Relational database fields are indeed unordered. But, SAS data sets are ordered, and it is often easier to use a proc SQL for something than a proc sort-data step-proc means/whatever. Also, if you are really using a relational database, you can use passthough SQL to access whatever custom row ordering function that database offers. With large complex datasets, there are often transactions that don't fit the "documented" rules. It is useful to put row numbers into a table so that you can easier find the duplicates values (monotonic unfortunately doesn't guarantee missing values). If you are already sub-selecting the primary table, then you can cut out an additional data step to just number them. ... View more

The following might not keep your data points ordered within each group, but will give you row numbers by group. This works in 9.4, don't know about earlier. PROC SQL;   CREATE TABLE foo AS   SELECT data, group, (MONOTONIC() - MIN(MONOTONIC()) + 1) AS row_n   FROM test1   GROUP BY group; QUIT; ... View more