Sampling a dataset without replacement.

[RJ99]

I have a small dataset ( ~2000 obs) and ~20 vars of which the var X is really crucial. Var x has values ( a and b ) such that a:b =6:4. I want to create two samples ( 1800 Obs and 200 obs) of this dataset such that in final datasets it keeps the ratio of a 😛 in var x same (6:4). How shall i do that.

[RaviKommuri]

Can you please give some example of output datasets?

[Doc_Duke · Accepted solution]

Separate the original dataset into two, one for x=a and the other for x=b.

From x=a data set sample 1080 (60% of 1800) and 120 (60% of 200).  Do analogously for x=b data set.

Join the two subset to get the final file of 1800.  Ditto for 200.

When I do a sample without replacement, I will often add a variable to the data generated with ranuni (or one of the other random number generators in SAS), sort on that variable and then just do a DATA step to sequentially pull the samples.  You should also be able to do this with PROC SURVEYSELECT, but I like the simplicity in interpretation of the DATA step.

[Rick_SAS]

For more details on Doc's advice, see the article   Sample without replacement in SAS - The DO Loop

[RJ99]

Thanks a lot Doc@Duke and Rick. It is really helpful.