Data wrangling across columns

Warmup activity

Work on the activity (handout) with a neighbor, then we will discuss as a class

Warmup

Your friend writes the following code:

starwars |>
  summarize(across(where(is.numeric),
                   list("median" = median),
                   .names = "{.fn}_{.col}"))

# A tibble: 1 × 3
  median_height median_mass median_birth_year
          <int>       <dbl>             <dbl>
1            NA          NA                NA

Why are they getting NAs?

Ignoring NAs

starwars |>
  summarize(median_height = median(height))

# A tibble: 1 × 1
  median_height
          <int>
1            NA

What would I change to ignore missing values (NAs) when computing the median?

Ignoring NAs

starwars |>
  summarize(median_height = median(height, na.rm=T))

# A tibble: 1 × 1
  median_height
          <int>
1           180

Now let’s try with across…

Ignoring NAs

starwars |>
  summarize(across(where(is.numeric),
                   list("median" = median(na.rm=T))))

Error in `summarize()`:
ℹ In argument: `across(where(is.numeric), list(median = median(na.rm =
  T)))`.
Caused by error in `median.default()`:
! argument "x" is missing, with no default

Why is this code failing?

Ignoring NAs

starwars |>
  summarize(across(where(is.numeric),
                   list("median" = median(na.rm=T))))

median is a function:

median

function (x, na.rm = FALSE, ...) 
UseMethod("median")
<bytecode: 0x13cf29ab8>
<environment: namespace:stats>

median() is evaluating (calling) the function:

median()

Error in median.default(): argument "x" is missing, with no default

Ignoring NAs

starwars |>
  summarize(across(where(is.numeric),
                   list("median" = median(na.rm=T))))

Error in `summarize()`:
ℹ In argument: `across(where(is.numeric), list(median = median(na.rm =
  T)))`.
Caused by error in `median.default()`:
! argument "x" is missing, with no default

What should we do instead?

Writing a new function

We want a function that calculates the median without the NAs, so we can do something like

starwars |>
  summarize(across(where(is.numeric),
                   list("median" = median_no_na)))

However, this median_no_na function doesn’t exist. We have to write it ourselves!

Writing a new function

median_no_na <- function(x) {
  median(x, na.rm = T)
}

Writing a new function

median_no_na <- function(x) {
  median(x, na.rm = T)
}

What will each of the following lines return?

median(c(0, 1, 2, NA))

median(c(0, 1, 2, NA), na.rm=T)

median_no_na(c(0, 1, 2, NA))

Writing a new function

median_no_na <- function(x) {
  median(x, na.rm = T)
}

median(c(0, 1, 2, NA))

[1] NA

median(c(0, 1, 2, NA), na.rm=T)

[1] 1

median_no_na(c(0, 1, 2, NA))

[1] 1

Writing a new function

median_no_na <- function(x) {
  median(x, na.rm = T)
}

starwars |>
  summarize(across(where(is.numeric),
                   list("median" = median_no_na)))

# A tibble: 1 × 3
  height_median mass_median birth_year_median
          <int>       <dbl>             <dbl>
1           180          79                52

What would I change if I want to calculate the mean instead of the median?

Writing a new function

mean_no_na <- function(x) {
  mean(x, na.rm = T)
}

starwars |>
  summarize(across(where(is.numeric),
                   list("mean" = mean_no_na)))

# A tibble: 1 × 3
  height_mean mass_mean birth_year_mean
        <dbl>     <dbl>           <dbl>
1        175.      97.3            87.6

Will we need to use the mean_no_na or median_no_na functions many times?

Anonymous functions

If we don’t need a function repeatedly, we can make an anonymous function instead:

Anonymous function:

starwars |>
  summarize(across(where(is.numeric),
                   list("mean" = function(x) mean(x, na.rm=T))))

# A tibble: 1 × 3
  height_mean mass_mean birth_year_mean
        <dbl>     <dbl>           <dbl>
1        175.      97.3            87.6

Class activity

https://sta279-f25.github.io/class_activities/ca_08.html

• Work with a neighbor on the class activity
• At the end of class, submit your work as an HTML file on Canvas (one per group, list all your names)