You aren't doing anything wrong. Somebody around here has been giving you, if not bad advice, at least incomplete advice. Change OUTPRED to OUTPREDM, and include COUNTRY for the missing obs. OUTPRED gives a BLUP (narrow inference space), while OUTPREDM gives the marginal. And this turns out to be exactly what you had in the original ESTIMATE statement. Try this: DATA PRDSALE; SET SASHELP.PRDSALE END=EOF; IF COUNTRY='CANADA' THEN ACTUAL=ACTUAL+500; OUTPUT; IF EOF THEN DO; COUNTRY = 'U.S.A.'; ACTUAL=.; PREDICT=300; PRODUCT='BED'; OUTPUT; END; RUN; PROC MIXED DATA=PRDSALE; CLASS PRODUCT COUNTRY; MODEL ACTUAL=PRODUCT PRODUCT*PREDICT/NOINT solution outp=predictions outpm=mpredictions; RANDOM COUNTRY PRODUCT*COUNTRY; estimate "predicted value" product 1 0 0 0 0 product*predict 300 0 0 0 0/cl; RUN; If mpredictions has what you are looking for in terms of std err and df, then you should be on your way to actually answering the question where this all started. Steve Denham
... View more