topic Re: Oversampling with PROC SURVEYSELECT in Statistical Procedures

Oversampling with PROC SURVEYSELECT

degdeg — Mon, 02 Oct 2017 13:26:06 GMT

Hi,

i try to use the oversampling technique with PROC SURVEYSELECT.

I have a sample of 8500 obersvations with 1% of 0 and 99% of 1.

I achieved undersampling by droppings 1 but now i want to inflate the percentage of 0 with oversampling.

So i tried with PROC SURVEYSELECT to have a random sample of, for example, 12.5% of 0 and 87.5% of 1 with the URS method but he can't do a random sample greater with a number of 0 greater than in the original sample.

For example, i got like 99 times 1 and 1 time 0. What i want is to have like still 99 times 1 but 10 times 1 in my new sample, is it impossible with PROC SURVEYSELECT ?

Thanks you by advance.

Re: Oversampling with PROC SURVEYSELECT

Ksharp — Tue, 03 Oct 2017 13:37:23 GMT

Can you use TWO proc surveyselect to sample for these two level 0,1

and after that combine them together by

data want;

set a b;

run;

or try this one.

data have;
 set sashelp.class ;
 size=1;
run;
proc sort data=have;by sex;run;
proc surveyselect data=have out=want method=pps_wr sampsize=(100 100)  outhits;
strata sex;
size size;
run;

Re: Oversampling with PROC SURVEYSELECT

Ksharp — Tue, 03 Oct 2017 13:42:42 GMT

data have;
 set sashelp.class ;
run;
proc sort data=have;by sex;run;
proc surveyselect data=have out=want method=urs sampsize=(100 100)  outhits;
strata sex;
run;