Subsetting loses variable labels

[Crismoc]

When I import SPSS data with read_sav() variable labels are correctly imported.

data<-haven::read_sav("data.sav")
str(data$sexo)
> 'labelled' num [1:738] 1 0 0 1 1 0 1 0 0 0 ...
 - attr(*, "label")= chr "Sexo (0: Hombre, 1: Mujer)"
 - attr(*, "format.spss")= chr "F8.2"
 - attr(*, "display_width")= int 10
 - attr(*, "labels")= Named num [1:2] 0 1
  ..- attr(*, "names")= chr [1:2] "Mujer" "Hombre"

Then I create multiple subsamples of it and write new SPSS data frames with write_sav().

haven::write_sav(data[sample(nrow(datos),350),], "data_subsample.sav")

When opening the new dataframes in SPSS -or reading them back with read_sav()- variable labels are lost. Is there a way of fixing this?

Thanks!

R version 3.5.0 (2018-04-23)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows >= 8 x64 (build 9200)

Matrix products: default

locale:
[1] LC_COLLATE=English_United States.1252  LC_CTYPE=English_United States.1252   
[3] LC_MONETARY=English_United States.1252 LC_NUMERIC=C                          
[5] LC_TIME=English_United States.1252    

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] haven_1.1.1

loaded via a namespace (and not attached):
[1] compiler_3.5.0 magrittr_1.5   tools_3.5.0    pillar_1.2.3   tibble_1.4.2  
[6] yaml_2.1.19    Rcpp_0.12.17   forcats_0.3.0  rlang_0.2.1  

[huftis]

Ah, I see. The reason for this bug is that subsetting the variables loses their attributes. And the reason for this is that the variables with only a label but no labels don’t have custom class (e.g. haven_labelled) (for which a subsetting method is defined).

Note that this only applied to variables that have a label attribute but don’t have a labels attribute. (So, for instance, the fare variable doesn’t work, but the Residence variable does.)

A possible solution would be for the importer to store the variables as haven_labelled objects. The class haven_labelled would then have to be redefined to have either 1) labels, 2) a label, or 3) both labels and a label. The class name would fit this definition, but some functions using haven_labelled objects would probably have to be redefined to accept labels-less haven_labelled objects.

Here’s a reprex:

library(haven)
d <- read_sav("https://www.sheffield.ac.uk/polopoly_fs/1.547009!/file/Titanic_2.sav")

class(d$Residence) # OK
#> [1] "labelled"
class(d$fare) # Missing class
#> [1] "numeric"

# Subsetting ‘fare’ loses the attribute
attr(d$fare, "label")
#> [1] "Cost of ticket"
attr(d$fare[1:3], "label")
#> NULL

# But subsetting ‘Residence’ works fine
attr(d$Residence, "label")
#> [1] "Country of residence"
attr(d$Residence[1:3], "label")
#> [1] "Country of residence"

[hadley]

I think the proposal to always use haven_labelled would create more problems than it solves, because more people would be force to confront an exotic variable type that doesn't actually bear on their analysis.

Instead I think you're better off creating a custom restore labels functions, something like this (untested, but should be close):

restore_labels <- function(df, orig) {
  for (var in names(original)) {
    if (is.null(df[[var]]))
      next
    
    if (!is.null(attr(df[[var]], "label")))
      next
    
    if (is.null(attr(orig[[var]], "label")))
      next
    
    attr(df[[var]], "label") <- attr(orig[[var]], "label")
  }
  
  df
}

[larmarange]

labelled package propose a copy_labels function doing exactly this: http://larmarange.github.io/labelled/reference/copy_labels.html

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/