Combining two value labels into one

[Seungmin Lee]

Dear Statalist,

I am looking for a way to merge two variable labels into one.

I have two variables, var1 and var2, with corresponding value labels (lab1 and lab2).


- var1 ranges from 1 to 100, and lab1 has 100 different labels for each value ("a" for 1, "b" for 2, ..., "zzz" for 100).
- var2 ranges from 101 to 200, and lab2 has 100 different labels for each value ("aa" for 101, "bb" for 102, ..., "zzzz" for 200).
- For certain values of var1, var2 is non-missing. Otherwise, var2 is missing.

(If you are familiar with a survey, think of var1 as a regular response variable including "other", and of var2 as additional response variable conditional upon selecting "other" in var1)



Here's an example of 5 observations in my data. var2 is non-missing if var1 is 29 or 30, and missing otherwise.
(There are several other values in var1 which var2 is non-missing, but var2 is missing in most values of var1)

var1	var2
26	(missing)
27	(missing)
28	(missing)
29	130
30	122

My goal is to combine var1 and var2 into one variable (var3)
The combined variable has the value from var2 if var2 is non-missing, or from var1 if var2 is missing. Thus it will vary from 1 to 200. This is pretty easy.

gen var3 = var2
replace var3 = var1 if missing(var2)

But I am struggling with combining lab1 and lab2 into one value label (lab3) ranging from 1 to 200 for this new combined variable.
I first thought it would be easy because lab1 and lab2 do not have any overlapping values, but I could not figure it out how to do it.

I first crated lab3, a copy of lab1

lab copy lab1 lab3

The problem is that I cannot simply add value labels from lab2; I tried the following command, but it returned the syntax error (r198)

lab define lab3 lab2, add

It seems "lab define, add" requires me to manually write each value labels, but it would be tedious to enter 100 different values in lab2.


Is there a way to combine two value labels, with no overlapping values, into one value label?


// Edited
I checked Nick's method of generating composite categorical variable with label on Stata Journal (https://www.stata-journal.com/articl...article=dm0034), but it does not work well in my case - it generates so many missing values in the composite variable due to missing values in var2. I also think there's easier way to do this.
//


Thank you.

[Clyde Schechter]

If you install the -labutil- package from SSC, written by Nick Cox, I believe the -labvalcombine- command therein will do this for you.

[Seungmin Lee]

Thank you so much Clyde Schechter! Just a single line solved everything.

[Daniel Schaefer]

Here is a less serious (but working!) answer. It's Halloween today, so prepare to be horrified.

preserve // create preserve point
* Example labels
label define xlabs 1 "one" 2 "two" 3 "three" 4 "four" 5 "five"
label define ylabs 6 "six" 7 "seven" 8 "eight" 9 "nine" 10 "ten"
* Save label code to file system
label save xlabs ylabs using labsave, replace
* Import saved do file as data
import delimited "labsave.do", delim(" ") clear
* Modify each line so that one "combined" label is used
replace v3 = "combined"
* Need to use outfile instead of export delimited to handle quotes correctly.
outfile using "labsave.do", noquote replace
* run label builder do file
restore // restore original data
do labsave
label list combined

[daniel klein]

If the two value labels are mutually exclusive, i.e., do not define the same integers, then

elabel copy lab1 lab3
elabel copy lab2 lab3 , add

should work for you, too.

elabel is from SSC, SJ, or GitHib.

[Seungmin Lee]

Thank you Daniel Schaefer and daniel klein ! They are also very interesting ways to do it.