I'm afraid that proc sort with nodupkey and descending by variable won't produce what the OP requested, namely the LAST record with a tied key value. In data HAVE have below, presumably the user would want what shows in the log for dataset WANT (with seqnum=key). But the proc sort nodupkey … by descending key produces constant seqnum=1:
616 data have;
617 do key=1 to 10;
618 do seqnum=1 to key;
619 output;
620 end;
621 end;
622 run;
623
624 data want;
625 set have;
626 by key;
627 if last.key;
628 put key=z2. seqnum=;
629 run;
key=01 seqnum=1
key=02 seqnum=2
key=03 seqnum=3
key=04 seqnum=4
key=05 seqnum=5
key=06 seqnum=6
key=07 seqnum=7
key=08 seqnum=8
key=09 seqnum=9
key=10 seqnum=10
NOTE: There were 55 observations read from the data set WORK.HAVE.
NOTE: The data set WORK.WANT has 10 observations and 2 variables.
630
631 proc sort data=have out=want2 nodupkey; by descending key; run;
NOTE: There were 55 observations read from the data set WORK.HAVE.
NOTE: 45 observations with duplicate key values were deleted.
NOTE: The data set WORK.WANT2 has 10 observations and 2 variables.
632 data _null_;
633 set want2;
634 put key=z2. seqnum=;
635 run;
key=10 seqnum=1
key=09 seqnum=1
key=08 seqnum=1
key=07 seqnum=1
key=06 seqnum=1
key=05 seqnum=1
key=04 seqnum=1
key=03 seqnum=1
key=02 seqnum=1
key=01 seqnum=1
Edited addition: My take on this is that NODUPKEY takes the first encountered record with a tied key, not the first output sorted record that is tied. The de-duplication takes place at input time, not output time.
