forcats fct_lump() in R: Collapse Rare Factor Levels

The forcats fct_lump() function collapses rare factor levels in R into a single "Other" category, turning a long-tailed factor into a tidy handful of groups.

fct_lump(x)                          # auto-lump rare levels into "Other"
fct_lump(x, n = 5)                   # keep the 5 most common levels
fct_lump(x, prop = 0.1)              # keep levels above 10% frequency
fct_lump(x, n = -3)                  # lump the 3 most common, keep the rest
fct_lump(x, w = weights)             # rank levels by weighted counts
fct_lump(x, other_level = "Misc")    # rename the catch-all bucket
fct_lump_min(x, min = 10)            # keep levels seen at least 10 times

What fct_lump() does in one sentence

fct_lump() collapses a factor's infrequent levels into a single "Other" level. It comes from the forcats package, part of the tidyverse. A categorical column with dozens of rare values, such as product names or cities, is hard to chart and model. fct_lump() keeps the levels that matter and sweeps the long tail into one labelled bucket, controlled by a count (n), a proportion (prop), or an automatic heuristic.

Syntax

fct_lump() takes a factor plus one rule for how many levels to keep. The signature is short:

The arguments are:

• f: a factor, or any vector that can be coerced to one (character, numeric, or logical).
• n: keep the n most common levels and lump the rest. A negative n flips it, lumping the n most common and keeping the rare tail.
• prop: keep levels that appear at least proportion prop of the time. A negative prop lumps the common levels instead.
• w: an optional numeric weight vector, one value per observation, so levels are ranked by summed weight rather than a plain row count.
• other_level: the name of the catch-all level. Defaults to "Other".
• ties.method: how to break ties when two levels share a count near the n cutoff.

Supply either n or prop, never both. With neither, fct_lump() uses a heuristic that lumps the least common levels as long as the "Other" level stays the smallest.

fct_lump() examples

Example 1 shows the automatic heuristic with no n or prop. Load forcats and call fct_lump() on a vector with a clear long tail.

The rare levels c, d, and e (one each) merge into Other, which totals 3. fct_lump() stopped there because lumping b as well would make Other larger than the levels it kept.

Example 2 keeps a fixed number of levels with n. This is the most common use: a bar chart or model needs a manageable count of categories.

The three busiest manufacturers stay named; the other twelve collapse into Other.

Example 3 keeps levels above a frequency threshold with prop. Use prop when the cutoff should scale with the data instead of a hard count.

prop = 0.1 keeps any manufacturer that makes up at least 10 percent of the 234 rows. Four brands clear the bar; the rest become Other.

Example 4 renames the catch-all bucket with other_level. The default label "Other" is generic, so set a name that reads well on a chart axis or in a report.

fct_lump() vs the fct_lump_*() family

fct_lump() is the umbrella; the fct_lump_*() variants each expose one rule. Reach for the variant that names the rule you want.

Common pitfalls

fct_lump() rejects n and prop supplied together. The two arguments are competing rules, so passing both is an error rather than a silent choice.

Pick one rule: use n for a fixed number of bars, prop for a frequency cutoff that scales with the data.

The "Other" bucket can outgrow every level you keep. When you lump aggressively, the catch-all sums many small levels and can become the single largest category.

With n = 2, Other holds 163 of 234 rows and dwarfs the two named brands.

Try it yourself

Related forcats functions

These forcats functions pair naturally with fct_lump() for level management.

• fct_other(): keep or drop named levels explicitly, lumping the rest into "Other".
• fct_collapse(): merge groups of levels into named categories by hand.
• fct_recode(): rename levels one by one.
• fct_infreq(): order levels by frequency, a natural follow-up after lumping.
• Categorical Data in R: the full guide to factors.

See the forcats reference for the official documentation.

FAQ

What is the difference between fct_lump() and fct_lump_n()?

They do the same job with different ergonomics. fct_lump(x, n = 5) and fct_lump_n(x, 5) both keep the five most common levels and lump the rest. fct_lump() is the older umbrella function that switches behavior based on which argument you pass. fct_lump_n() is the newer variant that names the rule directly. Recent forcats versions supersede fct_lump() in favor of the explicit fct_lump_n(), fct_lump_prop(), and fct_lump_min() functions, so new code reads better with the variant.

How do I rename the "Other" category created by fct_lump()?

Pass the other_level argument with the label you want, for example fct_lump(x, n = 4, other_level = "Smaller brands"). The catch-all level then carries your name instead of the default "Other". This is worth doing whenever the factor will appear on a chart axis or in a report, where a generic "Other" reads poorly. The argument works the same way across fct_lump_n(), fct_lump_prop(), and the other variants.

Does fct_lump() work on character vectors?

Yes. fct_lump() accepts a factor or any vector that can be coerced to one, including character, numeric, and logical vectors. When you pass a character vector, fct_lump() converts it to a factor first, then returns a factor with the rare levels collapsed. The original values are preserved for the levels that are kept. If you need the result back as plain text, wrap the call in as.character().

How do I keep the rarest levels instead of the most common?

Pass a negative n or prop. The call fct_lump(x, n = -3) lumps the three most common levels into "Other" and keeps the rare tail untouched. Likewise fct_lump(x, prop = -0.1) lumps any level above 10 percent frequency. This inversion is useful when the frequent categories are the noise and the rare ones are the signal, such as flagging unusual error codes or uncommon transactions.