I would approach this a bit differently.    I'd keep the case and control datasets separate and apply the rand_num as you have done. Drop the SORT and RANK.  (BTW, RANK is working correctly.  SSN is a unique key and it restarts the ranking every time the by level changes.) Do the SQL, but change to two datasets and drop the rand_num part of the JOIN.  Include the case.ssn with the control.ssn in the output dataset and drop the other case variables.. This will get you ALL of the matches for each case, with duplication, so now your task is to de-duplicate.  Non-trivial, but doable.   OR, you can use one of the man 1:N matching macros that people have already written.  Google search for sas matching macro and you will find a number.   Statistically, a 1:5 match isn't that much more informative than a 1:2 match.  Depending on your next steps, you could also use all fo the matches (1:1-1:x) and have a perfectly valid analysis.  The use of a limited number of matches is valuable if you have to do manual abstraction (e.g. extra labor), but if you are totally working with computer files, what's a few extra cycles to use them all. ... View more

In addition to Rick's comments, the Gamma, OR, and RR, and confidence limits are all based on asymptotic theory, so they should not be used with sample sizes this small.  PROC FREQ has a number of EXACT tests that are more appropriate. ... View more

The syntax you have is not technically correct (missing multiple semi-colons), so it is hard to understand what you are really trying to do.  The DATA step, as stated, does not require the data sets to be sorted; is there a BY statement missing? ... View more

The approach that I have used is to create a new variable in the data set based on the RANUNI funciton and then use that to make a pseudo-random selection of a prorportion of the records for audit.  I'm sure you could set up something similar in PROC SQL so you didn't have to do multiple passes of the data. ... View more

As you have learned the hard way, it is good not to modify the original data.  Now that you have done it, deleting the extra observations is a simple data step:   DATA haveless; SET have; IF _n_>2010 THEN DELETE;  *Assumes added rows at the end; RUN;   If you still have the original data, a better approach to augmentation might be:   DATA havemdd0; DO _I_ =1871 to 2010;   mdd=0;   OUTPUT;   END; DROP _i_; RUN;   DATA have2; SET have havemdd0; RUN; ... View more