Creating dataset that matches profile of another dataset

[andrewjmdata]

Hi, I have mailing file which I have profiled on age, gender, ethnicity, occupation.

From the main database I want to select as big a sample as possible that has the same characteristics as the mailing file based on age, gender, ethnicity and occupation.

I can do this in a very long winded way by matching % of the database that are same in the mailing file but it would be better to if I could find a sas procedure/methodology to do it as I need an element of statistical rigour. 

I am using SAS EG v7.

Thanks

Andrew

[pau13rown]

could use proc sort nodupkey to reduce the mailing file to all the possible combinations of age/sex/ethnicity/occupation. And then merge this onto the master file.

proc sort data=mailingfile out=template (keep=age sex ethnicity occupation) nodupkey;

by age sex ethnicity occupation;

run;

proc sort masterfile

by age sex ethnicity occupation;;

run;

data new;

merge masterfile template (in=a);

by age sex ethnicity occupation;

if a then output;

run;

it deponds a lot on exactly what you're doing ie how they should be matched. You could do something sophisticated ie proc psmatch https://support.sas.com/documentation/onlinedoc/stat/142/psmatch.pdf

[andrewjmdata]

Mmm! Not sure that will work, I probably haven't explained the guts of this too well.

If in the mail file I have the following profile of age sex ethnicity occupation...

Age	Sex	Ethnicity	Occupation	% of total
20-30	Male	Black	Plumber	3.5%
30-40	Male	White	Painter	2.9%
40-50	Male	White	Lawyer	1.7%
20-30	Female	White	Teacher	2.1%
30-40	Male	Asian	Doctor	2.9%
40-50	Male	White	Plumber	1.7%

What I want in the control I select from the main database to have the same profile by age, sex, ethnicity and occupation in terms of the %.

Hope this makes more sense now, and thanks for your help so far.

Andrew

[PGStats · Accepted solution]

I think you need two columns: nMail and nPop with the mail sample sizes and the population sizes respectively. From those, you could do:

data counts;
input Age $ Sex $ Ethnicity $ Occupation $ nMail nPop; 
datalines;
20-30 Male Black Plumber 350 2000
30-40 Male White Painter 290 5000
40-50 Male White Lawyer 170 1200 
20-30 Female White Teacher 210 5000
30-40 Male Asian Doctor 290 500
40-50 Male White Plumber 170 3500
;

proc sql;
create table sampsize as
select
    *,
    min(nPop/nMail) * nMail as SampleSize
from counts;
select * from sampsize;
quit;

proc surveyselect data=pop out=control sampsize=sampsize;
strata age sex ethnicity occupation;
run;

Note: The population and sample size datasets should be sorted the same way.

[andrewjmdata]

That looks like it might be what I want. I'll have a bash at it in the morning.

Many thanks!

Andrew

[Reeza]

And if you don't have the latest version of SAS to use PSMATCH, see the Mayo Clinic's macro for greedy match algorithms.