Date:         Mon, 20 Oct 2003 11:12:51 -0700
Reply-To:     Will Dwinnell <predictr@BELLATLANTIC.NET>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Will Dwinnell <predictr@BELLATLANTIC.NET>
Organization: http://groups.google.com
Subject:      Stratified sampling with 'proc surveyselect'
Content-Type: text/plain; charset=ISO-8859-1

Problem: Select a sample from a data set, so that distributions of
particular variables are "similar" to the original distribution.

Solution (?): Use 'proc surveyselect' to generate a new data set,
thus:


data LMN;
      input State $ Sex $ X Y @@;
      datalines;
PA      M       2       7
PA      M       7       4
PA      M       1       7
PA      M       5       10
 some number of data lines here...
NY      F       9       1
NY      F       6       1
NY      F       1       3
NY      F       1       2
NY      F       7       6
   ;
run;

proc sort data=LMN;
  by
    State
    Sex;
run;


proc surveyselect data=LMN method=sys seed=8810 rate=0.1 out=XYZ;
  strata
    State
    Sex;
run;



This appears to work, but with a rate of 0.1, and 100 total original
data lines (in data set LMN), surveyselect occasioanlly delivers 9 or
11 observations, not 10.  Is this appropriate?  Am I doing this
correctly?

Many thanks in anticipation...