Re: st: Puzzled by behavior of -recode varlist (.a/.z=.)-

[Richard Williams <Richard.A.Williams.5@nd.edu>]

At 12:53 PM 4/23/2004 -0400, Renzo Comolli wrote:
> Dear Statalist,
>
> I don't understand the behavior of
> . recode varlist (.a/.z=.)
One other tidbit to what I had before:  Stata 7's -recode- command did not 
use -inrange-.  As a result, Stata 7 and Stata 8 can sometimes produce 
different results for a -recode- command.  Example:

Stata 7:
. list

         var1
  1.        1
  2.        2
  3.        4
  4.        7
  5.        8
  6.        9
  7.       12
  8.        .
  9.        .
 10.        .

. recode var1 9/. = 9
(4 changes made)

. list

         var1
  1.        1
  2.        2
  3.        4
  4.        7
  5.        8
  6.        9
  7.        9
  8.        9
  9.        9
 10.        9

The researcher presumably wanted all values that were 9 or greater (where 
greater includes missing) to be recoded to 9.  That is what happened.  But 
in Stata 8, using the exact same data and syntax,

Stata 8:

. recode var1 9/. = 9
(var1: 1 changes made)

. list

     +------+
     | var1 |
     |------|
  1. |    1 |
  2. |    2 |
  3. |    4 |
  4. |    7 |
  5. |    8 |
     |------|
  6. |    9 |
  7. |    9 |
  8. |    . |
  9. |    . |
 10. |    . |
     +------+


Stata 8 interpreted the recode command as saying that any nonmissing values 
between 9 and positive infinity should be recoded as 9.  This is probably 
not what the researcher intended, i.e. the researcher wanted the 3 missing 
cases recoded too; but if the researcher did want what actually happened, 
he or she could have done it with the command

recode var1 9/max = 9

Now, my guess is that Stata 8's recode behavior will only cause problems 
under rare and esoteric conditions (although I don't think the above is all 
that unlikely a scenario). Nonetheless, this behavior strikes me as a bug, 
and at a minimum it should be documented.  Renzo's original command, recode 
varlist (.a/.z=.), winds up being equivalent to -recode varlist 
(min/max=.)- which does exactly the opposite of what he intended and is 
certainly not what people would expect.

Also, -inrange-'s behavior (treating missing as being either negative 
infinity or positive infinity) was ok when there was only one possible 
value for missing; but now that Stata has a range of MD values it is not so 
logical.

Whether you can fix these quirks in -recode- and -inrange- without creating 
other problems, I don't know.

*
*   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/