topic Re: Sampling with conditions in SAS Programming

Sampling with conditions

Myurathan — Tue, 07 May 2019 10:10:44 GMT

Hi Experts,

I have never done sampling in SAS. I have a population table with a million rows. I will have to extract 500 entries that meet the following criteria.

The sample should include all available distinct values from three column

How can I do this in SAS.

Thanks in advance.

Re: Sampling with conditions

Onizuka — Tue, 07 May 2019 10:15:10 GMT

Hi, you can use the proc surveyselect :

Data have ;
set Yourtable ;
keep var1 var2 var3 ;
run ;

proc surveyselect data=have
   method=srs n=500 out=SampleSRS;
run;

Re: Sampling with conditions

colabear — Tue, 07 May 2019 16:23:30 GMT

Hello,

I have a follow up to this questions. I need random samples. However if the number of observations from a 15% sample are less than 10 records, then I need the program to select 10 random records. Is there any way to do this?

Re: Sampling with conditions

Watts — Tue, 07 May 2019 16:34:26 GMT

You can use the NMIN= option to specify the minimum sample size.

proc surveyselect rate=.15 nmin=10;

Re: Sampling with conditions

colabear — Tue, 07 May 2019 20:07:59 GMT

Thank you very much!

Re: Sampling with conditions

Rick_SAS — Mon, 13 May 2019 18:00:15 GMT

I just want to point out that the question says

> The sample should include all available distinct values from three columns

The PROC SURVEYSELECT code that you marked as "correct" does not necessarily "include all available distinct values." It simply extracts 500 random observations.

In general, it might be impossible to satisfy that constraint. For example, if X1 = _N_, then there are 1 million distinct values and no subset of 500 observations can include all distinct values. If you want to include all distinct values, you would have to sort the data, then use the FIRST.VAR technique to extract the distinct combinations:

proc sort data=have;
by x1 - x3;
run;

data distinct;
   set have;
   by x1 - x3;
   if first.x1 | first.x2 | first.x3;
run;

This method is unlikely to create 500 observations.