Re: st: Accessing the number of changes made by a "replace" command

[David Kantor <dkantor@jhu.edu>]

At 10:03 AM 3/18/2005 +0000, David Fischer wrote:
> Hi all,
> I am attempting to create a "while" loop that executes, amongst other 
> things, a "replace" command on a particular variable.  I would like this 
> loop to continue until the "replace" command cannot make any more changes 
> to the data.  In order to do this I would like to be able to access the 
> number of (real) changes made by the "replace" command, so that I can say 
> something like "loop while (number of real changes)!=0".  Is there a way 
> to do this in Stata?  I thought there might be an underscore variable that 
> would do this, but I can't find one.
> Thanks!
> David Fisher.
Coincidentally, just yesterday, I wrote to Stata Tech Support, asking about 
this and suggesting that they provide a returned value.  It's potentially 
under consideration, but there are some issues involved.

Until then, the only way to correctly do it is ...
 gen savedcopy = myvar  // possibly with a type inserted after -gen-
 replace myvar = whatever...
 count if savedcopy != myvar
and test r(N)

One other suggestion would be along the lines of...
 count if condition
 ... and pick up the value of r(N) here ...
 replace myvar = whatever if condition

This doesn't work in general, because you may count more observations that 
actually get a changed value.  It is correct only if you are certain that 
every observation that qualifies under the condition will, indeed, receive 
a changed value in the -replace- -- as is the case in the example that 
Michael Blasnik presented.

The "issues" I alluded to above (as I see it) are that, if they were to 
place the value into an r() returned value, it could break some existing 
programs and do-files.  Such a change would cause other r() values to be 
wiped out, and your program might be expecting to find it:

summ thisvar
replace someothervar = something
...code that looks at the r() values from the above -summ- command...

This would cease to work if the -replace- were made r-class.  (The above 
example may be poor coding practice, but it would work, in Stata as we know 
it now.)  Making it s-class, as Michael suggested, would cure the above 
example, but still not solve the problem in general.  Thus, if such a 
change gets implemented, it would need some warnings to go with it.  But, 
of course, -version- ought to take care of it as well.

I hope this has been interesting.
--David

David Kantor
Institute for Policy Studies
Johns Hopkins University
dkantor@jhu.edu
410-516-5404
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/