dplyr slice_sample() in R: Random Rows From a Tibble

The slice_sample() function in dplyr returns a random sample of n rows (or a fraction) from a data frame, optionally per group, with or without replacement. It supersedes the older sample_n() and sample_frac().

slice_sample(df, n = 5)                       # 5 random rows
slice_sample(df, prop = 0.1)                  # 10% of rows
slice_sample(df, n = 3, by = cyl)             # 3 per group (stratified)
slice_sample(df, n = 5, replace = TRUE)       # bootstrap
slice_sample(df, n = 5, weight_by = w)        # weighted sample
df |> group_by(g) |> slice_sample(n = 3)      # equivalent grouped form
set.seed(42); slice_sample(df, n = 5)         # reproducible

What slice_sample() does in one sentence

slice_sample(.data, n) returns a random sample of n rows; slice_sample(.data, prop = 0.1) returns 10% of the rows. On a grouped tibble (or with by = g), the sampling happens within each group.

This is the modern, group-aware random sampler. For new code, prefer it over sample_n() (deprecated).

Syntax

slice_sample(.data, n = NULL, prop = NULL, weight_by = NULL, replace = FALSE, by = NULL). Pass n OR prop, not both.

Five common patterns

1. Random n rows

2. Random fraction

20% of 32 = ~6 rows.

3. Stratified sample (per group)

3 random cars per cylinder group, regardless of group size.

4. Bootstrap (with replacement)

Standard bootstrap: same row count as original, with duplicates allowed.

5. Weighted sample

Rows with higher mpg are more likely to be picked. Useful for importance sampling.

slice_sample() vs sample_n() vs sample()

Three sampling functions in R, with different scope.

When to use which:

• slice_sample for data frames in dplyr pipelines.
• sample for sampling from vectors or generating random indices.
• Avoid sample_n in new code.

A practical workflow

The "stratified sample" pattern is the most common slice_sample use case. Examples:

• Train/test split with balanced classes: slice_sample(prop = 0.8, by = class)
• Sample customers per region equally: slice_sample(n = 100, by = region)
• Bootstrap CI estimation: loop slice_sample(n = nrow(df), replace = TRUE) 1,000 times

For one-off samples, slice_sample(n = 5) is the quick interactive tool. For production analysis, set the seed and document the sample size.

Common pitfalls

Pitfall 1: forgetting to set seed. slice_sample is non-deterministic. Reports / tests / docs should set.seed(42) (or any fixed integer) right before to get reproducible output.

Pitfall 2: per-group surprise. On a grouped tibble, slice_sample(n = 5) returns 5 rows PER GROUP. Often what you want for stratification, sometimes not. Use ungroup() first if you mean a global sample.

Reproducibility and seeds

Random samples are only useful if they are reproducible across runs. Always call set.seed(N) immediately before slice_sample() when the result will be used in a report, plot, or test. Different code paths that need DIFFERENT samples should use different seeds (e.g., set.seed(1) for train, set.seed(2) for test) so the splits are reproducible AND independent. R's RNG state is global, so any operation between set.seed() and slice_sample() that consumes randomness will desync the result. Keep them adjacent.

Try it yourself

Related slice functions

After mastering slice_sample, look at:

• slice_head() / slice_tail(): positional first/last n
• slice_max() / slice_min(): top/bottom n by value
• slice(): specific row indexes
• sample_n() / sample_frac(): deprecated; do not use
• rsample package: train/test splits with class balance, k-fold CV
• base::sample(): vector sampling

For machine-learning splits, the rsample package builds on slice_sample with initial_split() and vfold_cv().

FAQ

What is the difference between slice_sample and sample_n?

sample_n() is deprecated since dplyr 1.0. slice_sample() is the replacement: clearer name, supports prop, weight_by, and by arguments, group-aware.

How do I do a reproducible random sample in R?

Call set.seed(42) (or any fixed integer) immediately before slice_sample(). The same seed always returns the same rows.

How do I sample with replacement (bootstrap) in dplyr?

Pass replace = TRUE: slice_sample(df, n = nrow(df), replace = TRUE). The size matches the original; rows can repeat.

How do I do a weighted random sample?

Pass weight_by = column: slice_sample(df, n = 5, weight_by = price) makes rows with higher price more likely to be picked.

How do I do a stratified sample per group?

slice_sample(df, n = 3, by = group_col) returns 3 random rows per group. For a fraction per group, use prop = 0.1 instead of n.