topic Random Sample of a non-event in SAS Data Management

Random Sample of a non-event

murugan — Wed, 13 Aug 2014 20:35:14 GMT

Hi All:

I am working on a logistic regression-- binary outcome. The 1 (or the event) has a sample size of 600, while the 0 (non-event) has a sample size of 8000.

How do I extract a random sample of 600 of the non-event? It seems like proc suveyselect only lets me take a random sample of all participants, not just those who fall in the non-event.

Thanks so much!

Re: Random Sample of a non-event

ballardw — Wed, 13 Aug 2014 20:54:45 GMT

On the input dataset name use the where option to select input records only for specific values. This completely subsets the data. The selection probability and weights will only work for that subset so be careful.

proc surveyselect data=lib.have (where=(eventvariable = 0)) ....

Re: Random Sample of a non-event

murugan — Fri, 15 Aug 2014 17:12:14 GMT

This code is not working. I keep getting an error message that says that my use of the where statement is incorrect.

Re: Random Sample of a non-event

ballardw — Fri, 15 Aug 2014 18:48:55 GMT

Post the code you're submitting

Re: Random Sample of a non-event

stat_sas — Fri, 15 Aug 2014 18:52:46 GMT

Hi,

Seems like trying to make a dataset containing equal number of event and non-event cases. Try this to get the desired sample dataset.

data sample;

set have;

if y=1 or (y=0 and ranuni(1234)<0.075) then output; /* Suppose y is the event variable */

run;

Re: Random Sample of a non-event

murugan — Fri, 15 Aug 2014 19:08:40 GMT

When I used the code that you suggested, I got: data set has 0 observations and 1 variable.

I am trying to make a dataset containing an equal number of event and non-event cases.

Re: Random Sample of a non-event

murugan — Fri, 15 Aug 2014 19:09:12 GMT

Thank you for all of your help!

Re: Random Sample of a non-event

ballardw — Fri, 15 Aug 2014 20:14:40 GMT

If you had clearly stated your need earlier then STRATA in survey select with event and nonevent sample sizes specified would have worked.