Alan's suggestion seems to be a logical approach for handling regression reference categories. More generally about missing/adding REGRESSION subcommands, I'd need to think about that -- there are other features I would prefer to see implemented in PSPP first.