case_when within mutate loses variable labels

[rempsyc]

Maybe this is expected/not a bug, but it seems that case_when() within mutate() loses variable labels, although mutate() alone or with replace() doesn't. Reprex:

suppressWarnings(suppressPackageStartupMessages(library(dplyr)))
suppressPackageStartupMessages(library(sjlabelled))

data <- readRDS(url("https://osf.io/d4mjk/?action=download"))

get_label(data$Status)
#>          Status 
#> "Response Type"

data <- data %>% 
  mutate(Status = Status + 1)

get_label(data$Status)
#>          Status 
#> "Response Type"

data <- data %>% 
  mutate(Status = replace(Status, Status == 1, 50))

get_label(data$Status)
#>          Status 
#> "Response Type"

data <- data %>% 
  mutate(Status = case_when(Status == 50 ~ 1))

get_label(data$Status)
#> NULL

^{Created on 2023-05-23 with reprex v2.0.2}

Could it be because case_when() is vectorized? And given that I wish to preserve labels, what workaround would you recommend? Falling back to replace()? Thank you.

Edit:

Linked to #5762. Which suggested recode() as the workaround. However, recode() has been superseded by case_match().

But case_match() does not seem to solve the issue:

suppressWarnings(suppressPackageStartupMessages(library(dplyr)))
suppressPackageStartupMessages(library(sjlabelled))

data <- readRDS(url("https://osf.io/d4mjk/?action=download"))

get_label(data$Status)
#>          Status 
#> "Response Type"

data <- data %>% 
  mutate(Status = case_match(Status, 1 ~ 50))

get_label(data$Status)
#> NULL

^{Created on 2023-05-23 with reprex v2.0.2}

[DavisVaughan]

sjlabelled just adds a label attribute onto a double or integer vector. That isn't enough for our underlying infrastructure to be able to figure out that that information is important. It needs to use a real class like haven::labelled(). Then the information carries over as expected:

library(haven)
library(dplyr)

x <- labelled(c(1, 2, 1, 3, 2), label = "response type")
x
#> <labelled<double>[5]>: response type
#> [1] 1 2 1 3 2

# Doesn't ever "see" `x`. It gets a logical vector and `50`
case_when(x == 1 ~ 50)
#> [1] 50 NA 50 NA NA

# Explicitly tell `case_when()` to use `x` as the type
case_when(x == 1 ~ 50, .ptype = x)
#> <labelled<double>[5]>: response type
#> [1] 50 NA 50 NA NA

# Or supply `x` as the default if you just want to replace part of it
case_when(x == 1 ~ 50, .default = x)
#> <labelled<double>[5]>: response type
#> [1] 50  2 50  3  2

# Similar thing here
case_match(x, 1 ~ 50, .default = x)
#> <labelled<double>[5]>: response type
#> [1] 50  2 50  3  2

So I'd suggest switching to haven if you can. Or possibly manually converting the sjlabelled results over to haven.

[rempsyc]

I see, thank you. You will notice that I did not use sjlabelled to define labels, only to read them. The labels were produced with the qualtRics package. Reprex:

suppressWarnings(suppressPackageStartupMessages(library(qualtRics)))
suppressWarnings(suppressPackageStartupMessages(library(sjlabelled)))
suppressWarnings(suppressPackageStartupMessages(library(haven)))

# Extract all surveys
surveys <- all_surveys()

# # Identify right survey
survey1.id <- surveys$id[
  which("Projet priming-aggression (Part 1)_Study 3" == surveys$name)]

# # Fetch right survey
data <- suppressMessages(fetch_survey(surveyID = survey1.id, verbose = FALSE))

# sjlabelled works
get_label(data$Status)
#>          Status 
#> "Response Type"

# haven doesn't work
print_labels(data$Status)
#> Error in `print_labels()`:
#> ! `x` must be a labelled vector.
#> Backtrace:
#>     ▆
#>  1. └─haven::print_labels(data$Status)
#>  2.   └─cli::cli_abort("{.arg x} must be a labelled vector.")
#>  3.     └─rlang::abort(...)

^{Created on 2023-08-07 with reprex v2.0.2}

Maybe @jmobrien from the qualtRics package can chip in on this?

[jmobrien]

@rempsyc happy to discuss more over on the qualtRics page. We're aware but there's some history/complexity around labels & the labelled class, so for up to this point we've not added extended classes like labelled. There are a couple of workflows to get around it, though, that might serve your use case(s).