[prev in list] [next in list] [prev in thread] [next in thread] 

List:       sas-l
Subject:    Re: PROC SURVEYSELECT results substantially different than true population
From:       "James C. Whanger" <james.whanger () GMAIL ! COM>
Date:       2013-06-28 20:30:07
Message-ID: CAGG=QuHgK85aS7SC5AmvwV247DkmpH3=VQ-0-FK5to-g597YZA () mail ! gmail ! com
[Download RAW message or body]

Randall -- It appears you are specifying both a 'sampsize' and a 'size'
statement which may be incongruent.  The sampsize pulls a given number of
cases for each replication, while the size statement pulls a number of
cases conditional on the proportion of the value of e1i for each
replication.  The result may not be a sample that is  representative of the
population.

I'm not sure how SAS resolves this -- it may be pulling each sample of 1
from the group with the first value of e1i, thus biasing the estimate of
the population mean.






On Fri, Jun 28, 2013 at 11:57 AM, Randall Powers <powers_r@bls.gov> wrote:

> I’m attempting to use PROC SURVEYSELECT to do sampling with replacement. My
> code is below. I’m getting 10,000 replications with a sample size of 1.
> Each time I do this, I’m getting a result where the population mean for
> variable e1i is significantly different than the true population mean for
> e1i in the it.reg3trimmed90 dataset that I’m sampling from. Additionally I
> looked at the percentiles 1,2,…99 and the sample quantiles are not aligning
> with the true population quantiles.
>
> Thanks!
>
> Code:
>
> proc surveyselect data = it.reg3trimmed90 stats METHOD=PPS_WR rep = 10000
> noprint
>                          sampsize =1 seed = 0 out =
> it.n721110trim90rep10000;
>   id _all_;
>   size e1i;
> run;
>



-- 
*James C. Whanger*
*
*
[prev in list] [next in list] [prev in thread] [next in thread]