dplyr summarise() in R: Aggregate Data with Stats

The summarise() function in dplyr collapses many rows into one summary row using aggregation functions like mean(), sum(), or n(). Combined with group_by() or the .by argument, it produces one summary per group.

summarise(df, avg = mean(mpg))                       # one summary
summarise(df, n = n(), avg = mean(mpg))              # multiple stats
summarise(df, avg = mean(mpg), .by = cyl)            # one row per group
group_by(df, cyl) |> summarise(avg = mean(mpg))      # same via group_by
summarise(df, across(where(is.numeric), mean))       # all numeric cols
summarise(df, p25 = quantile(mpg, 0.25), p75 = quantile(mpg, 0.75))  # custom
summarise(df, .by = cyl, n = n(), mn = min(mpg), mx = max(mpg))      # multi

What summarise() does in one sentence

summarise() collapses rows into a single summary row. You give it a data frame and one or more aggregation expressions of the form name = function(column). The result has one row total (or one row per group when grouped), with only the columns you named.

Note: dplyr accepts both summarise() (British) and summarize() (American). They are aliases. Use either.

Unlike base R aggregate(), summarise integrates into pipelines, computes multiple statistics in one call, names result columns clearly, and pairs naturally with group_by() or .by for per-group aggregation.

Syntax

summarise() takes a data frame plus aggregation expressions. Each expression must return a scalar (single value) per group. Use n() for row counts, n_distinct() for unique counts, and across() for applying the same function to many columns.

The full signature is:

summarise(.data, ..., .by = NULL, .groups = NULL)

.data is the data frame. The ... argument takes one or more name = aggregation_expr pairs. .by provides ad-hoc grouping. .groups controls what to do with the grouping after summarise (relevant only when chained from group_by()).

Seven common patterns

1. Single summary across all rows

The result is a one-row data frame with the column name you supplied.

2. Multiple statistics in one call

n() returns the row count. Each named expression becomes a column in the result.

3. Per-group summary with .by

.by = cyl groups for this single call only. The result is automatically ungrouped.

4. Per-group summary with group_by

Functionally identical to the .by version. Choose .by when grouping is local to this call; group_by() when subsequent verbs in the pipeline also need it.

5. Apply same function to many columns

across() plus a tidyselect helper applies one function to multiple columns. Replaces legacy summarise_at(), summarise_if(), summarise_all().

6. Custom quantile statistics

Any function that returns a scalar from a vector works inside summarise.

7. Distinct counts and presence checks

n_distinct() counts unique values. Predicates like any() and all() produce TRUE/FALSE summaries.

summarise() vs base R aggregation

Base R offers aggregate() and tapply() for per-group aggregation; summarise wraps these in pipeline-friendly syntax. The result of summarise is always a data frame; tapply() returns an array, aggregate() returns a data frame with awkward column names.

When to use which:

• Use summarise() for any multi-statistic or pipelined aggregation.
• Use base R mean(), sum(), tapply(), etc. for one-line scripts on a vector.

Common pitfalls

Pitfall 1: forgetting NA handling. mean(starwars$mass) returns NA because some rows are missing. mean(starwars$mass, na.rm = TRUE) ignores missing values. Inside summarise: summarise(starwars, avg = mean(mass, na.rm = TRUE)).

Pitfall 2: result is grouped after group_by() |> summarise(). dplyr 1.1+ peels off ONE level of grouping by default but the result may still be grouped if you grouped by multiple keys. Either chain ungroup() or use .groups = "drop" to fully ungroup. Or use .by instead.

Pitfall 3: trying to return multiple values per group. summarise(df, q = quantile(x, c(0.25, 0.75))) errors because each group expression must return a scalar. Either name each value (p25 = quantile(x, 0.25), p75 = quantile(x, 0.75)) or use reframe() (dplyr 1.1+) for vector returns.

Try it yourself

Related dplyr functions

After mastering summarise(), look at:

• group_by(), ungroup(): persistent grouping for chained operations
• count(): shortcut for summarise(n = n(), .by = ...)
• tally(), add_count(): variations on row counting
• reframe(): for summaries that return multiple rows per group
• n(), n_distinct(), cur_group(), cur_group_id(): helpers that work inside summarise
• across() with tidyselect: bulk column summaries

For very large data, also check data.table syntax which provides the same semantics with often faster execution.

FAQ

What is the difference between summarise and summarize in dplyr?

They are identical aliases. dplyr accepts both spellings to support British and American English users. Pick one and stick with it for consistency.

How do I summarise multiple columns in dplyr?

Use across() with a tidyselect helper: summarise(df, across(where(is.numeric), mean)) computes the mean of every numeric column. To apply multiple functions: summarise(df, across(where(is.numeric), list(mean = mean, sd = sd))).

How do I count rows in dplyr summarise?

Use n() inside summarise: summarise(df, n = n()). For unique-value counts use n_distinct(col). For a quick group-counted shortcut, count(df, group_col) is equivalent to df |> summarise(n = n(), .by = group_col).

What is the difference between summarise with .by vs group_by?

.by groups for the single summarise call and auto-ungroups the result. group_by() sets a persistent grouping that affects subsequent verbs (filter, mutate, summarise) until you call ungroup(). Use .by for one-off summaries; use group_by() for chains that need the grouping throughout.

Can I use custom functions inside summarise?

Yes, any function that returns a scalar per group works: summarise(df, p90 = quantile(x, 0.9)). For functions returning multiple values, use reframe() (dplyr 1.1+) instead of summarise().