Re: st: replacement task

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

At 03:54 PM 1/21/2004 -0600, Dimitriy V. Masterov wrote:
> I could not find a command that performs this simple task. I was hoping if
> someone might have some advice. I have a variable named emp that looks
> like this
>
> emp     Freq.
> 0       93,404
> 1       55,537
> 2       36,556
>
> I want to replace emp=2 at random with 0 or 1 depending on the relative
> number of 1s and 0s in the sample. My outcome should look like this
>
> emp     Freq.
> 0       116,329
> 1       69,168
A very similar question came up not too long ago.  This weird looking 
command will get you close to the above split but not quite (and you'll get 
slightly different results than I show here):

. replace emp=uniform() <=55537/148941 if emp == 2

(36556 real changes made)

. tab emp

        emp |      Freq.     Percent        Cum.
------------+-----------------------------------
          0 |    116,387       62.74       62.74
          1 |     69,110       37.26      100.00
------------+-----------------------------------
      Total |    185,497      100.00


Here is the logic:  the uniform fnc generates a uniformly distributed var 
that ranges from 0 to 1.  Because of the uniform distribution, about 37.29% 
of the time the uniformly distributed var will be less than (you guessed 
it) .3729 (i.e. 55,537/148,941).  The statement will be evaluated as true 
in such cases and emp will be assigned 1.  The rest of the time it will be 
false and emp will get assigned 0.  The if emp == 2 will limit the 
replacement to when emp = 2.  Because it is random, you won't get exactly 
the final outcome you described but it will be close.

Now, if you absolutely positively have to get the exact split you describe 
above, try something like this:

. gen id = _n

. gen x = uniform() if emp==2

(148941 missing values generated)

. sort x

. replace emp = 0 in 1/22925
(22925 real changes made)

. replace emp = 1 in 22926/36556
(13631 real changes made)

. sort id

. tab emp

        emp |      Freq.     Percent        Cum.
------------+-----------------------------------
          0 |    116,329       62.71       62.71
          1 |     69,168       37.29      100.00
------------+-----------------------------------
      Total |    185,497      100.00


What you are doing is generating a random variable (no particular reason it 
has to have a uniform distribution, by the way), sorting on the variable, 
and then recoding the first 22925 cases to 0 on emp and the next 13631 
cases to 1. (The lines involving id are necessary only if you want to 
restore your original sort order and have no other var that will let you do 
so).


-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
FAX:    (574)288-4373
HOME:   (574)289-5227
EMAIL:  Richard.A.Williams.5@ND.Edu
WWW (personal):    http://www.nd.edu/~rwilliam
WWW (department):    http://www.nd.edu/~soc

*
*   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/