topic Re: how do I do this? in SAS Programming

How to randomly impute gender values

K_S — Fri, 03 Feb 2017 05:15:38 GMT

What statements do I add after the DO statement to randomly impute gender values of 'F' and 'M' (with 60% of people being female).

data imputed;

set missing;

if geneder = ' ' then do;

..........????

end;

run;

Re: how do I do this?

anoopmohandas7 — Fri, 03 Feb 2017 03:06:24 GMT

The below sample code should guide you through.

data have ;
input id sex $ ;
datalines ;
23 M
25 M
26 F
27 F
28 F
;
RUN ;

DATA WANT;
length Y $10. ;
set have ;
if sex='M' then do Y='Male' ;
end ;
else if sex = 'F' then do Y='Female' ;
end ;
run ;

proc print data=want;
run;

Re: how do I do this?

K_S — Fri, 03 Feb 2017 03:20:01 GMT

anoopmohandas7, that didn't help. 😞

I need to replace the missing observations by generating Ms and Fs randomly in a 4 to 6 ratio....

Re: how do I do this?

PGStats — Fri, 03 Feb 2017 03:23:38 GMT

Use the random number generator RAND which returns a random number uniformly distributed between 0 and 1 :

data imputed;
set missing;
if missing(gender) then 
	if rand("uniform") < 0.6 then gender = "F"; else gender ="M";
run;

Re: how do I do this?

anoopmohandas7 — Fri, 03 Feb 2017 03:23:45 GMT

Can you show me how your data looks like and what you expect it to be.

Re: how do I do this?

K_S — Fri, 03 Feb 2017 04:05:59 GMT

Thank you, PG Stats! This does the trick! 😉

Re: how do I do this?

art297 — Fri, 03 Feb 2017 04:07:00 GMT

@PGStats: You're the statistician .. I'm only the Psychologist, but wouldn't LE .5 be the more appropriate cutoff?

Art, CEO, AnalystFinder.com

Re: how do I do this?

K_S — Fri, 03 Feb 2017 04:07:58 GMT

Please see PGStats reply for what I was looking for.

Thank you so much for your reply though! 🙂

Re: how do I do this?

K_S — Fri, 03 Feb 2017 04:09:56 GMT

My question stated 0.6-PGStats just gave me what I asked for 😉

Re: how do I do this?

art297 — Fri, 03 Feb 2017 04:12:19 GMT

I hadn't seen your 4 to 6 ratio post. @PGStats: my sincere appologies. I should have know better than to question you 🙂

Art, CEO, AnalystFinder.com

Re: how do I do this?

PGStats — Fri, 03 Feb 2017 04:18:46 GMT

@art297, you are welcome to question me any time

Re: how do I do this?

Reeza — Fri, 03 Feb 2017 05:05:35 GMT

@PGStats idea of using RAND is what I'd use, but not a Uniform Distribution, but a Bernoulli.

if rand('bernoulli', 0.6) = 1 then sex='F';
else sex='M';

Re: how do I do this?

PGStats — Fri, 03 Feb 2017 05:14:02 GMT

Same thing, under the hood. You could also use "table".

Re: how do I do this?

Reeza — Fri, 03 Feb 2017 05:15:00 GMT

@PGStats wrote:

Same thing, under the hood. You could also use "table".

I was wondering about that...it's been a while since I've studied random number generators 🙂

Re: how do I do this?

PGStats — Fri, 03 Feb 2017 05:25:05 GMT

Of course, I haven't seen what's under the hood, I can only guess. But as far as I know, all RAND generators are based on transformations of the uniform generator.

Re: how do I do this?

ballardw — Fri, 03 Feb 2017 15:27:34 GMT

@Reeza wrote:

@PGStats wrote:

Same thing, under the hood. You could also use "table".

I was wondering about that...it's been a while since I've studied random number generators 🙂

But since rand('TABLE', 0.6, 0.4) would return either a 1 or 2 then you still need a conversion to the text gender value requested so I'm not sure you gain much over the If/ then/ else approach with uniform.

But if gender had been coded as 1,2 (or just about any sequential pair of numbers) then that would be another story.