RE: st: 'Re-ordering' the labels of a variable

["Nick Cox" <n.j.cox@durham.ac.uk>]

A more specialised program for reversing variables, -vreverse-, is also
available on SSC. 

-recode- divides the Stata world. I imagine that several people have
internalised the entire syntax through repeated use. I use it rarely
because I find it easier to use some mix of other commands than to
(re-)learn its Swiss army knife syntax. (No disrespect to Swiss Army
knives, or programs written in that way: I've owned several and written
several, but they can be a pain for occasional use.) And, because of
that, I use it rarely. 

But to the point: I was under the impression that -recode- could manage
value labels too. 

Nick 
n.j.cox@durham.ac.uk 

Phil Schumm

On Mar 25, 2010, at 1:16 PM, Nick Cox wrote:
> 3. Although -sencode- is undeniably useful, and its existence  
> underlines a view elsewhere expressed that -encode- should be  
> revisited by StataCorp to make it more comprehensive, I see nothing  
> here that could not also be achieved directly with -encode-.


Nick is, of course, correct; by using the -label()- option, you have  
complete control over how -encode- assigns integers when encoding a  
string variable.  This appears to have been what the OP was asking  
about.

There is, however, a related problem which occurs frequently: You have  
a categorical variable which is already encoded, but you want to  
change the mapping.  For example, you might have a numeric variable  
called myvar with corresponding value label

myvar:
            1 yes
            2 no

and you want to change this so that the response "no" is represented  
by 0 instead of 2 (e.g., so that you can use boolean operators).  Or,  
you might have a variable encoded thusly:

mylab:
            1 a lot
            2 some
            3 a little
            4 not at all

and you want to reverse this so that the order of the integers  
reflects the natural ordering of the responses (e.g., so that when you  
fit an ordinal regression model, your coefficients have a  
straightforward interpretation).  Although you can often avoid these  
problems when creating your own dataset, those who work with secondary  
datasets in which variables come pre-encoded often wish to change the  
default encoding to facilitate analysis and interpretation.  Many (if  
not most) datasets available in social science archives contain  
encoded variables.

Now, I often see people attack this problem with a combination of - 
recode- and -label define, modify-.  This is a pain, leads to code  
that is difficult to read, and, most importantly, is error prone.  An  
alternative solution is provided by a command I wrote called -re2lab-  
(for "recode to label"), which can be obtained via

     net install re2lab, from(http://rcg-software.uchicago.edu/stata)

This comes with a help file explaining its features and usage.  The  
goal was to simplify the process of re-encoding variables, and to  
reduce the likelihood of making a mistake when doing so.


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/