dplyr slice_tail() in R: Take the Last N Rows (Per Group)

The slice_tail() function in dplyr returns the LAST n (or fraction) rows of a data frame, optionally per group. It is the dplyr-pipe-friendly replacement for base R tail().

slice_tail(df, n = 5)                       # last 5 rows
slice_tail(df, prop = 0.1)                  # last 10% of rows
df |> group_by(g) |> slice_tail(n = 3)      # 3 per group from end
slice_tail(df, n = 3, by = cyl)             # per group via .by (dplyr 1.1+)
df |> arrange(x) |> slice_tail(n = 5)       # largest by x (after sort)
tail(df, 5)                                 # base R alternative

What slice_tail() does in one sentence

slice_tail(.data, n = X) keeps the last X rows of a tibble; on a grouped tibble it keeps the last X per group. Use prop = 0.1 instead of n to take a fraction.

This is dplyr's pipe-friendly answer to tail(). The big advantage: it respects group_by() automatically and integrates cleanly with the rest of the dplyr verb set.

Syntax

slice_tail(.data, n = NULL, prop = NULL, by = NULL). Pass n OR prop, not both.

Five common patterns

1. Last n rows of an ungrouped data frame

Equivalent to tail(mtcars, 10), but pipeline-friendly.

2. Last n rows per group

Returns 3 rows per cyl group (9 total).

3. Use .by for one-step grouping

by = cyl scopes grouping to this verb only.

4. Last fraction with prop

Returns ~3 rows (10% of 32).

5. Latest record per group (sorted timeline)

A common time-series idiom: get the latest event per user.

slice_tail() vs tail() vs slice_max() vs slice_sample()

Four R functions for "rows from the end" or "extreme values".

When to use which:

• slice_tail when row order is meaningful (e.g., latest in chronological order after arrange()).
• slice_max when you want "top N by metric" without sorting first.
• tail for quick base-R inspection.
• slice_sample for randomness.

A practical workflow

The classic "latest per group" pattern uses arrange + group_by + slice_tail.

This returns the most recent record per group. Equivalent: slice_max(timestamp, n = 1, by = group_var) for newer dplyr.

Common pitfalls

Pitfall 1: per-group surprise. mtcars |> group_by(cyl) |> slice_tail(n = 5) returns 15 rows. Always ungroup downstream if you assumed a global tail.

Pitfall 2: implicit row order. slice_tail returns whatever rows happen to be last in current order. If the data is unsorted, the "tail" is arbitrary. Always arrange() first when order matters.

Try it yourself

Related slice functions

After mastering slice_tail, look at:

• slice_head(): first n rows (per group)
• slice_max() / slice_min(): top/bottom n by column value
• slice_sample(): random n rows
• slice(): specific positional rows
• last(): last value of a vector (not data frame)
• tail(): base R alternative

For "latest record per group", slice_max(timestamp, n = 1, by = group) is the cleanest pattern in modern dplyr.

FAQ

What is the difference between slice_tail and tail in R?

tail(df, n) is base R: returns last n rows of the WHOLE frame, ignores grouping. slice_tail(df, n) is dplyr: respects group_by() and works inside pipelines.

How do I get the last n rows per group in dplyr?

df |> group_by(g) |> slice_tail(n = 3) returns the last 3 rows per group. Or slice_tail(df, n = 3, by = g) in dplyr 1.1+.

What is the difference between slice_tail and slice_max?

slice_tail(n) returns the LAST n rows in current order. slice_max(col, n) returns the n highest BY a column value (sorts internally). Use slice_max when ranking is the criterion.

How do I get the latest record per group?

Sort by timestamp, group, take the tail of size 1: df |> arrange(ts) |> group_by(g) |> slice_tail(n = 1). Or directly: slice_max(df, ts, n = 1, by = g).

Why did I get more rows than I asked for?

Because the data frame was grouped. slice_tail(n = 5) on a grouped df returns 5 rows PER GROUP. Ungroup first if you want 5 total.