dplyr arrange(), slice(), and top_n(): Get Exactly the Rows You Want

In dplyr, arrange() sorts rows, slice() picks rows by position, and the slice_*() family grabs rows by value or at random. Together they answer one question analysts ask every day: "give me exactly these rows, in this order."

How does arrange() sort rows in dplyr?

When you need to rank, compare, or just eyeball the biggest and smallest values, sorting is the first move. arrange() reorders rows by one or more columns, ascending by default, wrap a column in desc() for descending. Here are the fastest cars in mtcars, ranked by quarter-mile time (smaller qsec = faster):

The Ford Pantera L comes out on top with a 14.5-second quarter mile, followed closely by the Maserati Bora at 14.6. Notice the row names stay attached, arrange() moves whole rows, not just the sort column. That's a key property: every column in every row travels together, so your table remains consistent after sorting.

How do you sort by multiple columns at once?

Single-column sorting is fine until you hit ties. What if you want all four-cylinder cars first, then six-cylinder, then eight, and within each cylinder group, the highest-mpg car on top? Pass multiple columns to arrange() and dplyr sorts by the first column, then uses the second as a tie-breaker, and so on. The order you list the columns matters.

The four-cylinder block leads, and within it the Toyota Corolla (33.9 mpg) sits at the top. Scroll further and you'd see six-cylinder cars start, then eight-cylinder, always sorted high-to-low by mpg inside each group. This is exactly how SQL's ORDER BY col1, col2 works, and it's how you build leaderboards that respect natural categories.

What's the difference between arrange() and base R's order()?

If you've used base R you already know order(), it returns the indices that would sort a vector, and you use those indices to subset the data frame. arrange() skips that indirection: it takes the whole data frame, sorts it, and hands it back. Same result, half the keystrokes and none of the bracket gymnastics.

Both lines produce the same sorted table, confirmed by identical() on the row names. The dplyr version reads left-to-right like English, take mtcars, arrange by cyl then descending mpg, while the base R version needs you to parse nested bracket syntax and remember that -mtcars$mpg is the trick for descending. For interactive analysis, arrange() wins on readability every time.

How do you pick rows by position with slice()?

Sometimes you don't care about values, you just want "row 5" or "rows 10 through 15" or "everything except the first row." That's what slice() does: it subsets rows by their integer position in the table. It accepts a single index, a range with :, a vector with c(), or negative indices to exclude rows.

That's the first five rows of mtcars, position 1 through 5, in their original order. To grab specific non-contiguous rows, pass a vector: slice(c(1, 3, 5)). To drop the first row instead of keeping it, use a negative index: slice(-1). And to drop several, slice(-c(1, 2)). The pattern mirrors base R indexing, but it returns a proper tibble and plays nicely with pipes.

When should you use slice_head(), slice_tail(), slice_min(), slice_max()?

dplyr ships a whole family of slice_*() helpers, each one answering a different flavor of "give me rows." They spare you from writing arrange() |> head() every time and they handle ties and sampling gracefully. The right choice depends on how you want to pick:

Figure 1: Choosing the right slice_() variant based on how you want to pick rows.*

Here's slice_max() in action, no arrange() needed:

Five lines of output, ranked heaviest first, with Jabba unsurprisingly dominating at 1358 kg. Behind the scenes slice_max() sorts by mass descending and takes the top 5, but the verb name reads directly as intent: slice the max. Prefer these helpers over arrange(desc(mass)) |> head(5) when you want self-documenting code.

How do you get the top N rows per group?

Here's the pattern that makes slice_max() genuinely powerful: combine it with group_by() and you get top-N-per-group in one short pipeline. "Top 2 most fuel-efficient cars within each cylinder class", that's a question analysts ask constantly, and it would take an ugly loop in base R.

Figure 2: Getting the top rows within each group by combining group_by() and slice_max().

Six rows total, two per cylinder class, ranked highest mpg first within each group. The Toyota Corolla leads the 4-cylinder class, a Hornet tops the 6-cylinder, and a Pontiac leads the 8-cylinder. The ungroup() at the end is a good habit: it clears the grouping so downstream operations act on the whole table.

Is top_n() still the right way to get the top rows?

If you've read older dplyr tutorials you've seen top_n(), a function that grabs the top N rows by some column. It still works, but as of dplyr 1.0.0 (May 2020) it's officially superseded by slice_max() and slice_min(). Superseded means "still supported forever, but not the recommended choice anymore", new code should use the slice family.

Both lines return the three highest-mpg cars. Why was top_n() superseded? Two reasons: its argument order (n first, then the column) was inconsistent with the rest of the slice family, and slice_max() offers explicit with_ties and prop arguments for controlling ties and proportional sampling. The migration is a one-for-one replacement, no behavior changes to worry about.

Practice Exercises

These capstones combine arrange, slice, and group_by patterns. Use my_* prefixed variables so exercise code doesn't clobber tutorial state.

Exercise 1: Second-heaviest per species

In starwars, find the second-heaviest character of each species (not the heaviest, exactly the second). Ignore rows with missing mass. Save to my_second_heaviest with columns name, species, and mass.

Exercise 2: Bottom 3 and top 3 in one table

From mtcars, build a single table containing the 3 worst-mpg and 3 best-mpg cars, with a rank column marking "bottom" or "top". Save to my_extremes.

Exercise 3: Random sample stratified by group

Take a random sample of 2 cars from each cylinder group in mtcars, reproducible with a seed. Save to my_stratified_sample. Result should have 6 rows total (2 per cylinder class × 3 classes).

Complete Example

Here's a realistic end-to-end pipeline that threads every verb from this tutorial together. Question: what are the 3 tallest Star Wars characters in each homeworld that has at least 2 characters, sorted by homeworld and height?

Every line corresponds to one intent: drop rows with missing height or homeworld, for each homeworld, keep only homeworlds with 2+ characters, take the 3 tallest, sort nicely, clean up grouping, show just the columns we care about. That's the tidyverse in its natural habitat, each verb small and focused, the pipeline reading top to bottom like a recipe.

Summary

References

1. dplyr reference, arrange(). Link
2. dplyr reference, slice() and the slice_*() family. Link
3. dplyr reference, top_n() (superseded). Link
4. Posit tidyverse blog, dplyr 1.0.0 release notes introducing slice_*(). Link
5. Wickham, H., & Grolemund, G., R for Data Science, Chapter 4: Data Transformation. Link
6. R documentation, order() base function. Link
7. dplyr grouping vignette, grouped operations. Link

Continue Learning

1. dplyr filter() and select(), Row and column subsetting, the verbs you reach for alongside arrange and slice.
2. dplyr group_by() and summarise(), Aggregate data by group, pairs naturally with grouped slice_max for top-N-per-group.
3. dplyr mutate() and rename(), Add computed columns, useful for creating the columns you'll sort or slice by.