Produce and count modes of combinations of variables

[alex_philby]

Hello everyone !

My data looks like this :

occupation	employer	qualification
dentist	Private	9
construction worker	Private	2
construction worker	Public	2
dentist	Private	9
dentist	Public	8
zoologist	Public	8
dentist	Public	6
dentist	Public	6
police officer	Public	4
police officer	Public	4
police officer	Public	4
police officer	Public	4

I want to have the modes of the qualification variable for each occupation/employer couple. I'd also like to have counts for the modes. If there is an equality, the lowest qualification would be selected (lowest is 1 and highest is 9).

Result would be something like this :

occupation	employer	qualification	count
dentist	Private	9	2
dentist	Public	8	1
dentist	Public	6	2
construction worker	Private	2	1
construction worker	Public	2	1
zoologist	Public	8	1
police officer	Public	4	4

My data has 1 million observations, so an sql approach is probably best.

[novinosrin]

data have;
input occupation & $25.	employer :$20.	qualification ;	
cards;
dentist		Private	9
construction worker		Private	2
construction worker		Public	2
dentist		Private	9
dentist		Public	8
zoologist	Public	8
dentist		Public	6
dentist		Public	6
police officer		Public	4
police officer		Public	4
police officer		Public	4
police officer		Public	4
;
proc sql;
create table want as
select   occupation,employer,qualification,count( qualification ) as count
from have
group by occupation,employer,qualification;
quit;

[Astounding]

To get counts (before limiting results to modes):

proc freq data=have;

tables occupation * employer * qualification / out=want (drop=percent);

run;

[alex_philby]

I already tried this but it takes hours before SAS finally crash down.

[Astounding]

It's difficult to believe that would happen with only 1M observations, but there is a workaround:

proc sort data=have;

by occupation;

run;

proc freq data=have;

by occupation;

tables employer * qualification / out=want (drop=percent);

run;

[alex_philby]

I think it's because at work we have to run SAS on some crappy virtual machine. I always have this problem when I run proc freq on 1M+ tables.

Ok so now I've got the counts but no way to find how to get the modes on SAS. Proc univariate only allows you to have the mode of one variable, not the modes of occupation*employer*qualification.

Any idea ?

[Astounding]

If your data set holding the counts is named COUNTS, you would just need to sort and subset:

proc sort data=counts;

   by occupation employer descending count qualification;

run;

data want;

set counts;

by occupation employer descending count qualification;

if first.count;

run;