topic Re: Sampling in SAS Programming

Sampling

Sth19 — Fri, 01 Dec 2017 21:19:20 GMT

I am working on a predictive model with a sample size of about 430K, of which the target is binary. The rare event (1) has about 1000 samples. I need to come up with a way so that the ratio is about 50%. Currently I only have access to SAS Enterprise Guide.

What are some recommendations on how to handle this with sample code?

Thanks

Re: Sampling

Reeza — Fri, 01 Dec 2017 21:24:19 GMT

Which Task/Proc are you using? I'm assuming Logistic regression but depends partially on your variables.

If you define your model/proc you can find full examples in the SAS documentation for that procedure.

Re: Sampling

PGStats — Fri, 01 Dec 2017 22:07:24 GMT

Use proc surveyselect, stratify by your target variable, and request sampling rates of 1 and 1/430 for your target and non target strata, respectively.

Re: Sampling

ballardw — Fri, 01 Dec 2017 22:31:46 GMT

@PGStats wrote:

Use proc surveyselect, stratify by your target variable, and request sampling rates of 1 and 1/430 for your target and non target strata, respectively.

Or instead of sample rate you can specify an exact sample size. If you request 100 from reach strata your sample will be half of one and half of the other.

It is helpful to know that Surveyselect will provide both the selection probability and the sampling weight for each record.

Re: Sampling

Ksharp — Sat, 02 Dec 2017 10:29:38 GMT

You want oversample as ratio 1:1 ?

data class;
 set sashelp.class;
run;
proc sort data=class;
by sex;
run;
proc surveyselect data=class out=want sampsize=(5 5) seed=12345678;
strata sex;
run;

Re: Sampling

Sth19 — Mon, 04 Dec 2017 14:50:31 GMT

I do want to oversample as a 1:1 ratio.