read_dta: import value labels for labeled negative values

[kwenzig]

Example:

tmp <- tempdir()
x1 <- labelled(
  sample(-1:5), 
  c(Good = -1, Bad = 5)
)
df <- tibble::data_frame(x1, z = 1:7)
as_factor(df)
write_dta(df, paste0(tmp,"/df.dta"),version=14)
df1 <- read_dta(paste0(tmp,"/df.dta"))
as_factor(df1)

Here the result of as_factor(df) contains a row Good 7, while the corresponding row of as_factor(df1) is -1 7.00.

The produced Stata file is correct:

. label list x1
x1:
          -1 Good
           5 Bad

[cschwem2er]

We're currently having trouble with saving negative value labels in sjlabelled, which relies on havens functions for reading and writing. But we're not 100% sure whether this is the same problem (see strengejacke/sjlabelled#7). Maybe @hadley has an idea.

[kwenzig]

In my example constructing a data set and saving with write_dta works within R: The Stata file has a negative value with the correct label. The problem is reading this file with read_dta.

[hadley]

Slightly more minimal reprex:

library(haven)

tmp <- tempfile()
x1 <- labelled(
  sample(-1:3), 
  c(Good = -1, Bad = 3)
)
df <- tibble::data_frame(x1)
as_factor(df)$x1
#> [1] Bad  Good 1    2    0   
#> Levels: Good 0 1 2 Bad

write_dta(df, tmp, version = 14)
df1 <- read_dta(tmp)
as_factor(df1)$x1
#> [1] Bad -1  1   2   0  
#> Levels: -1 0 1 2 Bad Good

Created on 2018-06-20 by the reprex package (v0.2.0).

[hadley]

library(haven)

x1 <- labelled(-1:3,  c(Good = -1, Bad = 3))

tmp <- tempfile()
write_dta(data.frame(x1), tmp, version = 14)
x2 <- read_dta(tmp)$x1

x1
#> <Labelled integer>
#> [1] -1  0  1  2  3
#> 
#> Labels:
#>  value label
#>     -1  Good
#>      3   Bad
x2
#> <Labelled double>
#> [1] -1  0  1  2  3
#> 
#> Labels:
#>  value label
#>  NA(z)  Good
#>      3   Bad

Created on 2018-06-20 by the reprex package (v0.2.0).

[hadley]

@evanmiller I think this is a readstat issue — readstat_value_is_tagged_missing() is returning TRUE for the value passed to the readstat_set_value_label_handler().

[evanmiller]

Fix: WizardMac/ReadStat@42c5212

Test: WizardMac/ReadStat@68d7a59

Looks like there was a problem with signed/unsigned comparisons. The fix involved ripping out code, which I am always happy about.

[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/