Date:         Fri, 2 Oct 1998 02:46:34 +0100
Reply-To:     John Whittington <medisci@POWERNET.COM>
Sender:       "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From:         John Whittington <medisci@POWERNET.COM>
Subject:      Re: random sample selection
Comments: To: MelliJ@AOL.COM
Content-Type: text/plain; charset="us-ascii"

Melanie,
 
At 19:06 01/10/98 EDT, Melanie Barish wrote:
>Hello everyone!  Can anyone give me some suggestions as to the proper code for
>a random sample selection.  Thus far, at the recommendation of a co-worker, I
>have used a ranuni command, followed by a data step with a retain statement.
>However, when attempting to select a 20% random sample, I am not getting
>exactly 20%.  For instance, when I select 20% of a data set that contains 36
>observations, I get 9 observations.  Thanks for the help...
 
You'll find this question addressed frequently in the SAS-L Archives.  The
most common solution to get a random sample of precise specified size is
along the lines of:
 
  data sample (drop = k) ;
    k = 1000 ;                          /* specify sample size required */
    if 0 then set bigdata nobs = n ;    /*  get nobs, without reading
anything */
    do i = 1 to n while (k > 0) ;
      if ranuni(123456) < k/n then do;
        k = k-1;
        set bigdata point = i ;
        output ;
      end ;
      n=n-1 ;
    end ;
    stop ;
  run ;
 
If you want to specify the exact proportion to be sampled, rather than the
exact size of the sample, the start of that can be modified, something like:
 
  data sample (drop = k) ;
    if 0 then set bigdata nobs = n ;    /*  get nobs, without reading
anything */
    k = n * 0.XXX ;                     /* 0.XXX = desired proportion */
 
Any help ?
 
Kind Regards
 
John
 
----------------------------------------------------------------
Dr John Whittington,        Voice:    +44 (0) 1296 730225
Mediscience Services        Fax:      +44 (0) 1296 738893
Twyford Manor, Twyford,     E-mail:   medisci@powernet.com
Buckingham  MK18 4EL, UK              mediscience@compuserve.com
----------------------------------------------------------------