Variable labels get dropped by mutate

[torfason]

Mutate does not preserve variable labels specified using the labelled package, making it more cumbersome to work with variable labels in a dplyr workflow. Admittedly, this is an interaction between two packages, but they are both part of the tidyverse, so very often used together I would imagine.

This seems to apply to mutate functions in general, which is a bummer, given how convenient in-place processing using mutate_at or the new across() syntax.

Using base R notation instead of mutate confirms that the issue is not with the intermediate functions but with mutate. In other words, assignment to columns using base R preserves the variable labels as expected. Demonstration below:

library(dplyr)
library(labelled)

# Mutate drops haven_labelled variable labels, but this seems wrong
f <- factor(c("yes", "yes", "no", "no", "don't know"))
value_labels <- c("yes" = 1, "no" = 2, "don't know" = 9)
dat <- tibble(responses = to_labelled(f,value_labels)) %>% 
  set_variable_labels(responses="Variable labels get dropped by mutate")
dat <- dat %>%
  mutate(responses_mut = to_factor(responses))

# Assignement using base R notation does not drop labels
dat$responses_base <- to_factor(dat$responses)
sapply(dat,var_label)
#> $responses
#> [1] "Variable labels get dropped by mutate"
#> 
#> $responses_mut
#> NULL
#> 
#> $responses_base
#> [1] "Variable labels get dropped by mutate"

# This seems to apply to mutate_* functions in general
# (note that the variable label now dropped from responses_base)
dat %>% 
  mutate_if(is.factor,to_labelled) %>%
  sapply(var_label)
#> $responses
#> [1] "Variable labels get dropped by mutate"
#> 
#> $responses_mut
#> NULL
#> 
#> $responses_base
#> NULL

# It also applies to the new across() notation
# (note that the variable label now dropped from responses_base)
dat %>% 
  mutate(across(is.factor,to_labelled)) %>%
  sapply(var_label)
#> $responses
#> [1] "Variable labels get dropped by mutate"
#> 
#> $responses_mut
#> NULL
#> 
#> $responses_base
#> NULL

[hadley]

labelled isn't part of the tidyverse, but this will by and large be fixed in dplyr 1.0.0.

[QQuantmod]

labelled isn't part of the tidyverse, but this will by and large be fixed in dplyr 1.0.0.

Mister hadley, now dplyr 1.1.3 will still drop variable's label using mutate. And I didn't see related description in mutate's Arguments.

[torfason]

Dear @QQuantmod,

As the original submitter, I can verify that this has indeed been fixed. The original reprex now looks as follows, confirming that all relevant dplyr functions preserve variable labels:

library(conflicted)
library(dplyr)
library(labelled)

packageVersion("dplyr")
#> [1] '1.1.4'

# Mutate no longer drops haven_labelled variable labels
f <- factor(c("yes", "yes", "no", "no", "don't know"))
value_labels <- c("yes" = 1, "no" = 2, "don't know" = 9)
dat <- tibble(responses = to_labelled(f,value_labels)) %>% 
  set_variable_labels(responses="Variable labels no longer dropped by mutate")
dat <- dat %>%
  mutate(responses_mut = to_factor(responses))

# Assignement using base R notation does not drop labels
dat$responses_base <- to_factor(dat$responses)
sapply(dat,var_label)
#>                                     responses 
#> "Variable labels no longer dropped by mutate" 
#>                                 responses_mut 
#> "Variable labels no longer dropped by mutate" 
#>                                responses_base 
#> "Variable labels no longer dropped by mutate"

# mutate_* functions now also preserve labels
dat %>% 
  mutate_if(is.factor,to_labelled) %>%
  sapply(var_label)
#>                                     responses 
#> "Variable labels no longer dropped by mutate" 
#>                                 responses_mut 
#> "Variable labels no longer dropped by mutate" 
#>                                responses_base 
#> "Variable labels no longer dropped by mutate"

# As does the (no longer so new) across() notation
dat %>% 
  mutate(across(is.factor,to_labelled)) %>%
  sapply(var_label)
#> Warning: There was 1 warning in `mutate()`.
#> ℹ In argument: `across(is.factor, to_labelled)`.
#> Caused by warning:
#> ! Use of bare predicate functions was deprecated in tidyselect 1.1.0.
#> ℹ Please use wrap predicates in `where()` instead.
#>   # Was:
#>   data %>% select(is.factor)
#> 
#>   # Now:
#>   data %>% select(where(is.factor))
#>                                     responses 
#> "Variable labels no longer dropped by mutate" 
#>                                 responses_mut 
#> "Variable labels no longer dropped by mutate" 
#>                                responses_base 
#> "Variable labels no longer dropped by mutate"