topic Re: Distribute a variable according to another in SAS Programming

Distribute a variable according to another

Mathis1 — Mon, 23 Nov 2020 10:37:00 GMT

Hello,
I would like to allocate a (categorical) variable ( Y=1,2,3,4) to each observation with a precise repartition, in function of another variable (X). I explain myself :

If X=0, I want :

- 1% of the population with Y=1

- 10% of the population with Y=2

- 40% of the population with Y=3

- 49% of the population with Y=4

And if X=1, I want :

- 2% of the population with Y=1

- 12% of the population with Y=2

- 56% of the population with Y=3

- 30% of the population with Y=4

Etc until X=10

I know how to make the do loop and the conditional statement :
Do X=0 to 10 ;
If X=i then ...
i=i+1;

But I don't know how to allocate each observation to Y in the right proportions.
Of course, I know for each X what the repartition (in percentage) should be. I just need to know how to allocate the observations to each level of Y.

Thank you by advance for your help !

Re: Distribute a variable according to another

PeterClemmensen — Mon, 23 Nov 2020 10:38:55 GMT

How many different values can X have?

Also, can you provide some sample data and what your desired result looks like? Makes it much easier to provide usable code.

Re: Distribute a variable according to another

PaigeMiller — Mon, 23 Nov 2020 10:44:31 GMT

Use the RAND('table', ...) function

https://documentation.sas.com/?cdcId=pgmsascdc&cdcVersion=9.4_3.5&docsetId=lefunctionsref&docsetTarget=p0fpeei0opypg8n1b06qe4r040lv.htm&locale=en#p0rssof65ddoyyn1s3kfdap849lr

Example:

if x=1 then y=rand('table',0.01,0.1,0.4,0.49);

Re: Distribute a variable according to another

ballardw — Mon, 23 Nov 2020 15:50:58 GMT

Exactly how many observations are you going to process?

If you only have 25 records where X=1 it is going to be very hard to get any distribution that "exactly matches" the stated percentages. (1 value of y=1 is 4% in this case). Other similar issues will exist with other numbers of records with any given X value.

So what would expect for a "distribution" if you only have 25 X=1 records? or 50 X=0 (1% = half a record or Y=1 is 2%).

It is unlikely to be able to match any "exact distribution" if the number of X= records is not exactly 100 or a multiple of 100. Also and "exact distribution" likely is not going to be a true random distribution if that is what is desired.