janitor adorn_rounding() in R: Round Tabyl Cells Cleanly
The janitor adorn_rounding() function rounds the numeric cells of a data frame to a chosen number of decimal places while keeping every column numeric. It is the cleanup step in the adorn chain when you want rounded values you can still pipe into ggplot(), joins, or further math, instead of the percent strings produced by adorn_pct_formatting().
adorn_rounding(df) # default: 1 decimal, "half to even" adorn_rounding(df, digits = 0) # round to integers adorn_rounding(df, digits = 3) # high precision adorn_rounding(df, rounding = "half up") # 0.5 rounds up, not to even adorn_rounding(df, , q1, q2) # round only q1 and q2 tabyl(x) |> adorn_percentages() |> adorn_rounding(digits = 2) # clean numeric chain mtcars |> group_by(cyl) |> summarise(mpg = mean(mpg)) |> adorn_rounding()
Need explanation? Read on for examples and pitfalls.
What adorn_rounding() does in one sentence
adorn_rounding() rounds every numeric column of a data frame to a fixed number of decimal places and returns the same data frame class. The first column is treated as an identifier and skipped, mirroring the rest of the janitor adorn family.
Unlike adorn_pct_formatting(), it does not multiply by 100, does not paste a "%" symbol, and never converts cells to character strings. Use it whenever you need rounded values for further computation, plotting, or joining, not just for display.
Syntax
adorn_rounding() takes the data frame plus three optional arguments that control precision, rounding mode, and which columns are rounded. Only the data frame is required.
The full signature:
adorn_rounding(dat, digits = 1, rounding = "half to even", ...)digits sets how many decimal places survive (default 1). rounding is either "half to even" (banker's rounding, the default) or "half up". The ... slot accepts tidyselect expressions when only specific columns should be rounded. The first column is always skipped, so identifiers like gear stay untouched even when numeric.
Five common patterns
These five patterns cover every rounding decision you will face in a tabyl chain. Each block reuses props from the syntax section, so you can run them in order.
1. Default rounding
Calling adorn_rounding(props) rounds each numeric column to 1 decimal place. This is rarely what you want for proportions (it collapses too much), but it is the right default for summary tables of integers and dollar amounts.
The cells are still numeric (class numeric), so you can pipe the result into ggplot() or dplyr::mutate() without converting back.
2. Increase digits for precision
Set digits to 2 or 3 when the proportions span a wide range and 1 decimal loses meaningful detail. Two decimals is the typical sweet spot for tabyl outputs.
The same call with digits = 3 keeps three decimals (0.333, 0.667), which matches what most reports want when the audience cares about percentage points below 1.
adorn_rounding(digits = 2) with adorn_percentages() for clean numeric output. This is the "I still want to do math on these values" alternative to adorn_pct_formatting(). The result is a data frame you can join, plot, or write to a CSV without parsing percent strings.3. Switch rounding mode to "half up"
The default "half to even" (banker's rounding) sends 0.5 to the nearest even integer to reduce cumulative bias, but it surprises readers who expect 0.5 to always round up. Pass rounding = "half up" for the conventional behavior.
Half-to-even sends 2.5 to 2, 3.5 to 4, 4.5 to 4, and 5.5 to 6 (each toward the nearer even integer). Half-up sends every .5 away from zero. The choice matters most when a column has many midpoint values; for typical messy data the two modes agree almost everywhere.
4. Round selected columns only
The ... argument accepts a tidyselect expression so non-percent columns survive untouched. This is the right call when a frame mixes proportions with raw counts or currency.
Only share was rounded. The revenue column kept its full precision and the region identifier was skipped.
5. Round any summary data frame
adorn_rounding() is not exclusive to tabyls. Any data frame with an identifier column and numeric value columns will round cleanly.
Month stayed as an integer because it is the first column. The other three columns rounded to 1 decimal.
Compare with alternatives
Base R offers round(), the tidyverse offers dplyr::mutate(across()), and signif() rounds by significant digits. adorn_rounding() wins on convenience inside janitor pipelines and loses on flexibility outside them.
| Approach | Best for | Watch out for |
|---|---|---|
janitor::adorn_rounding() |
Tabyl chains, identifier-first frames | Always skips column 1 |
base::round(x, digits) |
One-off vector or matrix rounding | No tidyverse pipe friendliness |
dplyr::mutate(across(where(is.numeric), round, 2)) |
Column subsetting by predicate | Verbose for the common case |
base::signif(x, digits) |
Significant figures, not decimals | Different semantics than digits-after-decimal |
adorn_rounding() is math that stays numeric, adorn_pct_formatting() is presentation that becomes character. Pick the rounder when the values feed into another calculation, plot, or join. Pick the formatter when the next step is a rendered report or kable table.Common pitfalls
Pitfall 1: rounded percentages do not sum to 100. Three proportions of 0.3333 rounded to digits = 1 give 0.3, 0.3, 0.3 totalling 0.9 (90 percent), not 1. Append adorn_totals() before adorn_rounding() to compute totals from unrounded values, or accept the visible mismatch and explain it in the table caption.
Pitfall 2: digits = 0 on proportions collapses everything. A 0.4-versus-0.6 split becomes 0-versus-1, which is misleading. Use digits = 0 only on count-scale data (counts, currency, integer summaries), not on 0-to-1 proportions.
Pitfall 3: banker's rounding looks wrong on contrived examples. Inputs like 0.5, 1.5, 2.5, 3.5 round to 0, 2, 2, 4, which feels off compared to the conventional 1, 2, 3, 4. On real, noisy data the difference disappears because exact midpoints almost never occur. Document the choice in the table caption if your audience expects half-up rounding.
Try it yourself
Try it: Take mtcars, build a tabyl(cyl, gear) cross-tab, convert to row percentages, and round to 3 decimal places using "half up" mode. Save the result to ex_rounded.
Click to reveal solution
Explanation: denominator = "row" divides each cell by its row total so rows sum to 1. adorn_rounding(digits = 3, rounding = "half up") keeps three decimals using conventional half-up rounding. The cells remain numeric, so you can pipe the result into a join or a plot without parsing strings.
Related janitor functions
adorn_rounding() is one of seven adorn helpers that polish tabyl output. Each call preserves the input class so the next call works.
tabyl(): the upstream frequency builderadorn_percentages(): convert counts to proportions before roundingadorn_pct_formatting(): alternative when you want "47.2%" strings instead of numericsadorn_totals(): append totals row or column before roundingadorn_ns(): paste raw counts onto rounded percentage cellsadorn_title(): attach a banner row above a tabylclean_names(): standardize column names before any of the above
See the janitor reference on tidyverse.org for the full argument list.
FAQ
What is the difference between adorn_rounding() and adorn_pct_formatting()?
adorn_rounding() keeps cells numeric and does not multiply by 100 or paste a "%" symbol. adorn_pct_formatting() rounds, multiplies by 100, appends "%", and returns character strings. Use the rounder when downstream code needs numbers (plots, joins, further math). Use the formatter when the next step is rendering a kable table or pasting into a report.
Why do my rounded proportions not sum to 1?
Rounding loses information. Three cells of 0.3333 rounded to 1 decimal give 0.3, 0.3, 0.3, which sum to 0.9, not 1. The mismatch is mathematical, not a janitor bug. Compute adorn_totals() first and round afterward if you want the totals row to read exactly 1.00, or accept the visible gap.
Can I use adorn_rounding() without a tabyl?
Yes. The function accepts any data frame whose first column is an identifier and whose remaining columns are numeric. Group summaries from dplyr::summarise(), descriptive statistics from psych::describe(), or any handmade frame works. The tabyl class is convenient but not required.
What does "half to even" rounding mean?
"Half to even" (banker's rounding) rounds exact midpoints like 0.5 to the nearest even integer. So 0.5 becomes 0, 1.5 becomes 2, 2.5 becomes 2, 3.5 becomes 4. The aim is to remove the upward bias that "half up" introduces across many roundings. The cumulative bias is usually negligible, but the difference is visible on small, contrived examples like a 4-row demo table.
How is adorn_rounding() different from base R round()?
round() operates on a single numeric vector or matrix. adorn_rounding() operates on a whole data frame, skips the first column, lets you pick rounding mode ("half to even" or "half up"), and chains cleanly with other adorn_* helpers. Inside a tabyl pipeline adorn_rounding() is the natural choice. Outside one, plain round() is shorter.