How to select random rows from SQL all-to-all join?

[jmoody77]

Hi -

I'm doing a conditional all-to-all merge in SQL with the code below.  It works; but seems silly to have to create the table twice.  I'm looking for a way to select randomly from the rows of the joined table before writing it out.  

proc sql;
 create table inftrans as select
  a.patientid as patientid, a.hospsize as hospsize, b.patientid as inf_source, 1 as newinf,
  a.hospid as whereinf, a.daysin as dayinf 
  from work.patients (where=(infected=0)) as a, work.patients (where=(infected=1)) as b
  where a.hospid=b.hospid & a.patientid^=b.patientid;
  create table infrtrans as select * from inftrans
    where ranbin(0,1,10/(hospsize**2))=1;
quit;

When I try adding the condition to the first where statement like this:

proc sql;
 create table inftrans as select
  a.patientid as patientid, a.hospsize as hospsize, b.patientid as inf_source, 1 as newinf,
  a.hospid as whereinf, a.daysin as dayinf 
  from work.patients (where=(infected=0)) as a, work.patients (where=(infected=1)) as b
  where a.hospid=b.hospid & a.patientid^=b.patientid & ranbin(0,1,10/(hospsize**2))=1;
quit;

inftrans will apply the ranbin() condition to a single row from the input dataset & thus be true multiple times.  

So the question is if there is a cleaner way to get a random subset of the dyad-level dataset?

Thank you!

Jim

[FreelanceReinh]

Hi @jmoody77 and welcome to the SAS Support Communities!

You could apply the condition involving RANBIN via a WHERE= dataset option in the first CREATE TABLE statement:

create table infrtrans(where=(ranbin(...)=1)) as ...

(Personally, I would prefer a positive random seed value to obtain a reproducible dataset, but that's your decision.)

Another option would be to create a view (CREATE VIEW statement) and to perform the random selection based on the view (be it with PROC SQL as you did or with PROC SURVEYSELECT if SAS/STAT is available or in a DATA step).

[jmoody77]

Thank you!!