`write_dta` can't save numeric labelled variables without value labels

[domingoscisci]

Hi all,
I have a data.frame of double (no integer) variables that I want to label (only variable label, not value labels). If I try to use the labelled() function, a vector of class dbl+lbl is correctly created, but it cannot be saved in a Stata file using write_dta(). If I use the attr() function the problem does not arise.
It is probably not correct to use labelled() to label the variable only. In this case, perhaps it might be useful to make this explicit within the documentation (maybe I missed something?).

Here is a reprex:

library(tibble)
library(haven)


dat <- tibble(
  a = c(0.1, 0.2, 0.3, 0.4, 0.5)
)

# label variable with haven::labelled
dat$a <- labelled(dat$a, label = "test label")

# write_dta fails
write_dta(dat, tempfile(fileext = "dta"))
#> Error: Stata only supports labelling with integers.
#> Problems: `a`


dat <- tibble(
  a = c(0.1, 0.2, 0.3, 0.4, 0.5)
)

# label variable with attr
attr(dat$a, "label") <- "test label"

# write_dta ok
write_dta(dat, tempfile(fileext = "dta"))

Thank you for your help,
Domingo

[jmobrien]

Seconding this issue, and noting that it is uniquely a problem with non-integer numerics lacking value labels:

require(haven) # 2.4.1
#> Loading required package: haven
df <- 
  data.frame(
    x1 = labelled(1:10),
    x2 = labelled(as.numeric(1:10), labels = c(one = 1)),
    x3 = labelled(as.character(1:10)),
    x4 = labelled(as.numeric(1:10))
  )

path <- tempfile()
write_dta(df, path)
#> Error: Stata only supports labelling with integers.
#> Problems: `x4`

^{Created on 2021-06-03 by the reprex package (v2.0.0)}

[jmobrien]

Problem seems to be here:

haven/R/haven-stata.R

Line 150 in dbd65a3

!is_integerish(attr(x, "labels"))

is_integerish() returns FALSE if the labels are missing, producing this issue. Submitted PR #609 to address.

[dusadrian]

If you don't mind a workaround, this solves your example:

install.packages("declared")
devtools::install_github("dusadrian/DDIwR")
library(declared)
library(DDIwR)
df <- as_declared(df)
convert(df, to = "Stata")

# or convert(df, to = "path/to/file.dta")

I would highly recommend using declared objects instead of labelled. Apart from being superior in treating missing values, they also allow variables without value labels, and they are fully compatible with the traditional labelled objects.

Naturally, you can create your df directly:

df <- 
  data.frame(
    x1 = declared(1:10),
    x2 = declared(as.numeric(1:10), labels = c(one = 1)),
    x3 = declared(as.character(1:10)),
    x4 = declared(as.numeric(1:10))
)
convert(df, to = "Stata")