Venita DePuy <depuy001@NOTES.DUKE.EDU> sagely replied: > Unless you're talking about Proc Robustreg in v9 . . > Haven't used it personally, but will do robust regression. > Not sure if it's applicable to the subject at hand.

Excellent point (as I would expect from you).

PROC ROBUSTREG is *experimental* in SAS 9.0, and so you do have to watch out for any surprises, like, say, a complete failure with tracebacks showing up in your log file. (I have not played enough with this proc to see this, but I have seen fun like this with other experimental procs, like PROC SURVEYFREQ in SAS 9.0 .)

PROC ROBUSTREG does M estimation (a la Huber) and Rousseeuw's LTS estimation, as well as S estimates and MM estimates. Our original poster would probably do better to try one of these. (I would recommend starting off with LTS estimation if he is worried about distribution contamination.) I'm still not sure *why* our OP needed such an unusual minimization function. The SAS Online Docs for PROC ROBUSTREG might provide him with enough information that he could make some decisions on which sort of approaches could give him the best protection from the types of problems he is dealith with.

David -- David Cassell, CSC Cassell.David@epa.gov Senior computing specialist mathematical statistician