Functions that take variables

So far, our functions have taken vectors

new_mean(x), prop(x), and your new_sd(x) all take a vector of numbers.

But most of what you actually do all day takes a dataset and a variable:

summarize(nlsy, mean = mean(income, na.rm = TRUE))
summarize(nlsy, mean = mean(age_bir, na.rm = TRUE))
summarize(nlsy, mean = mean(nsibs, na.rm = TRUE))

This is the “copied it more than twice” signal. Silly example, as you could do it all in one summarize() call, but let’s write a function.

The obvious attempt

summarize_var <- function(data, variable) {
  data |>
    summarize(mean = mean(variable, na.rm = TRUE))
}

summarize_var(nlsy, income)

Error in `summarize()`:
ℹ In argument: `mean = mean(variable, na.rm =
  TRUE)`.
Caused by error:
! object 'income' not found

Well, that didn’t work.

Why not?

Here’s what we know works:

summarize(nlsy, mean = mean(income))

But income isn’t an object in your environment – if you typed income in the console you’d get an error. It only means something inside nlsy.

• {dplyr} does this on purpose. It’s called data masking, and it’s why you don’t have to write nlsy$income in the mean() call.
• But that means that when you write your own function, {dplyr} looks for a column literally named variable (that’s what you have in your function body!), doesn’t find one, and gives up.

The fix: {{ }}

Wrap the argument in curly-curly (I didn’t make up this name, it’s actually what they call it!) to say “don’t look for a column called variable, look for whatever the user passed in”:

summarize_var <- function(data, variable) {
  data |>
    summarize(
      n    = sum(!is.na({{ variable }})),
      mean = mean({{ variable }}, na.rm = TRUE),
      sd   = sd({{ variable }}, na.rm = TRUE)
    )
}

Now it works

summarize_var(nlsy, income)

# A tibble: 1 × 3
      n   mean     sd
  <int>  <dbl>  <dbl>
1 10195 14707. 12507.

summarize_var(nlsy, age_bir)

# A tibble: 1 × 3
      n  mean    sd
  <int> <dbl> <dbl>
1  5943  24.4  6.11

And because it takes the data first, it pipes:

nlsy |> summarize_var(nsibs)

{{ }} works anywhere {dplyr} does

Including .by =:

summarize_by <- function(data, variable, group) {
  data |>
    summarize(
      mean = mean({{ variable }}, na.rm = TRUE),
      n    = n(),
      .by  = {{ group }}
    )
}

summarize_by(nlsy, income, sex_cat)

# A tibble: 2 × 3
  sex_cat   mean     n
  <fct>    <dbl> <int>
1 Female  14523.  6283
2 Male    14880.  6403

You can even name the output after the variable

Use "{{ variable }}" inside a string, with := instead of =:

mean_named <- function(data, variable) {
  data |>
    summarize("mean_{{ variable }}" := mean({{ variable }}, na.rm = TRUE))
}

mean_named(nlsy, income)

# A tibble: 1 × 1
  mean_income
        <dbl>
1      14707.

It works in ggplot() too

plot_hist <- function(data,
                      variable) {
  ggplot(
    data,
    aes(x = {{ variable }})) +
    geom_histogram(bins = 30) +
    theme_minimal()
}

plot_hist(nlsy, income)

And in {gtsummary}

table_by <- function(data,
                     group) {
  data |>
    tbl_summary(
      by = {{ group }},
      include = c(
        race_eth_cat,
        eyesight_cat, age_bir)
    ) |>
    add_overall() |>
    bold_labels()
}

table_by(nlsy, sex_cat)

Same function, different grouping variable

table_by(nlsy, region_cat)

You only have to write the formatting once! Changing the stratifying variable is now just one word, not a copy-paste.

Careful: {{ }} is a tidyverse feature

It works because dplyr, ggplot2, and gtsummary are built to understand it. Base R functions are not.

fit_income <- function(data, predictor) {
  lm(income ~ {{ predictor }}, data = data)
}

fit_income(nlsy, age_bir)

Error:
! object 'age_bir' not found

A note on asking AI about this

This is a corner of R that changed a lot, and models are frequently out of date here.

If you see aes_string(), !!sym(), .dots =, or standardise_call() in a suggestion, it’s giving you a solution from before {{ }} existed (2019).

Exercises

The website has a script to download with these examples, plus some additional exercises.