reshape and (value) labels

[daniel klein]

A recent item on the Wishlist for Stata 18 claims that reshape does not preserve (value) labels. The problem is not exactly clear from that post. However, a probably related problem was reported here.

According to this FAQ, in earlier versions of Stata, reshape did not preserve (value) labels. New versions of Stata should do that and, as far as I am concerned, they do.

I will use the example dataset from the linked FAQ:

clear
input id year answer inc
1 80 0 5000
1 81 1 5500
1 82 0 6000
2 80 1 2000
2 81 0 2200
2 82 1 3300
3 80 0 3000
3 81 1 2000
3 82 1 1000
end

label define answer 0 "Yes" 1 "No"
label define year 80 "1980" 81 "1981" 82 "1982"
label values answer answer
label values year year
label variable id "Identification"
label variable year "Year of study"
label variable answer "Answer to question"
label variable inc "value of inc"

Here is what I get with Stata/SE 17.0 for Windows (64-bit x86-64), update level 06 Apr 2022 (outdated) on Windows 11:

. describe

Contains data
 Observations:             9                  
    Variables:             4                  
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Variable      Storage   Display    Value
    name         type    format    label      Variable label
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
id              float   %9.0g                 Identification
year            float   %9.0g      year       Year of study
answer          float   %9.0g      answer     Answer to question
inc             float   %9.0g                 value of inc
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Sorted by: 
     Note: Dataset has changed since last saved.

. label list
year:
          80 1980
          81 1981
          82 1982
answer:
           0 Yes
           1 No

. reshape wide inc answer, i(id) j(year)
(j = 80 81 82)

Data                               Long   ->   Wide
-----------------------------------------------------------------------------
Number of observations                9   ->   3           
Number of variables                   4   ->   7           
j variable (3 values)              year   ->   (dropped)
xij variables:
                                    inc   ->   inc80 inc81 inc82
                                 answer   ->   answer80 answer81 answer82
-----------------------------------------------------------------------------

. describe

Contains data
 Observations:             3                  
    Variables:             7                  
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Variable      Storage   Display    Value
    name         type    format    label      Variable label
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
id              float   %9.0g                 Identification
answer80        float   %9.0g      answer     80 answer
inc80           float   %9.0g                 80 inc
answer81        float   %9.0g      answer     81 answer
inc81           float   %9.0g                 81 inc
answer82        float   %9.0g      answer     82 answer
inc82           float   %9.0g                 82 inc
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Sorted by: id

. label list
answer:
           0 Yes
           1 No

. 
end of do-file

I cannot see any problem here.

Now, let's reshape long. You may or may not know that reshape stores lots of information in characteristics. The information facilities quickly change between wide and long. We will get rid of all those characteristics and pretend that we have started with the wide layout.

mata : st_local("reshape_chars", invtokens(st_dir("char", "_dta", "*")'))
foreach c of local reshape_chars {
    char define _dta[`c'] // void
}

If I now reshape long, again, I get

. reshape long inc answer , i(id) j(year)
(j = 80 81 82)

Data                               Wide   ->   Long
-----------------------------------------------------------------------------
Number of observations                3   ->   9           
Number of variables                   7   ->   4           
j variable (3 values)                     ->   year
xij variables:
                      inc80 inc81 inc82   ->   inc
             answer80 answer81 answer82   ->   answer
-----------------------------------------------------------------------------

. describe

Contains data
 Observations:             9                  
    Variables:             4                  
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Variable      Storage   Display    Value
    name         type    format    label      Variable label
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
id              float   %9.0g                 Identification
year            byte    %10.0g                
answer          float   %9.0g      answer     
inc             float   %9.0g                 
-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Sorted by: id  year
     Note: Dataset has changed since last saved.

. label list
answer:
           0 Yes
           1 No

. 
end of do-file

I cannot find any problems with (value) labels after reshape. Perhaps, I am missing the problem; maybe the problem only occurs in a specific edition of Stata and/or on a specific OS. I encourage everyone, but especially those claiming that there is a problem, to try and replicate my results. Please state the version, edition, and update level of Stata that you are using. Also, please state the OS you are on.

[Clyde Schechter]

Your results are replicated on my setup: Stata/MP 17.0 for Windows (64-bit x86-64), revision 10 May 2022 (up to date), Windows 10 .

[Asjad Naqvi]

Dear Daniel,

The example you have provided works perfectly but this is not the issue.

Have a look at this example:

sysuse auto, clear
keep make mpg foreign

Let's say we want to reshape by "foreign" which has a value label of 0 = Domestic and 1 = Foreign:

reshape wide mpg, i(make) j(foreign)

After the reshape, the value label for "foreign" gets dropped. If it is still existed somewhere, it would be then easy to rename these variables or assign them variable labels.

Currently what one needs to do is preserve the labels information somewhere (either locals or external files), do the reshape, and call back the information to "fix" the variables.
So the suggestion is to let the variable labels stay wherever they are stored. Maybe as an option in the reshape command, for example:

reshape wide mpg, i(make) j(foreign) preservelabel

Hope this clarifies the issue.

[daniel klein]

This is a more specific description of what you seek. Thanks.

You might want to type

char list

after reshape. You will find, among other information,

 _dta[__JValLab]:             0 `"Domestic"'  1 `"Foreign"'

from which you could reconstruct the value label information.

Unfortunately, the various characteristics are not documented and, thus, might be subject to change in the future.