write_dta: numeric labelleds not written correctly

[diogocp]

I caught this in #140 but have not yet discovered the cause of the problem.

> num <- labelled(c(1, 2), c(a = 1, b = 3))
> num
<Labelled>
[1] 1 2

Labels:
 value label is_na
     1     a FALSE
     3     b FALSE
> roundtrip_var(num)
<Labelled>
[1] 1 2

Labels:
 value label is_na
     0     a FALSE
     0     b FALSE

I opened the file in Stata and verified that the labels are not written correctly:

. label list
x..i..:
           0 a
           0 b

It might be a bug in ReadStat, since the label-writing code in DfWriter seems to be basically the same as the code for integer labelleds, which do work fine.

[evanmiller]

Stata only supports value labels for integers:

http://www.stata.com/help.cgi?dta#value_labels

You could try creating the label set for integer values (readstat_label_int32_value) but I am not sure what the Stata behavior is when an integer label set is applied to a double-precision column.

[diogocp]

You're right. I thought I had tried to label a double in Stata, but apparently I did not.

. label define test 3.14 "pi"
may not label 3.14

Applying an integer label to a double variable does work, but is probably not very useful.

. list, clean

          x  
  1.      1  
  2.   3.14  

. label define test 1 "one"
. label values x test
. list, clean

          x  
  1.    one  
  2.   3.14 

The easiest way to fix this would be to warn and discard labels when writing numerics. Writing only those labels with integer values could be very surprising for users.