Problem with -reshape long- applying value label to j values (2023 version)

[Castor Comploj]

Based on a similar discussion in Problem with -reshape long- applying value label to j values - Statalist, the j(newvar) variable after -reshape long- takes on strange value labels.

The following is reduced dataset in long format after reshaping:

* Example generated by -dataex-. For more info, type help dataex
clear
input byte(panelid2 wave) str6 hhid str8 id float var
1 1 "000064" "00006401"   .6229963
1 2 "000064" "00006401"   .6187591
1 3 "000064" "00006401"   .0485496
1 4 "000064" "00006401"  .21651682
1 5 "000064" "00006401"   .6109347
2 1 "000064" "00006402"   .3473928
2 2 "000064" "00006402" .007756131
2 3 "000064" "00006402"   .6543282
2 4 "000064" "00006402" .008100444
2 5 "000064" "00006402"   .8569513
3 1 "000214" "00021401"   .4994775
3 2 "000214" "00021401"   .3270546
3 3 "000214" "00021401"  .07871727
3 4 "000214" "00021401"   .6700594
3 5 "000214" "00021401"   .1014784
4 1 "000214" "00021402"   .7776048
4 2 "000214" "00021402"  .05866977
4 3 "000214" "00021402"  .26950312
4 4 "000214" "00021402"   .6614817
4 5 "000214" "00021402"  .13838032
end
label values wave agecl
label def agecl 1 "51-52", modify
label def agecl 2 "53+", modify
label def agecl 3 "tot", modify

Before reshaping, the data looked like this:

* Example generated by -dataex-. For more info, type help dataex
clear
input str6 hhid str8 id float(var1 var2 var3 var4 var5) byte panelid2
"000064" "00006401" .7488847 .4792806 .29489282  .789392  .3904327 1
"000064" "00006402" .7396652  .409892   .474054 .4045655  .5660811 2
"000214" "00021401"   .83459 .3540865  .9541889 .6604541  .9541239 3
"000214" "00021402" .1537002 .8376353  .8663253 .2670887 .51788366 4
end

To which I applied the simple command, as I usually do:

reshape long var, i(panelid2) j(wave)

I do not understand why in the dataset above, value labels "51-52", "53+", and "tot" come from. These values are nowhere in my code, nor in my dataset. However, I now see that dataex reports the section:

label values wave agecl
label def agecl 1 "51-52", modify
label def agecl 2 "53+", modify
label def agecl 3 "tot", modify

Which I did not write myself. This seems to be a bug in stata reshape.

[Hemanshu Kumar]

Your problem does not reproduce. I start with your data extract and then type your reshape command, and I do not get any value labels.

. d

Contains data
 Observations:            20                  
    Variables:             5                  
------------------------------------------------------------------------------------------------------------------------------
Variable      Storage   Display    Value
    name         type    format    label      Variable label
------------------------------------------------------------------------------------------------------------------------------
panelid2        byte    %8.0g                
wave            byte    %10.0g                
hhid            str6    %9s                  
id              str8    %9s                  
var             float   %9.0g                
------------------------------------------------------------------------------------------------------------------------------
Sorted by: panelid2  wave
     Note: Dataset has changed since last saved.

It is likely the labels were already in your Stata memory from some previous code you executed.

You may want to try

label drop _all

to eliminate any labels straggling in your Stata memory, or even just

clear

[Castor Comploj]

Thanks. Executing #2 before reshape does not solve the issue. I tried

set trace on

and found the relevant section here:

     - $ReS_Call sort $ReS_i $ReS_j
      = version 17: sort panelid wave
      - }
      - restore, not
      - local isstr : char _dta[ReS_str]
      - local labn : char _dta[__JValLabName]
      - if "`labn'" != "" & `"`isstr'"' == "0" {
      = if "agecl" != "" & `"0"' == "0" {
      - local lab : char _dta[__JValLab]
      - capture label define `labn' `lab'
      = capture label define agecl  0 `"-50"'  1 `"51-52"'  2 `"53+"'  3 `"tot"' 
        -------------------------------------------------------------- begin label ---
        - version 10.0
        - local vv : display "version " string(_caller()) ", missing:"
        - gettoken val : 0
        - if (strpos("`val'", "val") > 0 ) {
        = if (strpos("define", "val") > 0 ) {
          gettoken val 0 : 0
          syntax anything [, nofix]
          if "`fix'" != "" {
          local fix ", nofix"
          }
          gettoken var rest : anything
          while `"`rest'"' != "" {
          gettoken lab rest : rest
          local label "`lab'"
          }
          local vlist : list anything - lab
          if "`lab'" == "." {
          local lab ""
          }
          foreach var of varlist `vlist' {
          `vv' _label `val' `var' `lab' `fix'
          }
          }
        - else {
        - `vv' _label `macval(0)'
        = version 5, missing: _label define agecl  0 `"-50"'  1 `"51-52"'  2 `"53+"'  
> 3 `"tot"' 
        - }
        ---------------------------------------------------------------- end label ---
      - label values $ReS_j `labn'
      = label values wave agecl

So I am puzzled where this issue comes from, because it is data I have generated from within Stata.
Anyway, if there is no solution to this, relabeling the values manually solves the issue. Important to note is that this is only a labeling issue, not a data issue.

[Hemanshu Kumar]

The example in the other thread suggests that the problem comes up when a second reshape is attempted. Is that what is happening with you?

Also, and this is certainly unsatisfactory and hack-y, but doing the following between the two reshapes will obviate the problem:

char _dta[__JValLabName]
char _dta[__JValLab]
char _dta[__JVarLab]

[Castor Comploj]

I do not do two reshapes. however, maybe the data was reshaped previously, leaving some info inside the dataset somewhere.
The variable label of "wave" also is named "coverscreen id of respondent" after the reshape. So the issue may be more complicated.
Anyway, the issue lies in the data in such a case - I stop persuing correction from here onwards. Thanks for the help anyway.

[Hemanshu Kumar]

I would consider this a bug, and I think you should inform technical support.

But in the meantime, could you try putting the commands in #4 at the top of your code and see if it fixes the problem?

[Castor Comploj]

Yes, #4 works, also without -label drop _all-