lesson follows one small sales table through grouped summaries, grouped mutation and explicit ungrouping. The placement of .groups matters: it is a summarise argument, not a group_by option.

Tested with R 4.5.1 and dplyr 1.1.4. Run the following code blocks in order. All inputs are included.

Create groups and inspect their state

The table includes a missing amount and a missing region key. Keep these cases visible rather than treating them as zero or discarding them silently.

sales <- data.frame(region = c("East", "East", "West", "West", NA),
                    store = c("A", "A", "B", "C", "D"),
                    amount = c(10, 20, NA, 40, 5))
grouped <- dplyr::group_by(sales, region, store)
print(dplyr::group_vars(grouped))  # "region" "store"
stopifnot(nrow(grouped) == 5, dplyr::n_groups(grouped) == 4)

A missing grouping key forms its own group. Decide whether that is acceptable for the report. Merely grouping does not reorder the underlying records into a promised chronological order.

Put .groups on summarise

The summary counts records and observed amounts, and returns NA for a group with no observed amounts.

summary <- dplyr::summarise(grouped,
  records = dplyr::n(), observed = sum(!is.na(amount)),
  observed_total = if (all(is.na(amount))) NA_real_ else sum(amount, na.rm = TRUE),
  .groups = "drop")
summary <- dplyr::arrange(summary, region, store)
print(as.data.frame(summary))
stopifnot(nrow(summary) == 4, length(dplyr::group_vars(summary)) == 0)
stopifnot(identical(summary$observed_total, c(30, NA_real_, 40, 5)))

The values are East/A 30, West/B NA, West/C 40 and missing-region/D 5. A partial observed total does not establish a complete total. Putting .groups = "drop_last" inside group_by would create a grouping column called .groups, not control the summary result.

Distinguish retained groups from ungrouped results

After summarizing two keys, drop_last retains the first key. Explicitly remove it before an operation intended to span all regions.

by_store <- dplyr::summarise(grouped, rows = dplyr::n(), .groups = "drop_last")
stopifnot(identical(dplyr::group_vars(by_store), "region"))
all_stores <- dplyr::ungroup(by_store)
stopifnot(length(dplyr::group_vars(all_stores)) == 0)
global <- dplyr::summarise(all_stores, stores = dplyr::n(), records = sum(rows))
stopifnot(global$stores == 4, global$records == 5)

Use .groups = "keep" when the next step intentionally needs both group keys. Inspect group_vars() at a boundary where the intended scope changes; relying on a hidden grouped state can change later calculations.

Mutate keeps the original rows

A grouped mutation computes each region total and repeats it on that region’s rows. It has a different output shape from summarise.

with_totals <- sales |>
  dplyr::group_by(region) |>
  dplyr::mutate(region_observed_total = sum(amount, na.rm = TRUE)) |>
  dplyr::ungroup()
stopifnot(nrow(with_totals) == nrow(sales))
print(with_totals$region_observed_total)  # 30 30 40 40 5
stopifnot(identical(with_totals$region_observed_total, c(30, 30, 40, 40, 5)))
empty <- sales[FALSE, ] |>
  dplyr::group_by(region) |>
  dplyr::summarise(rows = dplyr::n(), .groups = "drop")
stopifnot(nrow(empty) == 0)

Each displayed region has at least one observed amount; use an explicit all-missing policy for broader input. For factors, group_by(..., .drop = FALSE) can preserve unused levels as empty groups. Keep grouping, missingness and output order explicit before considering performance changes.

Primary references

• group_by
• summarise