You can use (as already suggested by others) the CATX function to string the names - but I would much prefer to use CALL CATX, for two reasons: You get a warning if the output variable is too short to hold the result It is a bit more efficient, CPU-wise So, a solution could be something like this: data want; set have; array scores(*) APbio2015--APhist2015; length MET_AP $300; do _N_=1 to dim(scores); if scores(_N_)>=3 then call catx(' - ',MET_AP,scores(_N_)); end; keep ID MET_AP; run; ... View more

@Saba1 As you state that your data has millions of observations, it may be more efficient if you can do the calculation without sorting (assuming that you initial data is sorted by ID and YEAR): data want; array values(100) 8 _temporary_; npos=0; nneg=0; n0=0; do _N_=1 by 1 until(last.id); set have(keep=id ret); by id; values(_N_)=ret; if ret<0 then nneg+1; /* number of negative values */ else if ret>0 then npos+1; /* number of positive values */ else n0+1; /* number of zeroes */ end; call sortn(of values(*)); do until(last.id); set have; by id; _N_=whichn(ret,of values(*)); if ret=0 then rank=0; else if ret<0 then rank=_N_-100+npos+n0-1; else rank=_N_-100+npos; values(_N_)=.; /* set missing, so that next record with same value gets next rank */ output; end; run; If two years have the same ID and RET value, the program assigns the lowest rank to the earliest record. The array dimension (100) should not be smaller than the number of records for a single ID, as you appeared to have one record per ID and YEAR, I assumed that 100 was a safe value. I added the possibility of zeros, always giving a rank of 0. You may want to take missing values into account also (I would suggest giving them a missing rank, currently they are getting NPOS+N0-100). ... View more