topic Re: Random Sample Selection with a Condition in Statistical Procedures

Random Sample Selection with a Condition

c8826024 — Wed, 11 Apr 2012 19:57:28 GMT

Hi, everybody,

Did anybody know how I could select a random sample with a condition? Say, I have a 1 MM customers loan dataset. I wanted to select two random samples with a condition that the average loan balance in each sample should close to $800. Did anybody know how I could do it?

Thanks in advance.

Re: Random Sample Selection with a Condition

Reeza — Wed, 11 Apr 2012 20:08:19 GMT

What does your data look like? and What are your other criteria?

Would a sample of 3 loans, say $10, $800, 1590 be a valid sample?

Re: Random Sample Selection with a Condition

Hima — Wed, 11 Apr 2012 20:09:31 GMT

You probably would have to adjust the code according to your requirement but this would help as start up for you.

Code:

DATA TEMP;
INPUT SAMPLE LOAN;
CARDS;
1 800
1 800
2 100
3 800
4 800
3 800
4 800
;
RUN;

PROC SQL;
SELECT SAMPLE, AVG(LOAN) AS AVG_LOAN FROM TEMP
WHERE RANUNI(111) < 0.55
GROUP BY SAMPLE HAVING AVG_LOAN = 800;
QUIT;

Output:

sample avg_loan

1 800

4 800

Re: Random Sample Selection with a Condition

Patrick — Wed, 11 Apr 2012 20:19:11 GMT

Not being a statistician in my simple world I would expect a RANDOM sample to have the same characteristics than the universe you draw it from. If this is true then you would first have to "tailor" yourself a universe with the desired characteristics.

I assume you first would have to sub-set your source table and then draw the sample from this sub-set (http://support.sas.com/kb/24/722.html).

Re: Random Sample Selection with a Condition

Rick_SAS — Thu, 12 Apr 2012 13:26:55 GMT

A similar question was asked at https://communities.sas.com/message/122173

Lots of suggestions there. It's not clear how your $800 requirement relates to the properties of the data. Is $800 the average balance among all customers?

Re: Random Sample Selection with a Condition

PGStats — Thu, 12 Apr 2012 14:56:43 GMT

Two things to consider that would considerably change the approach taken:

What should the variance of the sample sets be? The same for both sets or the same as the population?
What is the purpose of the control over the mean (=800$). To eliminate the effect of a cofactor or to match the overall loan population?